Today’s article will dive into the impressive performance of the ‘Plan B’ chip designer and the neocloud, which is barely making any money yet borrowing loads and loads to meet demand.

AMD

AMD has long struggled with being the far second option after Nvidia, but now that seems to be working in their favour as the demand and concentration risks skew in their favour.

Financials

All figures below are non-GAAP unless otherwise stated.

AMD has spent the better part of three years making the case that it can be a serious competitor in AI infrastructure. This quarter was the clearest evidence yet that the case is being made in revenue, not just roadmaps. Revenue came in at $10.3bn, up 38% YoY and ahead of the $9.9bn consensus, with adjusted EPS of $1.37, up 43% YoY and clearing the $1.29 estimate.

The Data Center (DC) segment was the clear engine of the quarter, generating $5.8bn, up 57% YoY and now representing well over half of total company revenue. Server CPU revenue grew more than 50% YoY with both cloud and enterprise customers each expanding at that rate, a breadth of adoption that makes the growth look structurally durable rather than concentrated in a handful of hyperscaler commitments. AMD’s CPU business has long been underappreciated relative to its GPU story, but the emergence of agentic AI has changed the picture significantly. These workloads don’t only need accelerators, they also need host CPUs to orchestrate the connections between GPU-based servers and traditional data centre infrastructure, and EPYC is increasingly filling that role at scale.

The Meta partnership, which includes plans to deploy up to 6 gigawatts of Instinct GPUs across multiple generations, including a custom MI450-based accelerator, is among the largest public commitments in AMD’s history, with the OpenAI co-engineering relationship adding a second anchor at scale. It’s also worth noting that AMD doesn’t need to displace Nvidia to generate significant value here. The AI infrastructure market is too large and too strategically sensitive for hyperscalers to source from a single supplier, which means a credible second platform has genuine commercial running room regardless of how performance benchmarks play out.

On the competitive landscape, Intel’s next-generation Xeon processor, Diamond Rapids, has slipped to mid-2027, handing AMD a largely uncontested window for its sixth-generation EPYC Venice ramp. Intel’s in-house manufacturing, long considered a potential advantage, is working against it in the near term. AMD’s reliance on TSMC provides more flexible capacity access at a moment when supply is the binding constraint, and switching costs within the x86 ecosystem mean the installed base AMD is building now will be difficult for competitors to dislodge even once that window closes.

The Client and Gaming (CG) segment contributed $3.6bn, up 23% YoY. Client revenue of $2.9bn grew 26% on Ryzen processor demand and continued PC market share gains, with commercial Ryzen Pro sell-through up more than 50%. Gaming revenue of $720m grew 11% YoY on solid Radeon GPU demand, though semi-custom revenue remained a drag. AMD has flagged that gaming revenue will decline more than 20% in the second half due to higher memory and component costs, so CG will likely shift from tailwind to headwind as the year progresses. The Embedded segment added $873m, up 6% YoY, continuing a gradual recovery after two years of inventory-driven declines.

Adjusted gross margin came in at 55%, with adjusted operating income of $2.5bn representing a 25% operating margin, up 1 percentage point YoY. It’s worth flagging that MI450 is expected to carry below-average gross margins at launch and ramps significantly in Q4, meaning AMD’s strongest revenue quarter of the year may also be its most margin-pressured. Management sees the richer server CPU mix as sufficient offset, and the Q2 guidance of 56% adjusted gross margin supports that view for now, but it’s a dynamic worth watching into the back half.

Free cash flow more than tripled to a record $2.6bn, up from $727m in the same period last year.

Price Action

Shares went ballistic after the report, opening up roughly 25%. At this point, I’ve run out of regions to draw support and resistance lines on! It goes without saying that I won’t be entering at these levels. I’ve missed the boat on this one for now. If you’re already holding AMD, though, happy days. This quarter proved what I think a lot of investors were waiting to see confirmed. The market is large enough for AMD to take a meaningful seat at the table without needing to go toe-to-toe with Nvidia for GPU supremacy. AMD’s GPU business displacing Nvidia from the throne? Never say never, but I wouldn’t bet on it anytime soon. That’s not the thesis anyway, and it doesn’t need to be.

Outlook

AMD guided Q2 revenue to approximately $11.2bn, up 46% YoY and ahead of the $10.5bn consensus, with adjusted gross margin of 56%. Server CPU revenue is expected to grow more than 70% YoY in the quarter, with double-digit sequential growth in both server CPU and data centre AI. The bigger event for the rest of the year is the MI450 and Helios ramp, with initial volume in Q3 scaling significantly in Q4, underpinned by AMD’s decision to double its server CPU market forecast to a $120bn TAM by 2030, up from $60bn just six months ago.

The most interesting moment on the call came when Su was pressed on CPU to GPU ratios. Historically, data centre configurations ran one CPU per four to eight GPUs, with the CPU acting purely as a host node. Su’s view is that agentic AI is fundamentally changing that ratio, pushing deployments closer to 1-to-1, and she didn’t rule out a world where CPUs eventually outnumber GPUs as the volume of agentic tasks compounds. It’s a bold claim, but it’s grounded in something real. Every AI agent spawned by an inference workload requires CPU processing for orchestration, data movement, and parallel execution. As those workloads scale, the CPU demand they generate scales with them. If Su’s read on the trajectory is anywhere close to correct, the server CPU TAM revision from $60bn to $120bn by 2030 isn’t management optimism. It’s just the maths playing out.

CoreWeave

If you’ve been subscribed for a while, you’ll know that I’m not the biggest fan of CoreWeave, and my position hasn’t changed in that I refuse to buy shares. However, I’ve grown to appreciate the fact that they’re useful within the ecosystem as compute is so highly sought after and is in a demand crunch. Investors weren’t so pleased with the latest print. Let’s find out why.

Financials

CoreWeave doesn’t have segments in the traditional sense. It’s a single-product business renting GPU compute to whoever needs it most. So the financials here read differently to a company like AMD, and the metrics that matter are different too.

Revenue came in at $2.1bn, up 112% YoY and 32% sequentially, beating the $1.97bn consensus. That’s the good news, and it’s genuinely impressive for a company that didn’t exist at meaningful scale three years ago. The problem is everything below the revenue line. The net loss widened to $740m from $315m a year ago, with interest expense alone hitting $536m in the quarter, up from $264m in the same period last year, a direct consequence of the debt load CoreWeave has taken on to fund its build-out. CoreWeave closed the quarter with nearly $25bn in total debt, having raised $8.5bn in a new non-recourse term loan facility alongside a $2bn equity investment from Nvidia.

Adjusted EBITDA came in at $1.16bn at a 56% margin, which sounds healthy until you note that it was 62% a year ago. The compression is happening because technology and infrastructure costs jumped 127% YoY to $1.27bn, growing faster than revenue. Adjusted operating income was just $21m on $2.1bn of revenue. The reason margins look this thin despite strong top-line growth is structural rather than operational. CoreWeave incurs depreciation and financing costs on installed capacity from the moment it’s commissioned, but revenue only follows once customers begin drawing on it. As the proportion of pre-revenue capacity shrinks relative to the overall fleet, the margin profile should improve. One detail that tends to get overlooked in the cost discussion is that when CoreWeave signs a customer contract, it locks in pricing on the overwhelming majority of the infrastructure required to deliver it upfront, including GPU prices and energy costs. That means the capex guidance raise doesn’t necessarily damage the economics of existing contracts the way it might appear at first glance.

The figure CoreWeave wants investors to focus on is the revenue backlog, which reached $99.4bn at quarter end, up nearly 50% sequentially and roughly 4x YoY. It’s worth noting that much of that sequential jump was already publicly telegraphed, with the Meta, Jane Street, Anthropic and Perplexity deals all announced during the quarter. The revenue guidance for the full year stayed flat despite the backlog step-up because CoreWeave’s 2026 capacity was already booked before those commitments landed. The backlog growth is largely a 2027 story. Of the total, 36% is expected to convert to revenue within 24 months and 75% within four years, with weighted average contract lengths for new capacity sitting at approximately five years. A noteworthy shift within the business is that more than 50% of CoreWeave’s compute is now being used for inference rather than training, meaning customers are actively monetising the models they trained rather than simply continuing to build them. That’s a healthier demand signal than it might look, and it’s also why pricing on older-generation GPUs is actually rising sequentially as inference clients come in and find that newer hardware isn’t always necessary for their workloads.

Operating cash flow was $3bn, though that was more than consumed by the $7.7bn capital expenditure. For a business carrying nearly $25bn in debt and guiding to up to $35bn in annual capex, the path to sustainable free cash flow remains a long one.

Price Action

Sold off ~11% on the back of earnings. As mentioned, I don’t own this, nor am I looking to get in, but if I were to pick this up, the lower price the better, given its highly volatile and uncertain nature. Somewhere in the 80s would be interesting, given it seems to be a region of support.

Outlook

CoreWeave guided Q2 revenue to $2.45bn to $2.6bn, with the $2.53bn midpoint well below the $2.69bn consensus, which is what sent the stock lower after hours. Full-year revenue guidance of $12bn to $13bn was maintained alongside the $900m to $1.1bn adjusted operating income target, with management reiterating that Q1 was the margin trough and that sequential expansion follows through the rest of the year, exiting 2026 at low double-digit adjusted operating margins.

The most interesting thing said on the call was that CoreWeave is now in a position where it allocates capacity rather than sells it. Demand is so far ahead of supply that the company gets to be selective about which customers it brings onto its infrastructure. That’s a remarkable place to be, and it does speak to genuine scarcity value in what CoreWeave has built. The problem is that scarcity value only translates to shareholder value if the margin story actually inflects when management says it will. Adjusted operating income of roughly $820m to $1bn is implied in just the second half, against $51m in the first. That’s a very large step-up for a business carrying nearly $25bn in debt and over $500m of quarterly interest expense. To be blunt, I’ll believe the margin story when I see it.

Final Thoughts

AMD's trajectory is genuinely impressive, and the CPU-to-GPU ratio argument only adds to the conviction. If agentic AI continues to scale the way Su is describing, the server CPU TAM revision isn't a stretch, it's a consequence. CoreWeave is a different story. The debt burden is enormous, operating income is barely visible, and while the demand environment is clearly real, I'm not yet sold on the long-term durability of the business model at this level of leverage. I would be happy to be proven wrong on that one. For now, however, I remain on the sidelines.

Share The Tech Investor

Thank you for reading. If you found this useful, please share it with anyone you think would enjoy it!

Disclaimer: This article is for informational purposes only and should not be considered financial or investment advice. The views expressed are my own and based on publicly available information, market trends, and personal analysis. Readers should conduct their own research and consult a financial professional before making any investment decisions.

But back to the tech side of things, it was a big week for earnings with Alphabet being the clear winner, whilst Meta lagged and was once again punished by investors for their capex. Without further ado, let’s get into it.

All figures are GAAP unless otherwise stated

Alphabet

Google has always been my favourite horse in the AI race, quite simply for two reasons. First, they invented the transformer architecture, which OpenAI implemented and made popular with ChatGPT. Second, the DeepMind team and Sir Demis Hassabis. These guys have been working on AI for a very long time, and as far as I’m aware, are the only team to have broken down major walls in the space. Yes, OpenAI changed the course of the world’s future with ChatGPT, but beyond that, they haven’t achieved much else in the space. I guess what I’m trying to say is if we focus on pure computer science & science, DeepMind is a true lab catered to that, whereas the rest are mainly set up as businesses trying to milk cash from one product. Take my opinion with a pinch of salt, though, as a Londoner, it’s only right for me to bet on a fellow Londoner.

Beyond my opinion of the team, Google has a much more diversified business, as I’ll get into below, than the other hyperscalers and have their fingers in pretty much all the pies, namely ads and cloud.

Financials

Revenue came in at $109.9bn, up 22% YoY, marking the company’s 11th consecutive quarter of double-digit growth. EPS came in at $5.11 against a consensus of $2.62. Net income jumped 81% YoY to $62.58bn, with a significant chunk of that reflecting unrealised investment gains rather than operating performance. Strip that out, and the underlying business was still comfortably ahead of expectations.

Google Services, which includes Search, YouTube, and subscriptions, generated $89.6bn, up 16%, with Search leading the way at 19% growth to $60.4bn. YouTube advertising came in at $9.88bn, up 11%, slightly below expectations, though the shortfall was largely driven by users migrating to paid subscriptions rather than any weakness in the underlying business.

Google Cloud was the standout, as per usual. Revenue reached $20.03bn, up 63% YoY and ahead of the 48% growth posted last quarter, so this was an acceleration, not just a continuation. The segment's operating income roughly tripled, driven by surging demand for AI solutions and the inclusion of TPU hardware agreements as a new revenue stream. The backlog nearly doubled quarter-on-quarter to over $460bn, which is a leading indicator of where Cloud revenue is heading over the next several quarters as those commitments convert to recognised revenue. To put that in perspective, the backlog stood at $92bn a year ago. Pichai noted on the call that Alphabet is compute-constrained right now, and that Cloud revenue would have been higher had it been able to meet demand. That's a remarkable thing to say, and it tells you something about the scale of opportunity sitting in front of them.

Gemini is now processing more than 16 billion tokens per minute via direct API use, up 60% from last quarter, and Gemini Enterprise saw 40% quarter-on-quarter growth in paid monthly active users. Alphabet is also beginning to deliver TPU hardware directly into customers' own data centres, opening up a new revenue stream beyond traditional cloud consumption. The eighth-generation TPUs are split into two variants, one optimised for training and one for inference, with the inference chip delivering 80% better performance per dollar than its predecessor. Analysts estimate the Gemini API alone is now running at around $15bn in annual revenue, up from a $9bn run rate last quarter. This is where I think Alphabet's full-stack approach really sets it apart. OpenAI and Anthropic are largely dependent on third parties for their compute infrastructure, whereas Alphabet owns the chips, the models, and the distribution. That's a compounding advantage that's hard to replicate.

On the capex side, Alphabet spent $35.7bn in the quarter, up from $17.2bn in the same period a year earlier. Full-year 2026 capex guidance was raised to between $180bn and $190bn, up from $175bn to $185bn, partly reflecting the acquisition of Intersect, a data centre company that closed in March. The CFO also flagged that 2027 capex is expected to increase further still. Depreciation costs from new data centres are compounding, and analysts are forecasting negative free cash flow for Alphabet in 2026, which on the surface sounds alarming. However, this capex is being deployed against contracted demand, not speculation. The two metrics worth tracking over the coming quarters are Cloud operating margin and the trajectory of the Cloud backlog. Both moved in the right direction this quarter, which gives me comfort that the spend is ahead of revenue rather than ahead of demand.

Price Action

Shares were up ~7% after results. I've been topping up my position on the climb up as I'm quite bullish on Google. As I've mentioned, I like the diversification of the business, which means it isn't a one-trick pony like some of its peers. Beyond that, the Cloud growth trajectory is hard to argue with, the TPU programme gives them a structural cost and performance advantage that compounds over time, and I back the team behind Google's AI push. I’m happy to continue adding to my position on any weakness.

Outlook

Alphabet didn't provide explicit revenue guidance on the call, which is pretty standard for them. What they did say is that TPU hardware revenues won't become meaningful until 2027, so that's a tailwind that's still largely in front of us. The Wiz acquisition is worth keeping an eye on too, as it'll create a bit of margin pressure in Cloud for the rest of the year, though management seems pretty comfortable with where that's heading. The deal struck earlier this year with Apple is the one that really excites me. Getting Gemini embedded into Apple Intelligence across two billion devices isn't something that shows up in next quarter's numbers, but it's exactly the kind of deal that quietly reshapes the competitive landscape. All in all, a strong quarter, and I'm happy to keep adding here.

Meta

It’s starting to become a repetitive story with Meta. Revenues are growing, BUT capex is also growing, leading to investor unhappiness and selloffs. This quarter was more of the same.

Financials

Revenue came in at $56.31bn, up 33% YoY and ahead of the $55.49bn consensus, making it Meta’s fastest quarterly growth rate since 2021. The headline EPS of $10.44 includes an $8.03bn income tax benefit, which added $3.13 to that figure. Strip it out and adjusted EPS was $7.31, still beating the consensus of $6.67 by around 10%. Operating income rose 30% to $22.87bn, with operating margin holding flat at 41%.

The advertising numbers were the cleanest part of the print. Ad impressions delivered across the Family of Apps (FoA) grew 19% YoY, while the average price per ad rose 12% YoY. Both metrics moving in the same direction simultaneously is an unusually strong result. AI-driven improvements to content ranking drove a 10% lift in Reels time spent on Instagram and an 8% increase in total video time on Facebook, the largest quarterly gain on that platform in four years, with conversion rates improving 1.6% across both platforms. The prevailing concern heading into this print was that Meta's AI buildout would weigh on margins well before the revenue benefits materialised. The Q1 data makes that narrative harder to sustain.

Family daily active people (DAP) came in at 3.56 billion for March 2026, up 4% YoY but below the 3.62 billion the market was expecting. The sequential decline was attributed to internet disruptions in Iran and WhatsApp restrictions in Russia, so not a structural concern.

Reality Labs posted $402m in revenue against an operating loss of $4.03bn. The loss narrowed slightly from $4.21bn a year earlier, but the segment remains the drag it always has been. There’s no near-term path to profitability here, and Zuckerberg has never pretended otherwise.

Beyond the ad business, the most significant recent development is the launch of Muse Spark, the first model out of Meta Superintelligence Labs, which dropped in early April just ahead of the earnings call. The model was rebuilt from the ground up and reaches the same capabilities as its predecessor Llama 4 Maverick, with an order of magnitude less compute, which is a meaningful efficiency gain. It's natively multimodal, proprietary rather than open-source, and will roll out across WhatsApp, Instagram, Facebook, Messenger, and Ray-Ban glasses. Meta is also exploring a paid API for third-party developers. Further Muse-class models are expected through 2026 and 2027, which will not only demonstrate Meta's frontier model credentials but also allow it to leverage its own models internally, from ad-tech tasks through to core engineering, with real implications for the cost structure over time.

On capex, Meta spent $19.84bn in the quarter and raised its full-year 2026 guidance to between $125bn and $145bn, up from the prior range of $115bn to $135bn. Total expenses for the full year are guided at $162bn to $169bn. Operating margin is forecast to compress from 41.4% in 2025 to around 37.9% this year as those costs weigh on PnL. Meta is trying to offset this by cutting costs elsewhere in the business, replacing headcount with AI where possible and using depreciation rather than salaries as the primary cost driver going forward.

Price Action

shares fell ~8% following the capex raise, and are now trading around 20% below their all-time highs. That discount could prove to be a bargain depending on execution, and I'm staying in the position. That said, this isn't a high conviction hold for me. Meta has a history of grand ambitions that don't always deliver, and while the AI capex cycle is a very different proposition to the Metaverse bet, the pattern of aggressive spending is familiar enough to keep me cautious. I'll be adding in small increments, but the Muse Spark rollout is the thing I'm watching most closely. The ad metrics suggest the AI investment is working. Now I want to see the model itself back that up at scale.

Outlook

Meta guided Q2 2026 revenue of between $58bn and $61bn, implying growth of around 25% at the midpoint. On the capex raise, Zuckerberg was direct. Meta has consistently underestimated its compute needs, and the increase reflects higher component pricing rather than a change in strategy. The planned headcount reduction in May is part of a deliberate effort to run a leaner organisation as infrastructure spend scales. The key things to watch over the coming quarters are how the next generation of Muse models performs and whether the capex commitment starts to show up more visibly in the margin structure.

Final Thoughts

Alphabet was the standout of the week. Cloud continued to grow at a rapid pace, the Search narrative held up better than most expected, and the full-stack approach is starting to pay off in a way that’s hard to argue with. Meta delivered on the fundamentals but continues to struggle with investor trust around the capex trajectory, and until the Muse Spark rollout starts showing up in the numbers, that’s unlikely to change.

More broadly, I’d be cautious about reading too much into the market rally. The oil situation hasn’t gone away, and the buffers that absorbed the initial shock of the Strait closure are running thin. At some point, that will have to be priced in. For now, the AI trade remains intact, but I’d be keeping one eye on the macro.

Thank you for reading. If you found this useful, please share it with anyone you think would enjoy it!

Share

Disclaimer: This article is for informational purposes only and should not be considered financial or investment advice. The views expressed are my own and based on publicly available information, market trends, and personal analysis. I own GOOG and META shares, which may change at any time. Readers should conduct their own research and consult a financial professional before making any investment decisions.

In this article, I’ll be covering the latest earnings from Intel and Texas Instruments, two companies whose histories are as intertwined as any in the semiconductor industry, even if they now occupy very different corners of it. If you missed my previous deep dive on Intel and/or my intro to Texas Instruments, both are worth reading alongside this one.

Intel

Given that I just wrote about Intel, I will go straight into it and keep things tight.

Financials

All figures are Non-GAAP where applicable.

Intel posted Q1 2026 revenue of $13.6bn, up 7% YoY and $1.4bn ahead of the midpoint of its own guidance, marking the sixth consecutive quarter in which the company has beaten its revenue outlook. That consistency matters. A year ago, Intel was a company investors struggled to trust on forward guidance, so six straight beats is a meaningful signal that the operational reset under CEO Lip-Bu Tan is gaining traction. EPS came in at $0.29, up 123% YoY against prior guidance of breakeven, though it’s worth noting that roughly $0.06 of that figure came from a one-time interest and other gain.

The segment picture is where things get interesting. The Data Center and AI (DCAI) segment was the standout, growing 22% YoY and 7% sequentially to $5.1bn, with operating profit of $1.5bn representing 31% of segment revenue. AI-driven businesses now account for 60% of total Intel revenue and grew 40% YoY. The shift from AI model training to inference is a key part of that story. As Tan put it on the call, for the past few years, the AI narrative was “almost exclusively” about GPUs and accelerator chips, but the CPU is proving to be an indispensable part of the AI stack. Each GPU now requires more CPUs to run inference workloads than it did during the training era, and Intel is the primary beneficiary of that dynamic. The Client Computing Group (CCG) contributed $7.7bn, up 1% YoY, with AI PC revenue up 8% sequentially and now representing over 60% of client CPU mix.

Intel Foundry grew 16% YoY and 20% sequentially to $5.4bn, ahead of the $4.6bn analysts had expected. The caveat is that the vast majority of that revenue still comes from Intel manufacturing its own chips rather than external customers, which is the harder and more valuable part of the foundry thesis to prove. Management is targeting external customer wins in the second half of 2026. On the 18A process node, Intel said it is hitting internal yield targets, and notably flagged that its next-generation 14A process is already showing better yields than 18A was at the same stage of development. That’s an encouraging data point, and Elon Musk’s stated intention to use 14A for Terafab gives it a commercial anchor that 18A never had at this stage.

Gross margin expanded to 41%, roughly 650 basis points above guidance, driven by higher volumes, improved pricing, and execution on the 18A ramp. Operating expenses (R&D and MG&A) fell 9% YoY to $3.9bn, reflecting Tan’s cost discipline since taking charge, which has included cutting 15% of the workforce and walking away from manufacturing projects in Germany and Poland. In terms of cash flows, operating cash flow was $1.1bn, with gross capex of $5bn, leaving adjusted free cash flow at -$2bn. Notably, tool spending within that capex envelope is projected to rise around 25% YoY, which matters because tool spend is what drives manufacturing capability and yield improvement rather than just square footage. Net income came in at $1.5bn on an adjusted basis, up 156% YoY, a meaningful sign of how far the profitability trajectory has shifted under Tan, though the reported figure was a loss of $3.7bn driven by a $3.8bn Mobileye goodwill writedown. With approximately $23bn in cash and equivalents on the balance sheet, the negative free cash flow is a deliberate consequence of the foundry strategy rather than a sign of distress.

Price Action

is trading at ATHs following one of the most explosive months in its history. It's delivered me a return of +100%, and at this stage, I'm considering trimming my position. My reasoning is straightforward: at a forward P/E of ~125x, a significant amount of the good news is already reflected in the price, and the risk/reward profile from here is less compelling than it was. The upside from current levels requires continued flawless execution on the foundry ramp, external customer wins, and sustained AI demand. All of those are plausible, but none are guaranteed, and the market is currently pricing them as though they are. I remain bullish on Intel's longer-term trajectory, but at these levels I'd rather take some chips off the table and look to re-enter at a more attractive entry point, potentially around $50-$60s.

Outlook

For Q2, Intel is guiding revenue of $13.8bn to $14.8bn, up 2% to 9% sequentially, with DCAI expected to grow double digits and both CCG and DCAI forecast to grow sequentially. Gross margin is guided to 39%, stepping back from Q1’s 41% due to a meaningfully larger contribution from 18A, still early in its ramp, and some inventory tailwinds in Q1 that won’t repeat. Non-GAAP EPS is guided to $0.20.

On the full year, Intel is prudently planning for PC unit TAM to fall into the low double digits in the second half, a sensible assumption given the tariff environment and consumer spending uncertainty. Offsetting that, the server CPU outlook has materially improved over the last 90 days, with management now guiding for strong double-digit unit growth for both the industry and Intel, extending into 2027. That revision upward is meaningful, and the CPU-to-GPU ratio shift Tan described on the call, moving from roughly one CPU per eight GPUs in training environments to approaching parity in agentic workloads, gives that growth a structural rather than cyclical foundation.

For most of the AI boom, the narrative was almost entirely about GPUs, which are purpose-built for the parallel computation required to train large models. But as AI shifts from training into inference and agentic workloads, where multiple models must be coordinated, orchestrated, and managed in real time, the CPU reasserts itself as the natural control plane of the stack. Zinsner noted on the call that the GPU-to-CPU ratio is already moving from around eight to one in training environments to three or four to one in inference, with agentic deployments potentially approaching parity, and that structural shift is a meaningful tailwind for Intel that has nothing to do with the PC market. I'm happy with where Intel is headed, and I buy into the CPU story as a structural rather than temporary phenomenon, though execution risks here carry real weight and shouldn't be dismissed.

Texas Instruments

Texas Instruments is one of the oldest names in semiconductors, tracing its roots back to 1951 with a history that includes Jack Kilby inventing the integrated circuit within its walls in 1958. Today it operates as the world's largest analog semiconductor company, designing and selling chips that manage power, sense data, and control systems across industrial, automotive, and consumer electronics markets. It's not a flashy AI name, but it's an essential one.

TI reports across two primary segments: Analog, which is by far the larger of the two and covers power management, signal chain, and high-volume analog products, and Embedded Processing (EP), which covers microcontrollers and processors used to run and control electronic systems. Together, these two segments account for the vast majority of revenue, with a small Other category covering legacy products.

Analog chips don't get the attention that GPUs or CPUs do, but they are the silent backbone of almost every electronic system. Where digital chips process ones and zeros, analog chips interface with the real world, converting physical signals like temperature, pressure, sound, and voltage into binary data that digital systems can use, a process handled by components called analogue-to-digital converters (ADCs). TI's dominance in this space, with roughly 80,000 products and a market share that no competitor comes close to matching, was on full display in Q1 FY26.

Financials

All figures below are on a GAAP basis unless otherwise stated.

TI posted Q1 FY2026 revenue of $4.83bn, up 19% YoY and 9% sequentially, with net income of $1.55bn and EPS of $1.68, coming in $0.31 ahead of consensus. The beat was broad-based and driven by two end markets in particular: industrial, up roughly 30% YoY and over 20% sequentially with growth across all geographies, and data centre, up approximately 90% YoY and more than 25% sequentially. That data centre figure deserves a moment’s attention. TI doesn’t make GPUs or CPUs for hyperscalers, but its analog chips manage power regulation and signal conversion inside the racks that do, and the AI infrastructure build-out is pulling that demand higher at a pace the market hadn’t fully anticipated.

At the segment level, Analog revenue grew 22% YoY to $3.92bn, with an operating margin of 41.7%, contributing to a company-wide gross margin of 58% for the quarter. Embedded Processing (EP) rose 12% YoY to $723m, but the more striking number was operating profit, which surged to $122m from just $40m in Q1 2025, lifting the segment’s operating margin from 6.2% to 16.9% in a single year. That kind of margin recovery reflects both the industrial cyclical recovery and TI’s significant operating leverage at higher volumes. Automotive grew mid-single digits YoY but was flat sequentially, with China down and the rest of the world up, a dynamic worth watching as the year progresses.

The industrial recovery story is worth unpacking further. Industrial is TI’s largest end market and the one that was most severely hit during the 2023-2024 inventory correction. The fact that it’s now growing 30% YoY while management notes it is still trending below previous cycle peaks suggests there is more runway in this recovery than a single strong quarter might imply. It isn’t a fully recovered market yet, which is actually the more encouraging read for investors.

On manufacturing, TI’s 300mm fab strategy is the structural advantage that separates it from most analog peers. Producing chips on larger 300mm wafers rather than the older 200mm standard delivers meaningfully lower cost per chip, and because TI owns its fabs outright rather than outsourcing to a foundry, it controls both the cost structure and the supply chain. Q1 capex came in at $676m, roughly half the $1.1bn spent in Q1 2025, reflecting a deliberate step-down after several years of heavy investment. Full-year 2026 capex guidance remains at $2bn to $3bn, with management flagging a growing focus on internalising back-end assembly and test capacity where external bottlenecks have emerged. The $555m in CHIPS Act incentives received in Q1, tied to production milestones at the new Sherman, Texas fab, partially offset that spend and contributed to TTM free cash flow of $4.35bn, up 154% YoY. TI returned $1.45bn to shareholders in the quarter through dividends and buybacks, consistent with its long-standing capital return framework.

Price Action

is also trading at ATHs. TXN isn't a portfolio position for me, though it sits near the top of my watchlist as a name I'd look to enter on a meaningful pullback. The investment case is compelling precisely because its core end markets haven't fully recovered yet. Industrial output remains cyclically depressed relative to prior peaks, and automotive demand, particularly in Europe, where passenger car production has been under sustained pressure for several years, adds a further layer of caution around near-term upside. That said, the AI infrastructure angle provides a clearer and more structural revenue tailwind than TI has historically enjoyed, and the market appears to be pricing that in. The risk is that investors are paying a full price for a recovery that is real but not yet complete, and macro deterioration could delay the industrial normalisation that underpins the longer-term thesis.

Outlook

For Q2, TI is guiding revenue of $5bn to $5.4bn, up 2% to 9% sequentially, with EPS of $1.77 to $2.05. CEO Haviv Ilan described the guide as slightly above seasonal, driven by continued strength in industrial and data centre. Gross margin is expected to drift modestly higher from Q1’s 58%, with incremental fall-through of 75% to 85% on a long-term basis excluding depreciation, though near-term margins will be partly absorbed by recurring acquisition charges related to the Silicon Labs deal until it closes in H1 2027.

On the full year, management is cautiously optimistic but deliberately guarded. The key risk Ilan flagged repeatedly is the sustainability of the industrial recovery, which is a view I share. He noted that industrial remains roughly 15% below its 2022 peak despite the strong Q1, which is the more encouraging read for investors, suggesting the recovery has room to run rather than being close to exhaustion. At the same time, he was transparent about last year’s false start, where a strong first half gave way to second-half deceleration, and made clear he wants to see Q2 play out before making any stronger full-year call. On free cash flow, CFO Rafael Lizardi was more forthcoming, stating that assuming the demand signal holds, the company is very likely to exceed $8 free cash flow per share for 2026, a figure that would represent a significant step up from prior years and underscores the earnings power of the business as revenue recovers and capex begins to moderate.

The automotive picture is worth watching. Ilan drew a parallel to the COVID cycle, where automotive was both the last segment to join the recovery and the last to peak. China was down sequentially in Q1 while the rest of the world was up, and management stopped short of calling a turn. With EMEA representing 22% of revenue and the European automotive market still operating well below pre-pandemic production levels, this remains a genuine overhang on the near-term outlook even if the secular content growth story remains intact.

Final Thoughts

After two strong earnings reports, it's hard to argue with the fundamentals, but at current valuations, a lot of that optimism is already in the price, and I'm not a buyer of either at these levels. Both companies have credible longer-term growth narratives for different reasons. Intel, through its CPU renaissance and foundry ambitions, and TI, through an industrial and automotive recovery that still has room to run, are both firmly on my watchlist. For now though, I'd rather wait for a better entry point than chase either name at all-time highs.

Thank you for reading. If you found this useful, please share it with anyone you think would enjoy it!

Share

Disclaimer: This article is for informational purposes only and should not be considered financial or investment advice. The views expressed are my own and based on publicly available information, market trends, and personal analysis. I own INTC shares, which may change at any time. Readers should conduct their own research and consult a financial professional before making any investment decisions.

The war has continued, although there has been some optimism in markets following the announcement of an extremely fragile ceasefire. Is Lebanon included in this or not? Apparently so to Iran, apparently not to the US and Israel. US Vice President JD Vance has left negotiations in Islamabad with no deal agreed, so it’ll be interesting to see how markets open up this week. I’m sure it’ll be another whipsaw week.

Today’s article will be covering the story of Intel, as there have been several progressions recently which are worth covering, given shares have rallied ~40% since the last week of March 2026.

Intel: The Rise

Robert Noyce and Gordon Moore were not first-time founders when they started Intel in 1968. The two had already helped build Fairchild Semiconductor into one of the most important companies in Silicon Valley. Noyce had co-invented the integrated circuit. Moore had articulated Moore’s Law, the principle that the number of transistors on a chip would double roughly every two years, which became the metronome by which the entire industry set its pace. By the late 1960s, they were legends who were frustrated. Fairchild’s parent company kept starving the semiconductor division of the reinvestment it needed. So they left.

They incorporated Intel on July 18, 1968, with $250,000 each of their own money and another $2.5 million raised in a matter of days, mostly on the strength of their names alone. The name came from a portmanteau of “integrated electronics,” though they had to pay $15,000 to buy it from a hotel chain that got there first. They started with a dozen engineers in a conference room and a simple goal: build semiconductor memory that nobody else could. Andy Grove joined almost immediately as their first hire, and the three of them together would go on to form one of the most formidable leadership teams in corporate history.

The memory business was good, but the pivot that changed everything came almost by accident. In 1971, Intel released the 4004, the world’s first commercially available microprocessor. It was originally designed to power a Busicom calculator and contained just 2,300 transistors. Intel’s own management was sceptical that it could be sold, but the scepticism was proven wrong. The 4004 was the seed from which the entire modern computing industry grew.

The truly defining moment came in 1981, when IBM chose Intel’s 8088 processor to power its first personal computer. That decision locked in the x86 architecture as the standard for personal computing, a standard that has never been dislodged. Intel didn’t just benefit from the PC boom. It became the PC boom. The “Intel Inside” campaign launched in 1991 turned a semiconductor company into a consumer brand, something almost unheard of in this industry. By 1993, revenue had crossed $10bn. By the end of the decade, it was pushing $34bn. Grove, who eventually became CEO in 1987, built the operational machine that turned brilliant engineering into one of the most dominant businesses in technology history.

The Fall

The decline didn’t happen overnight, and it didn’t have a single cause. It was a slow accumulation of strategic errors, missed markets, and manufacturing failures that took roughly two decades to play out fully.

The first and most consequential mistake came in 2005. Apple approached Intel about manufacturing chips for a new mobile device. Paul Otellini turned them down, deciding the margins and volumes weren’t worth it. Apple went to TSMC instead, and that relationship became the financial foundation that allowed TSMC to invest in leading-edge manufacturing at a scale Intel never matched. Intel missed mobile entirely. And in doing so, it handed TSMC the runway to eventually overtake Intel on process technology.

The manufacturing collapse followed. Intel had built its identity around being the best fab in the world. That identity crumbled at 10nm. The targets were too aggressive, the execution wasn’t there, and while TSMC delivered its 7nm process on schedule and AMD plugged straight into it with its Zen architecture, Intel was stuck on 14nm for years longer than it should have been. AMD sensed the opening and took it. Cloud providers started deploying AMD’s EPYC server chips. Intel’s data centre market share bled out. In 2020, Apple announced it was leaving Intel for its own ARM-based M1 chips. That same year, Intel buried a 7nm delay disclosure in an earnings release. The stock fell sharply.

And then the AI boom arrived, and Intel simply wasn’t in the conversation. Nvidia had CUDA. The industry reached for GPUs. Intel had nothing comparable. Its Gaudi AI accelerator existed but generated marginal revenue at a time when Nvidia was doing tens of billions in data centre GPU sales per quarter. Revenue peaked near $79bn in 2021 and then fell for seven consecutive quarters. By early 2025, the stock had fallen to around $18, they had recorded losses exceeding $22bn, and the cuts were visible everywhere. Construction on Ohio One, a $28bn campus in New Albany that was supposed to be Intel’s flagship US manufacturing site and begin production in 2025, was slowed to a crawl with completion pushed back to 2030 at the earliest. And in a sign of just how far things had deteriorated, they sold a 49% stake in Fab 34, their flagship facility in Leixlip, Ireland, to Apollo Global Management for $11.2bn just to shore up the balance sheet. The company that once defined semiconductor manufacturing was selling off pieces of its own factories to survive.

The Comeback

Then Lip-Bu Tan took over as CEO in March 2025, and the tone shifted almost immediately. A semiconductor veteran who had previously run Cadence Design Systems, Tan was the first Intel CEO in a long time who actually understood the business of making chips. He came in and did what needed doing. Jobs were cut, expansion projects were rationalised, and the focus returned to what Intel had always been best at: manufacturing. The 18A process node, which had been Intel’s great hope and frequent source of scepticism from the industry, entered high-volume production at the fab in Chandler, Arizona, in January 2026. That milestone alone was significant. It was the first time in years that Intel had delivered something on time.

The broader political backdrop also started working in Intel’s favour. The current administration has made onshoring semiconductor manufacturing a genuine priority, and Intel is the most obvious domestic beneficiary of that shift. The US government converted its remaining CHIPS Act grants into an $8.9bn equity stake in Intel, making Washington one of Intel’s biggest backers and adding a layer of national security tailwind that no competitor could replicate. This matters beyond the balance sheet. There is a growing view in Washington, one I find hard to argue with, that depending on TSMC for the world’s most advanced chips is a geopolitical risk that needs to be addressed. I personally think a Chinese invasion of Taiwan in the next five years is more plausible than markets are currently pricing. That is a highly speculative view, but if it ever materialised, the consequences for global semiconductor supply would be catastrophic. Intel being a viable alternative to TSMC is not just a business story. It is a strategic imperative.

Nvidia and SoftBank made multibillion-dollar investments alongside Washington’s stake. The balance sheet, which had looked genuinely precarious eighteen months earlier, started to stabilise. Intel ended 2025 with $37.4bn in cash and short-term investments and repaid $3.7bn of debt in Q4 alone. A business that had been selling stakes in its own fabs to survive was now paying down debt. And then in the space of ten days in early April, three announcements landed that changed the narrative entirely (for now).

Intel buys back the Apollo stake

On April 1st, Intel announced it was repurchasing Apollo's 49% stake in Fab 34 for $14.2bn, funded through a mix of cash on hand and around $6.5bn in new debt. That is a 27% premium over what Apollo originally paid. There is an awkward irony here that is worth acknowledging. Intel sold this stake to Apollo in 2024 for $11.2bn because it was desperate for cash, and is now buying it back two years later for $3bn more than it sold it for. That is the cost of having been in survival mode. You make decisions you later have to unwind at a premium. It's not pretty, but it needed to be done. Fab 34 makes Core Ultra and Xeon 6 processors, and with AI inference demand accelerating, Intel cannot afford to share the economics of that facility with anyone. Management expects the deal to be accretive to earnings and positive for the credit profile from 2027 onwards.

The Terafab partnership

On April 7th, Intel announced it was joining Elon Musk’s Terafab project alongside Tesla, SpaceX and xAI. The ambition is enormous. Terafab is targeting 1 terawatt per year of AI compute, positioning itself as the largest semiconductor production initiative ever attempted. Intel’s role is to design, fabricate and package custom chips for the project. No financial terms have been disclosed and there’s no official timeline, so it’s easy to be sceptical about how much commercial substance sits behind the headline right now. But the symbolic value is significant. The fact that Musk picked Intel to manufacture for Terafab is a meaningful vote of confidence in 18A. A year ago, that probably would’ve been unthinkable.

The Google deal

On April 9th, Intel announced a multiyear partnership with Google under which Google Cloud will continue deploying Intel Xeon processors across AI training, inference, and general-purpose compute workloads, with both of them also expanding co-development of custom infrastructure processing units (IPUs). No financial terms were disclosed. What makes this meaningful is that Google has simultaneously been building its own custom silicon. The fact that it’s still committing to multiple generations of Intel Xeon tells you x86 isn’t going anywhere in the data centre, and that Intel’s CPU roadmap is credible enough for one of the world’s largest cloud providers to stake its infrastructure on for years ahead.

If you’re keeping track of the ever-expanding universe of PUs, welcome to the IPU. The semiconductor industry has developed a habit of inventing a new processing unit every few years, and the infrastructure processing unit is the latest addition to the family. Where a CPU handles general compute, and a GPU handles parallel processing, an IPU is designed to take over the unglamorous but essential work that would otherwise eat into CPU capacity: networking, storage management, security functions, and packet processing. By offloading those tasks onto a dedicated chip, the CPU is freed up for higher-value work, and data centres become meaningfully more efficient at scale. For hyperscalers running millions of servers, that efficiency gain is not marginal; it compounds.

Price Action

Following these recent developments, is up ~22% in the last week. To think that was trading at $20 last September (2025) is crazy! It’s now nearing all-time high territory. I don’t like to add on huge moves like these, but as a shareholder, I’m pleased to see the jump! Of course, the investment is long-term, but it’s unrealised PnL; however, seeing a positive trajectory reassures the investment thesis.

Conclusion

Intel's story is not over, and the bear case hasn't disappeared entirely. Execution on 18A yields still needs to hold up over time, and Q1 2026 earnings on April 23rd will be the next real test. But the narrative has definitively shifted. A business that was selling pieces of itself to stay alive eighteen months ago is now buying them back, landing partnerships with Google, and manufacturing chips for the most ambitious AI compute project ever attempted. Whether you're a shareholder or just watching from the sidelines, that is a remarkable turnaround in a very short space of time. I'll be watching the April 23rd print closely.

As always, thank you for reading. If you found this useful, please consider sharing it with someone who might enjoy it.

Share

Disclaimer: This article is for informational purposes only and should not be considered financial or investment advice. The views expressed are my own and based on publicly available information, market trends, and personal analysis. I own Intel shares, which may change at any time. Readers should conduct their own research and consult a financial professional before making any investment decisions.

What Is a SPAC?

A Special Purpose Acquisition Company is a shell company that raises capital through an IPO with the express purpose of merging with a private company and taking it public. The SPAC raises money from investors, holds it in a trust account, and has a defined window to find and complete a deal. When it does, the private company effectively inherits the SPAC’s public listing. The result is a faster, cheaper route to public markets than a traditional IPO.

The mechanism has a mixed reputation. SPACs peaked in 2021 during a period of cheap money and speculative appetite, and many deals from that era left investors nursing significant losses. The structure has some inherent tensions worth understanding. SPAC sponsors typically receive a substantial "founder share" allocation as compensation, which can dilute existing shareholders even before the deal closes. Redemptions are another issue. If SPAC shareholders are unhappy with the proposed target, they can withdraw their capital from the trust before the deal closes, sometimes gutting the cash available to the company. In extreme cases, redemption rates have exceeded 80%, leaving companies with a fraction of the capital they had anticipated. PIPE stands for Private Investment in Public Equity, and it is essentially a direct placement of shares to institutional investors at a fixed price, often used to plug the gap left by redemptions.

For quantum computing specifically, SPACs have become the default route to public markets. D-Wave, Rigetti, IonQ, and Infleqtion have all gone this route. The reasons are understandable. These are pre-profitability, capital-intensive businesses that need to tell a long-horizon story to investors. The SPAC format, with its ability to include forward-looking financial projections in investor materials, gives companies more room to make that case than a traditional IPO prospectus allows. Whether investors have been well-served by that flexibility is a different question.

The Quantum Computing Landscape: Four Competing Approaches

Before getting into IQM specifically, it is worth understanding where superconducting fits relative to the other major hardware approaches. Quantum computing is not one technology. It is a set of competing bets on how to build a machine that exploits quantum mechanical properties like superposition and entanglement to solve certain classes of problems faster than any classical computer. The approaches differ in how they physically create and control qubits, the basic units of quantum information.

Superconducting qubits are the approach used by IQM, IBM, and Google. They work by fabricating circuits from superconducting materials cooled to temperatures near absolute zero. At those temperatures, electrical resistance vanishes, and certain circuit elements can be engineered to behave like two-state quantum systems. The primary advantage is that the underlying technology borrows heavily from classical semiconductor manufacturing. Superconducting chips can be fabricated using modified versions of existing chip fab processes, which gives the approach a clearer path to scale. The disadvantage is the cooling requirement. The dilution refrigerators needed to reach operating temperatures are large, expensive, and energy-intensive, which limits deployment flexibility. Qubit coherence times, the window in which a qubit maintains its quantum state before decohering, are also relatively short compared to some other approaches, which increases the demand for fast operations and error correction.

Ion trap systems take a different approach. They trap individual charged atomic ions using electromagnetic fields and manipulate them with lasers. IonQ is the most prominent public company in this space. Ion trap qubits tend to exhibit longer coherence times and higher gate fidelities than superconducting systems. Fidelity here refers to how accurately a quantum operation is executed, essentially the probability that a gate does exactly what it was supposed to do without introducing errors. Higher fidelity means cleaner computation, which matters enormously when you are chaining thousands of operations together and errors compound. The challenge is speed and scale. Gate operations in ion trap systems are slower, and physically scaling up the number of qubits while maintaining the precision of individual ion control remains an open engineering problem.

Photonic quantum computing, the approach taken by Xanadu, uses photons, individual particles of light, as qubits. Photons can, in principle, operate at room temperature, which eliminates the refrigeration requirement entirely and opens the door to very different form factors. Xanadu has specifically focused on a measurement-based approach using squeezed light and photon detection. The challenge is that photons interact weakly with each other, making it difficult to implement the two-qubit gate operations that most quantum algorithms require. Photonic systems also face loss challenges, where photons are lost during transmission or detection, introducing errors.

Neutral atom platforms trap neutral atoms, typically rubidium or caesium, using laser beams called optical tweezers. Companies like QuEra, Pasqal, and Atom Computing are active in this space. Neutral atom systems have attracted significant attention because they offer naturally identical qubits, since every atom of a given element is the same, and they can achieve relatively high qubit counts with promising coherence properties. Gate operations are improving but have historically been a limiting factor, and the control systems required are complex.

None of these approaches has yet demonstrated the fault-tolerant quantum advantage that would make quantum computing genuinely transformative for commercial applications. Each has a credible long-range case. The honest answer is that the winning architecture is still unclear.

Xanadu Goes Public

Xanadu, the Toronto-based photonic quantum computing company founded by Christian Weedbrook, began trading as XNDU on Nasdaq and the Toronto Stock Exchange on March 27th, 2026, making it the first publicly traded pure-play photonic quantum computing company. The deal valued the combined entity at a pro forma enterprise value of $3.1 billion and a market cap of $3.6 billion.

The transaction had a complicated path. Originally structured with approximately $225 million from Crane Harbor’s trust account alongside $275 million in PIPE financing, the deal was stress-tested when 88.1% of SPAC shareholders redeemed ahead of the shareholder vote, leaving just $27 million in trust. Weedbrook moved quickly, finalising the $275 million PIPE within four weeks and bringing in AMD as a strategic investor alongside BMO Global Asset Management and CIBC Asset Management. Ninety per cent of PIPE capital came from new investors, which the company framed as external validation of the photonic approach.

Xanadu’s stated ambition is to build one of the first quantum-powered data centres by 2030. The company’s PennyLane software platform has built meaningful developer mindshare, and the photonic approach theoretically sidesteps the refrigeration infrastructure problem that limits superconducting deployment flexibility. Whether $302 million in gross proceeds is sufficient to execute on a $1 billion data centre buildout, on top of ongoing hardware R&D, is a question investors will be watching closely. Generally, I offer my thoughts, but honestly, amid the current market environment, I can’t see many market participants other than retail taking a big bet on the equity right now. Closed at $11.5 on the NASDAQ and at CA$ 16.01 on the TSE. I will be keeping a close eye, but of all the technologies mentioned above, I think the photon approach is the least-proven, so I would prefer to invest in already somewhat-proven approaches, as ultimately none of the approaches currently work 100%. And when I say work, I mean being able to produce a system that isn’t frequently disturbed and a commercial one.

Now onto IQM…

IQM

IQM Quantum Computers announced on February 23rd, 2026, that it would merge with Real Asset Acquisition Corp. (Nasdaq: RAAQ), a SPAC based in Princeton, New Jersey. The deal values IQM at a pre-money equity valuation of $1.8 billion and, upon close, is expected to leave the combined entity with more than $450 million in cash. That breaks down as approximately $172 million in existing IQM cash, roughly $175 million from RAAQ’s trust account assuming no redemptions, $134 million in PIPE financing at $10 per share, and an estimated $24 million from warrant exercises, less $25 million in transaction expenses. American Depositary Shares are expected to list on either Nasdaq or NYSE, with a potential dual listing on the Helsinki Stock Exchange also under consideration. The transaction targets a Q2 2026 close, subject to shareholder approval and regulatory conditions. Upon completion, IQM becomes the first publicly traded European quantum computing company.

The Business

Founded in 2018 in Espoo, Finland, IQM has raised approximately €600 million in total capital. The most recent raise was a $320 million Series B in September 2025, led by Ten Eleven Ventures, with Goldman Sachs acting as sole placement agent.

IQM operates a full-stack model, meaning it designs and fabricates its own quantum chips, builds the control hardware, develops the software stack, and delivers and services systems on-premise. That vertical integration gives IQM tight feedback loops between hardware and software and reduces dependence on external suppliers. IQM operates its own cleanroom fabrication facility and can produce approximately 20 on-premise systems annually.

The product line runs from Spark, a 5-qubit system for universities and research institutes, through to the Radiance platform at 20, 54, and 150 qubits, which is the commercial workhorse targeted at high-performance computing centres. A 300-qubit system is planned for late 2026, designed as a testbed for quantum error correction, the process of using redundant physical qubits to simulate one reliable logical qubit and catch errors before they corrupt a computation. That system is destined for Finland’s national supercomputing infrastructure.

On architecture, IQM’s Constellation design tiles hexagonal modules across the chip, repeating a pattern of six qubits surrounding a central resonator to maximise qubit connectivity. More connectivity means error correcting codes run more efficiently. The roadmap targets hundreds of logical qubits by 2030, with million-qubit fault-tolerant systems as the longer-term ambition.

Commercial Traction

This is where IQM stands apart from most public quantum peers. The product line has three tiers. Spark is a 5-qubit on-premise system priced at around €1 million, aimed at universities and research institutes. Radiance is the commercial workhorse, available in 20, 54, and 150-qubit configurations, priced up to approximately €30 million for the most powerful version, and designed for integration into high-performance computing centres. Resonance is a cloud platform giving academia and enterprises remote access to IQM processors without needing to host hardware on-site.

IQM has sold 21 systems to 13 customers, four of which are among the top ten supercomputing centres globally, with deliveries across Germany, South Korea, Poland, Italy, Taiwan, Spain, and the United States. For 2025, IQM reported at least $35 million in unaudited revenue and over $100 million in bookings visibility. That said, $35 million in revenue against a $1.8 billion valuation is a significant multiple. The implicit assumption is that bookings grow, the 150 and 300-qubit cycles command larger contract values, and IQM locks in a leading position in institutional and government quantum infrastructure. The $450 million cash position provides meaningful runway to pursue that without near-term funding pressure.

What to Watch

Redemptions are the first risk. The RAAQ trust holds approximately $175 million, assuming no exits, but as Xanadu demonstrated this week, SPAC shareholders can and do walk. IQM has a $134 million PIPE backstop in place, but if redemptions run high, the $450 million headline cash figure becomes misleading. Beyond that, the 300-qubit delivery is the key technical milestone. The jump from 54 to 300 qubits is not linear in difficulty. More qubits means more points of failure, more demanding calibration, and more complex control systems. Slippage on timing or underperformance on gate fidelity would raise questions about the fault-tolerance roadmap more broadly.

On revenue, the $35 million in 2025 and $100 million in bookings visibility are encouraging, but the valuation demands a growth trajectory that compounds significantly from here. Large institutional and government contracts tend to move slowly, involve long procurement cycles, and carry political dimensions, particularly in Europe, where national quantum programmes are a factor. And then there is IBM, which has the deepest superconducting programme in the world, the largest installed base, and a developer ecosystem that dwarfs any pure-play competitor. IQM’s bet is that the on-premise, full-stack, open-architecture model serves a customer segment that IBM’s cloud-first approach does not fully address. That is a credible thesis for now. Whether it holds as IBM pushes further into HPC integration is an open question.

I think the issue Europe has faced across many sectors is a lack of investment and overregulation, with AI and data centres being the most recent example. But given the current dynamics, a push away from relying on the US for tech, amongst other things, will ramp up more than ever before. Given that all the major players in the quantum space are American, a thesis for investing in IQM is that they may receive quasi-governmental funding to keep them going should they run out of cash. Essentially, if they become important enough, the EU won't let them fail, and therefore, they would have deep enough pockets to build what they are trying to build. It's a longer-term view and a big if, given the EU's track record when it comes to these things.

Final Thoughts

Quantum computing is having an unprecedented public markets moment, with three companies announcing or completing listings in the first quarter of 2026 alone. The approaches are different, the valuations are significant, and the technology is still maturing across the board. None of these are straightforward investments, but the direction of travel is clear. This is a space worth watching closely, and we will be revisiting it as the milestones come in.

Thanks for reading. If you found this useful, feel free to share it with someone who might enjoy it

Share

I will be travelling next week, therefore I won’t be able to publish an article. The next article will be out on 12th April.

Disclaimer: This article is for informational purposes only and should not be considered financial or investment advice. The views expressed are my own and based on publicly available information, market trends, and personal analysis. Readers should conduct their own research and consult a financial professional before making any investment decisions.

Until Friday, the impact on equities had been somewhat contained, but the selling came in heavily on Friday with the NAS100 down ~1.5% and closing below its 200-day moving average. It’s hard to put my finger on what happens next as this move is entirely headline-driven. If Trump announces they’re done in Iran, then I believe markets will fly into the green, whereas if he decides to invade Kharg Island, this, in my opinion, will only be the beginning of the slide.

It’s easy to miss important developments in the tech space when the headlines are dominated by the ongoing war, so here’s what you may have missed. We will then get started on Micron’s latest earnings:

What You May Have Missed:

• SMCI co-founder was arrested for smuggling $2.5bn in Nvidia chips to China, slid at least 30% following the news.

• NVIDIA held its annual GTC conference, with Jensen projecting $1tn by 2027.

• Meta locked in a $27bn data centre infrastructure deal with Nebius Group. The Dutch cloud provider will supply $12bn of dedicated capacity starting in early 2027, with Meta committing to a further $15bn in additional capacity Nebius is building for third-party customers.

Now let’s dive into the latest figures from Micron.

Micron

Micron Technology is one of only three companies in the world that manufacture DRAM and NAND flash memory at scale, the other two being Samsung and SK Hynix. That oligopoly structure matters enormously for understanding why this company has become one of the most important stocks in the AI infrastructure trade. Micron designs and produces the memory chips that sit inside virtually every computing device on the planet, from smartphones and laptops to the most advanced AI servers being built today.

Micron reports across four business units: Cloud Memory (CMBU), which covers hyperscaler and AI server memory; Core Data Centre (CDBU), covering enterprise and traditional server demand; Mobile and Client (MCBU), covering smartphones and PCs; and Automotive and Embedded (AEBU), which serves automotive, industrial, and consumer electronics. For most of Micron's history, the mobile and PC segments drove the narrative. That is no longer the case.

This is because of High Bandwidth Memory, or HBM. Every Nvidia GPU that powers an AI data centre requires HBM stacked directly alongside it, delivering vastly higher memory bandwidth than conventional DRAM at lower power consumption. Each successive generation of Nvidia chip requires significantly more HBM than the last, and the architectural shift toward agentic AI and longer context windows is making every AI workload more memory-intensive over time. The result is a structural supply crisis. Producing HBM requires diverting cleanroom capacity away from standard DRAM and NAND, tightening supply across the entire memory market simultaneously. Micron currently holds roughly 24% of the HBM market behind SK Hynix and Samsung, but with its entire 2026 HBM supply already contracted and HBM4 now in volume production for Nvidia’s Vera Rubin platform, it is closing that gap fast. Understanding this dynamic is the key to understanding everything that follows in the financials.

Financials

Micron reported FQ2-26 revenue of $23.86bn, up 75% sequentially and 196% YoY. To put that YoY comparison in perspective, Micron did $8.05bn in revenue in the same quarter last year. In twelve months, quarterly revenue has nearly tripled. Every single business unit and every product line hit a new all-time high in the same quarter.

Breaking down revenue by technology, DRAM contributed $18.77bn, or 79% of total revenue, up 207% YoY and 74% sequentially. The growth was primarily price-driven. DRAM average selling prices rose in the mid-60s percentage range QoQ, with bit shipments only up mid-single digits. NAND contributed $5bn, or 21% of revenue, up 169% YoY and 82% sequentially. NAND average selling prices rose even more aggressively, up in the high-70s percentage range QoQ. This is the critical data point in the results. Volumes barely moved. What drove this quarter was pricing power, and pricing power comes from one thing. Demand that supply cannot keep up with.

By business unit, the data centre segments dominated. Cloud Memory (CMBU) was $7.75bn, up 163% YoY, with gross margins of 74%. Core Data Centre (CDBU) was $5.69bn, up 211% YoY, with gross margins also at 74%. Together, those two segments accounted for 56% of total revenue. Mobile and Client (MCBU) came in at $7.71bn, up 245% YoY, with gross margins of 79%, the highest of any segment. Automotive and Embedded (AEBU) was $2.71bn, up 162% YoY, with gross margins of 68%. The margin expansion across every single segment sequentially was dramatic, ranging from 23 to 25 percentage points QoQ, driven almost entirely by pricing rather than cost reduction.

Gross margin came in at 74.9%, operating income was $16.46bn for a 69% operating margin, and diluted EPS of $12.20 crushed the $9.31 consensus. Cash from operations was $11.9bn, representing 50% of revenue. Free cash flow of $6.9bn set a new quarterly record, surpassing the prior record set just last quarter by 77%. The balance sheet strengthened meaningfully, with net cash improving from $260m at the end of FQ1 to $6.5bn, and total liquidity sitting at $20.2bn at quarter end.

One question management did not address, and one worth watching closely given the current macro backdrop, is the exposure of Micron’s supply chain to helium. Semiconductor fabrication is heavily dependent on helium for cooling and inert atmospheres in cleanrooms, and with the Iran conflict disrupting regional supply chains and energy infrastructure, any tightening in helium availability could add cost pressure or operational risk that does not show up in today’s numbers. But it’s expected that the industry has adequate inventory for helium, so it isn’t a problem as of now.

Price Action

Shares dropped ~4% post the result, and given it was trading at ATHs, I think it’s mainly down to profit-taking amidst the war. A lot of the growth Micron is seeing is now baked into the share price. I missed the train with Micron, sadly, and I have no intention to catch a train that’s already left. I understand the importance of Micron amidst the current buildout, but in my opinion, the risk-reward is skewed against an entry now. I very well may be wrong, but I’d rather sit aside and miss out on more gains than forcing an entry I’m uncomfortable with and it melting down.

Outlook

Management guided FQ3 revenue of $33.5bn, gross margins of approximately 81%, and EPS of $19.15, with higher pricing, lower cost, and favourable mix all expected to contribute simultaneously to the margin expansion. On the call, CEO Sanjay Mehrotra confirmed that customers are still only receiving 50% to two-thirds of their requested supply, and that this tightness is expected to persist beyond calendar 2026. He also announced Micron’s first-ever five-year Strategic Customer Agreement, with discussions ongoing with multiple other customers. CFO Mark Murphy pushed back firmly on the idea that historical gross margin peaks in the low 60s % are a relevant reference point, arguing that AI has structurally changed the value of memory.

The demand driver here is compounding. Each new AI model generation requires more memory, agentic workloads are more intensive than inference alone, and context windows are only getting longer. New fab capacity will eventually close the gap, but Micron’s own sites in Idaho and New York do not deliver meaningful output until 2027 and 2028. For now, the supply data supports management's view that the tightness has further to run.

Final Thoughts

The bull case for Micron is well-supported by the data. The bear case is equally simple to articulate. Micron is a commodity business committing enormous capital at peak pricing, and when new supply from all three competitors arrives simultaneously in 2027 and 2028, the same pricing dynamic driving margins to 81% today can reverse hard. Memory has no moat. Cycles always turn. The question is not whether a downcycle comes, it is whether the structural AI demand floor is high enough to make the trough look nothing like 2023.

As for the broader picture, the war remains the dominant variable. Tech fundamentals are exceptional, but that matters little when oil is above $100, and rate hikes are back on the table. Micron just reported its best quarter in history, yet the stock fell. That tells you everything about where investor attention is right now.

Thank you for reading. If you found this useful, please share it with anyone you think would enjoy it.

Share

Disclaimer: This article is for informational purposes only and should not be considered financial or investment advice. The views expressed are my own and based on publicly available information, market trends, and personal analysis. Readers should conduct their own research and consult a financial professional before making any investment decisions.

Markets have remained nervy throughout the week, and Nvidia’s earnings did little to calm the jitters. In fact, it got swept up in it, despite a strong earnings report. There’s a lot of dispersion across equity markets, that is, some stocks are getting hammered, but the indices are staying relatively well put. If you look at DJ30 and the SP500, they haven’t moved much directionally since October 2025. NAS100 has also traded in a range, albeit wider. So, despite the headlines and the gloomy mood, things aren’t that bad. That is, unless your portfolio consists of software companies…

A rather dystopian report came out this week from Citrini, which spooked investors more than it probably should have. I haven’t read the entire piece, so I can’t comment much, but the author clearly states it’s just a scenario. It’s funny how the equity markets react, when there’s actually bad news, there seems to be little to no cares. But when a somewhat creative piece comes out on Substack, anything that includes AI is doomed and should be sold. Anyway, before I go off on a tangent, let’s dig into the latest Nvidia and IonQ earnings.

All financials will be Non-GAAP unless stated otherwise.

Nvidia

I think the market has gotten used to strong numbers from Nvidia, they need more. But how much more can they give after 4 years or so of crazy growth?

Financials

Q4 FY2026 revenue came in at $68.1bn, up 73% YoY and 20% sequentially, beating the $66.2bn consensus by nearly $2bn. EPS of $1.76 beat the $1.53 estimate, with net income reaching $43bn, up 94% YoY. Gross margin came in at 75%, ahead of the 74.8% guidance and up from 73.4% in Q3. The margin recovery the market had been watching for arrived cleanly. For the full fiscal year, revenue reached $215.9bn, up 65%, with operating cash flow of $102.7bn and free cash flow of $96.6bn. NVIDIA returned $41.1bn to shareholders through buybacks and dividends during the year and ended the quarter with $62.6bn in cash and marketable securities against $8.5bn in debt.

Data Centre was the driver, as it always is. Quarterly revenue of $62.3bn, up 75% YoY and 22% sequentially, representing 91% of total revenue. Full-year data centre revenue reached $193.7bn, up 68%, a business that has scaled nearly 13x since the emergence of ChatGPT in 2023. Networking within the segment was equally impressive, with quarterly networking revenue of $11bn, up more than 3.5x YoY, and full-year networking exceeding $31bn, up more than 10x versus the year Nvidia acquired Mellanox. Spectrum-X Ethernet is now running at roughly an $11-12bn annualised run rate and still growing. Sovereign AI revenue exceeded $30bn for the year, more than tripling, with Canada, France, the Netherlands, Singapore, and the UK among the primary contributors. Hyperscalers accounted for just over 50% of data centre revenue, with the remainder split across a diverse and expanding base of AI model makers, enterprises, and sovereign programmes, a customer mix that is broadening rather than concentrating. Gaming came in at $3.7bn, up 47% YoY, and Professional Visualisation was the surprise of the quarter at $1.3bn, up 159% YoY on exceptional Blackwell demand, well ahead of the $755m consensus. Automotive was the one soft spot at $604m, up 6% YoY but slightly below expectations.

The supply commitment figure is worth highlighting. Total supply-related commitments rose from $50.3bn at the end of Q3 to $95.2bn at the end of Q4, and management stated they have strategically secured inventory and capacity to meet demand extending into calendar 2027. CFO Colette Kress also noted that analyst estimates for 2026 capex across the top five hyperscalers are now approaching $700bn, up nearly $120bn since the start of the year, and that Nvidia expects to exceed the $500bn total chip manufacturing figure it had estimated for calendar 2026. The beat keeps beating the estimates that were already beating estimates.

One genuine risk that did not get enough airtime is the HBM shortage. Nvidia's Blackwell GPUs are inseparable from high-bandwidth memory, sourced from a small number of suppliers: SK Hynix, Samsung, and Micron. As data centre builders continue to scale, HBM supply has tightened, and this creates a margin unpredictability problem that sits entirely outside Nvidia's control. Kress acknowledged the company has inventory and supply commitments in place, but it is a constraint worth monitoring as the Blackwell ramp continues.

The agentic AI narrative on the call was the most developed we have heard from Jensen Huang. His framing was quite direct. In the new world of AI, compute is revenue. Without compute, there are no tokens, and without tokens, there is no growth. Whether you find that persuasive or promotional will shape how you think about the sustainability of the cycle. On the product side, Rubin samples have already shipped to lead customers, with volume production on track for the second half of 2026. Engineered to deliver a 10x reduction in inference token costs versus Blackwell, it matters enormously as the AI economy shifts from training to inference. The product roadmap has no visible gaps.

Price Action

Shares are down ~8% since reporting on Wednesday evening. On a P/E basis, it trades quite cheaply at 37x as of Friday’s close. I’m looking to add to my position around $170, but I will be watching how the momentum looks. If it’s really crashing past $170, I will hold fire. Although I’m investing for the longer term, I would rather get the best average price possible.

Outlook

Q1 FY2027 guidance came in at $78.0bn, plus or minus 2%, well ahead of the $72.1bn consensus, with gross margins guided to 75% through the April quarter and mid-seventies for the rest of the year. The Q1 guidance explicitly excludes any data centre compute revenue from China, a clean acknowledgement that the export control environment remains a headwind with no near-term resolution. H200 exports have been approved in small quantities, but as of quarter-end, Nvidia had generated zero revenue from them and does not know whether imports will ultimately be permitted. A resolution in either direction would move estimates meaningfully.

Sequential top-line growth through FY2027 is what management is pointing toward, with the Rubin ramp in the second half expected to sustain the trajectory. The Blackwell Ultra and Rubin roadmaps are overlapping rather than sequential, meaning there is no air pocket in the product cycle.

Michael Burry has drawn the Cisco parallel, and it is not an unreasonable one. Cisco was a genuine infrastructure monopoly during a real technological revolution and still fell 80% as the cycle turned. The difference is that Nvidia is profitable at a scale Cisco never was, and the AI cycle has multiple legs left in it. The question of what happens when hyperscalers have enough capacity will increasingly weigh on the multiple, but that is not a 2026 story, nor a 2027 one.

Now onto a different world of computing.

IonQ

A while back, I did a deep dive on IonQ, which you should check out for an overview of what IonQ does. Since then, they have acquired SkyWater Tech, a US-based foundry which I wrote about recently. This acquisition is strategically important, and I think it will be accretive to the business. Quantum as a space has been hit hard this year, with not many winners so far. This is attributed to, in my opinion, the more risk-off tone, which has led to investors realising their gains and rotating into safer bets.

Financials

Q4 2025 revenue came in at $61.9 million, up 429% YoY and more than 50% sequentially, beating the consensus of $40.4 million by 53% and exceeding the company’s own guidance midpoint by 55%. For the full year, revenue reached $130 million, up 202% YoY and 20% above the midpoint of IonQ’s own guidance range, making it the first publicly traded quantum computing company to exceed $100 million in annual revenue. Organic growth for the year was nearly 80%, which is the more important figure given the acquisitions. Adjusted EPS for Q4 came in at -$0.20, well ahead of the -$0.51 consensus. The full-year adjusted EBITDA loss was $186.8 million, better than guidance. Cash, cash equivalents, and investments at year end stood at $3.3 billion, providing substantial runway.

The revenue composition is also worth noting. More than 60% of 2025 revenue came from commercial customers, and international sales exceeded 30% of the total. That mix matters because it addresses one of the persistent criticisms of IonQ: that its revenue base was too concentrated in government and research contracts that would not scale. The commercial share growing to this level suggests the platform is being adopted outside of purely academic or defence settings. Remaining performance obligations exited 2025 at $370 million, up from $77 million at the end of 2024, a nearly fivefold increase that provides meaningful forward visibility and gives the 2026 guidance real credibility.

On the technology side, 2025 was a strong year. IonQ achieved 99.99% two-qubit gate fidelity, a world record and a significant step toward fault-tolerant quantum computing, where error reduction is the central challenge. The Tempo system hit its AQ 64 milestone three months ahead of schedule, a benchmark that indicates the system can evaluate over 18 quintillion possibilities simultaneously and address complex real-world algorithms beyond the reach of classical supercomputers. The first fifth-generation 100-qubit Tempo system was sold and delivered to KISTI in South Korea, anchoring the country's national quantum-classical compute platform alongside Nvidia acceleration. The next target is a 256-qubit demonstration in 2027, with the longer-term roadmap pointing toward 2 million physical qubits by 2030.

Price Action

had been under significant pressure heading into the print, down roughly 25% year to date on broader risk-off sentiment and uncertainty around the SkyWater acquisition. The post-earnings reaction was decisive, with shares surging around 22% to close around $40. The scale of a 53% revenue beat against consensus forces even sceptical investors to reassess, and short sellers who had piled in through February were squeezed out, amplifying the move. Despite the rally, the stock remains well below its 52-week high of $84.64 and the excitement that drove it there has visibly faded. I do not hold a position and am not looking to initiate one at current levels, but a pullback below $30 would get my attention, as a speculative investment, not a long-term one.

Management guided 2026 revenue of $225 to $245 million, a midpoint of $235 million, implying roughly 81% growth, ahead of Street expectations. The adjusted EBITDA loss is guided to $310 to $330 million, reflecting continued heavy R&D investment across the expanded platform. Both figures exclude any contribution from SkyWater, which has not yet closed.

The SkyWater acquisition is the most significant strategic development in IonQ’s history. The $1.8 billion cash and stock deal would give IonQ ownership of the leading U.S. quantum chip foundry, vertically integrating its supply chain and positioning it as a merchant supplier to the broader quantum industry. The strategic logic is particularly compelling for government and defence customers who need trusted domestic manufacturing. The deal is pending regulatory and shareholder approval, with a target close in Q2 or Q3 2026.

The open questions are the ones that have always surrounded IonQ. The path to profitability is distant, the valuation requires conviction on the commercial quantum timeline, and managing an increasingly complex platform is harder than operating a pure hardware business. None of that undermines what a genuinely strong quarter was, but it does mean position sizing here needs to reflect the risk honestly.

Final Thoughts

Both Nvidia and IonQ reported strong quarters, but sit at completely different points on the risk spectrum. Nvidia is the most profitable semiconductor company in history, still growing at a pace that should not be possible at its scale, and the bear case, while worth acknowledging, does not change the fundamental picture. IonQ is a different proposition entirely, with real commercial momentum and a genuinely impressive quarter, but profitability is years away, and the thesis requires patience. The debates surrounding both whether AI capex is sustainable and whether quantum will achieve commercial scale are questions that will define the next decade of technology investing. Nvidia stays a core holding. IonQ stays on the watchlist.

Share

Thanks for reading. If you found this useful, please share it with anyone you think would enjoy it.

Subscribe now

Disclaimer: This article is for informational purposes only and should not be considered financial or investment advice. The views expressed are my own and based on publicly available information, market trends, and personal analysis. I own NVDA shares, which may change at any time. Readers should conduct their own research and consult a financial professional before making any investment decisions.

Today’s article will take a look at two of the most important players within their given tech subsector. All figures highlighted are non-GAAP where applicable.

Cadence Design Systems (CDNS)

Cadence Design Systems is one of only two companies in the world, alongside Synopsys, that dominate the electronic design automation (EDA) market. In simple terms, EDA is the software and hardware used to design and verify semiconductors before they are manufactured. Without it, chips don’t get built. Cadence reports two segments: Product and Maintenance, which covers its core EDA software, semiconductor IP, and hardware systems; and Services, which encompasses implementation and other professional services. Within those segments, the business spans chip design software, verification hardware, pre-built IP building blocks, and system-level simulation tools. What makes Cadence particularly compelling right now is that the AI infrastructure buildout is directly fuelling demand for its tools. Every custom AI chip being designed by a hyperscaler, every next-generation GPU, every HBM memory stack requires EDA software to bring it to life. Cadence doesn’t build the chips, but it is an essential and largely unavoidable part of the process.

Financials

Q4 revenue came in at $1.44bn, up 6.2% YoY, beating consensus of $1.42bn. Within that, Product and Maintenance revenue was $1.3bn, up 7.5% YoY, while Services came in at $108m, down 7.8%, reflecting a deliberate shift toward annual subscription arrangements in the SDA segment that created noise in the year-over-year comparison. For the full year, revenue reached $5.3bn, up 14%. EPS of $1.99 beat the $1.90 consensus by $0.09, and full-year EPS came in at $7.14, up 20% on the year. Operating margin came in at 45.8% for the quarter and 44.6% for the full year, driven by productivity improvements across the business. Operating cash flow reached $553m in Q4 and $1.73bn for the full year, with free cash flow of $512m in Q4 and $1.59bn for the year. Cadence ended 2025 with $3.0bn in cash against $2.5bn in debt, and returned $925m to shareholders through buybacks during the year.

Across the product portfolio, performance was broad. Core EDA grew 13% for the year, with the recurring software segment re-accelerating to double-digit growth in Q4. The hardware business had another record year with 30+ new customers, and seven of the top ten buying both the Palladium and Protium systems. IP was the standout at nearly 25% growth for the year, driven by demand for HBM, UCIe, PCIe, and SerDes, all sitting right in the path of AI infrastructure investment.

One of the more interesting data points from the call was around how AI actually drives tool usage higher rather than lower. Management made the point that a manual user typically runs 3-5 design experiments in parallel, whereas agentic AI flows can run 10-100 experiments simultaneously, each of which calls Cadence’s underlying tools. More automation means more usage, not less, which is the right answer to the cannibalisation question that has been weighing on the entire software sector. The monetisation model for agentic flows is also worth noting. Rather than moving to pure consumption pricing, Cadence plans to price these as “virtual engineers,” a new revenue category on top of its existing subscription base, which, if executed well, could be a meaningful incremental growth driver. The company also highlighted that system companies, including hyperscalers designing their own chips, now represent roughly 45% of revenue, a mix shift Devgan described as irreversible and still accelerating.

The backlog number is the one that stands out most. Cadence exited 2025 with a record $7.8bn backlog, roughly 130% of annual revenue, with approximately 67% of 2026 revenue already covered from the starting backlog. That’s an unusual level of forward visibility and gives the guide real credibility.

Price Action

had been under pressure going into the print, down nearly 20% from its highs, with the entire EDA space selling off alongside broader software weakness. Synopsys had a similar experience, down nearly 19% over the same period. Post earnings, shares recovered ~3.5% at close with an initial after-hours move closer to 7%. As you can see from the chart, CDNS is not the smoothest ride, tending to pull back frequently and sharply, but with its moments of strong extension too. The valuation has come in meaningfully, with forward P/E now sitting around 35x based on the 2026 guidance midpoint, down from closer to 45x when the stock was trading above $350. I hold a position and am happy to continue doing so at these levels, given where in the value chain Cadence is positioned and would look to add on any further weakness.

Outlook

Management guided 2026 revenue of $5.9-6.0bn with an operating margin of 44.75-45.75% and EPS of $8.05-8.15. Q1 came in at $1.4-1.5bn revenue with EPS of $1.89-1.95, broadly in line with expectations. Operating cash flow is targeted at approximately $2.0bn, with roughly 50% of free cash flow allocated to buybacks. China is expected to remain at 12-13% of revenue, consistent with the last two years. H2 hardware pipeline visibility was flagged as less clear, normal for a hardware business at this point in the year, and management committed to updating the market mid-year as visibility improves.

Cadence also has the pending acquisition of Hexagon’s design and engineering business, carrying an annualised revenue run rate of around $200m, which is excluded from all 2026 guidance numbers. Once integrated, it extends Cadence’s reach further into physical AI and system-level simulation, areas that are becoming increasingly relevant as autonomous driving and robotics demand grows.

Overall, this was a clean quarter from Cadence with little to complain about, and with 67% of 2026 revenue already covered from starting backlog, the downside looks well protected. The more interesting question is whether agentic AI monetisation starts to show up more meaningfully through the year, which would be the catalyst to drive estimates higher.

Palo Alto Networks (PANW)

Palo Alto Networks is the largest pure-play cybersecurity company in the world, protecting over 70,000 organisations across network security, cloud security, and security operations. The business reports across two segments: Product, which covers hardware firewall appliances and related software, and Services, which covers subscriptions and support and is the dominant and faster-growing portion of the business. The key metric to watch is Next-Generation Security ARR, which captures the recurring revenue from cloud-delivered and AI-driven products across its three main platforms: Strata for network security, Prisma Cloud for cloud security, and Cortex for security operations. Within Cortex sits XSIAM, Palo Alto's AI-driven security operations platform that ingests telemetry data from across an organisation's entire environment, including firewalls, endpoints, cloud services, and identity providers, and uses it to automate threat detection and response in real time. It is arguably the most strategically important product in the portfolio, priced on data ingestion rather than seat count, and is the clearest expression of where Palo Alto wants to take the platform long-term. NGS ARR overall is the clearest signal of how well the platformization strategy is working.

Palo Alto recently closed two significant acquisitions: CyberArk, the leading identity security platform that manages and controls who and what can access critical systems, and Chronosphere, a cloud-native observability platform that helps organisations monitor and troubleshoot the health and performance of their infrastructure at scale.

Financials

Q2 FY2026 revenue came in at $2.59bn, up 15% YoY, just ahead of the $2.58bn consensus. EPS of $1.03 beat the $0.94 estimate by nearly 10%, and an operating margin reached 30.3%, the third consecutive quarter above 30%. NGS ARR grew 33% to $6.3bn, ahead of guidance, though four points of that growth came from the Chronosphere acquisition rather than organic momentum. Remaining performance obligation rose 23% to $16bn. Product revenue grew 22% to $514m, with software now representing 45% of product revenue versus 38% a year ago, a meaningful mix shift in the right direction. Services revenue grew 13% to $2.08bn. Adjusted free cash flow for the quarter was $502m, with the trailing twelve-month figure reaching $3.75bn.

The platformization data continues to impress. Net new platformizations reached 110 this quarter, up from 75 a year ago, and net retention for platformized customers came in at 119%. The number of two-platform customers grew over 50% YoY, and three-platform customers tripled. During the quarter, Palo Alto also closed its $25bn acquisition of CyberArk and the $3bn acquisition of Chronosphere, its two largest deals to date. CyberArk was funded primarily through a combination of cash and stock, with the share issuance meaningfully expanding the diluted share count to approximately 812-817 million shares going forward. The balance sheet reflects the weight of that activity, with cash ending the quarter at around $4.2bn and goodwill set to increase substantially as the CyberArk transaction is fully consolidated. Palo Alto carries minimal debt and is targeting a 37% adjusted free cash flow margin for the full year, which suggests the underlying cash generation remains healthy despite the near-term integration costs.

There is a broader point worth making here on AI and cybersecurity that I think the market is getting wrong. The bear case, that AI will automate threat detection and reduce security headcount, thereby shrinking budgets, misreads the supply and demand dynamics entirely. Cybersecurity talent has been structurally undersupplied for years, and automation addresses that gap rather than eliminating it. More importantly, pricing models are increasingly decoupled from headcount altogether. XSIAM is priced on data ingestion volume, not seats, and if AI turbocharges the frequency and intensity of threats, that drives more data ingestion, not less. The second bear argument, that AI labs could build security-focused models and displace incumbents, underestimates how data-dependent security really is. A generic LLM can understand security concepts, but it cannot replicate the real-time telemetry Palo Alto ingests across 70,000 customers from firewalls, endpoints, cloud services, and identity providers. That data is proprietary, it is real-time, and it is not replicable. We also saw with the shift to cloud computing that enterprises preferred third-party security tools precisely because they work across multi-vendor environments. With AI, the same dynamic is likely to play out: companies deploying multiple LLMs and needing an independent vendor to secure them, rather than buying security from the same AI labs selling them the models.

Price Action

has had a rough run, down nearly 33% from its highs of around $220 set in late 2025. The post-earnings reaction said it all. The stock beat on both revenue and EPS, yet fell 5-6% in after-hours as the market focused entirely on the Q3 EPS guidance of $0.78-0.80, well below the $0.92 consensus, citing integration costs from CyberArk and Chronosphere as the primary drag. The stock is now sitting just above the $140 support level that has held on multiple occasions, going back to 2024. That level is worth watching closely. A clean break below it would be a meaningful technical deterioration and likely bring in further selling. I do not currently hold a position in PANW, but I am watching it carefully at these levels. The fundamental case is compelling, and the valuation has come in significantly, but I want to see some stability around this support before considering an entry.

Outlook

For Q3, management guided revenue of $2.94-2.95bn, representing 28-29% growth, well ahead of the $2.60bn consensus, though the NGS ARR figure of $7.94-7.96bn includes a $1.47bn M&A contribution, so the organic picture needs to be read carefully. Full year FY2026 revenue is guided to $11.28-11.31bn, implying 22-23% growth, with an operating margin of 28.5-29.0% and full year EPS of $3.65-3.70. The company is targeting a 37% adjusted free cash flow margin for the full year, rising to 40% by FY2028, alongside a longer-term goal of $20bn in NGS ARR by FY2030.

The near-term EPS compression is real, and digesting two large acquisitions simultaneously is not trivial. That said, management noted that planning had been underway for months, with every CyberArk employee informed of their role within 48 hours of close, which suggests more readiness than a rushed deal would imply. The strategic rationale holds up, too. Arora’s framing on CyberArk was compelling: when AI agents start logging in at machine speed, identity becomes the primary attack vector, and Palo Alto now claims to be the only company that can verify the who and secure the what simultaneously. The next two quarters will determine whether that story translates into margin recovery. If it does, the current price looks very attractive in hindsight.

Final Thoughts

Both Cadence and Palo Alto sit in different corners of the tech stack, but they share a common thread. AI is a tailwind for both, and the market has been too quick to treat it as a threat. Cadence has the cleaner near-term picture, with a record backlog, strong execution, and a valuation that has come in to more reasonable levels. Palo Alto is the more complex story right now, with two major acquisitions to digest and a profitability dip that has spooked the market. But the platform is compounding, the strategic logic is sound, and if management delivers on margins through the back half of the year, the current price will look like an opportunity.

Thanks for reading. If you found this useful, please share it with anyone you think would enjoy it.

Disclaimer: This article is for informational purposes only and should not be considered financial or investment advice. The views expressed are my own and based on publicly available information, market trends, and personal analysis. I own CDNS shares, which may change at any time. Readers should conduct their own research and consult a financial professional before making any investment decisions.

What was most interesting this week was the sell-off in software as Anthropic released Claude Opus 4.6, which excelled at Excel tasks (pun intended), alongside capabilities clearly aimed at legal and finance use cases. The worries stemmed from the thought that AI will render SaaS businesses useless, which, to an extent, I agree with, but not anytime soon. But this is mainly the reason I don’t hold any software stocks; they have the least to offer in terms of growth and are the most structurally exposed to AI-driven changes.

There were 3 big earnings out this week, so I will cover them as crisply and concisely as possible! If you missed out on last week’s edition, make sure to check it out too, as it ties in with the bigger picture that I will cover in this article.

AMD

AMD was down bad post earnings. It was AMD’s largest single-day loss since 2022, and it was all because of the outlook. But before getting there, let’s have a look at the financials, which weren’t all that bad.

Financials

Revenue came in at $10.3bn, up 34% YoY, with gross margin at 54% GAAP and 57% non-GAAP, boosted by a one-off inventory reserve release tied to China MI308 shipments. Excluding that, margins were still around 55%, which suggests cost control and pricing held up. Operating income reached $1.75bn, with operating margin expanding to 17%, and non-GAAP EPS came in at $1.53. Execution wasn’t the issue.

Data Center did the heavy lifting, with revenue up 39% YoY to $5.4bn. EPYC continues to be the most dependable part of the story, benefiting from ongoing server share gains and rising CPU intensity alongside AI workloads. Instinct GPUs are the incremental growth lever within the segment, but remain in a ramp and investment phase, with meaningful acceleration pushed into the back half of the year. China MI308 revenue helped the quarter optically, but it doesn’t change the medium-term picture.

Client and Gaming revenue rose 37% YoY to $3.9bn, though the quality split matters. Client remains healthy, driven by Ryzen share gains and better commercial exposure. Gaming, however, is firmly late-cycle. The sequential slowdown in semi-custom reinforces the view that this segment will be a drag in 2026 rather than a contributor, which limits its relevance to the forward story.

Embedded was steady but uninspiring, growing in the low single digits. Long-cycle design wins still underpin the business, but near-term demand remains uneven, and margins are under mild pressure, keeping this segment in a holding pattern.

Cash flow was a clear positive. Operating cash flow came in at $2.3bn, with free cash flow of $2.1bn, both records. AMD exited the quarter with $10.6bn of cash and short-term investments, giving it balance sheet flexibility as AI-related investment ramps. Again, the numbers were fine. The market reaction was about confidence in the next leg of growth.

Price Action

Post reporting, was down 17%. It’s quite a large drop, but looking at the chart below, it wasn’t a catastrophic drop. Price is still above $185, a level it had struggled to break above previously. I don’t hold a position in AMD, but if shares were to drop below the $185 mark, I would be interested in starting a position.

Outlook

This is where the disappointment came from. AMD guided Q1 revenue to ~$9.8bn ±$300m with gross margin around 55%, implying sequential growth but no acceleration. Management expects Data Center to grow QoQ modestly, while Client, Gaming, and Embedded all decline, and China MI308 revenue steps down sharply from Q4. With Instinct GPU growth still back-half weighted, the near-term growth profile flattened, which was enough to reset expectations.

Alphabet (Google)

Here we arrive at my favourite holding. If you haven’t watched the documentary from the DeepMind team, I’d highly recommend it.

Financials

It was another fantastic quarter from Google. I will just refer to Alphabet as Google for simplicity. Revenue came in at $113.8bn, up 18% YoY, marking the first time annual revenues have pushed beyond $400bn.

Starting with Google Services, revenue grew to $95.9bn, up 14% YoY. Search was once again better than feared, growing 17%, which continues to push back on the idea that generative AI is cannibalising query volumes. If anything, AI appears to be expanding usage rather than compressing it. YouTube ads grew 9%, slower than Search but still healthy given tougher comps, while subscriptions, platforms and devices rose 17%, reinforcing that Google is steadily diversifying away from pure ad exposure. The takeaway here is that the core cash engine remains extremely robust.

Google Cloud was the standout. Revenue surged 48% YoY to $17.7bn, driven by strong demand for AI infrastructure and enterprise AI solutions. More importantly, operating income more than doubled to $5.3bn, pushing Cloud margins materially higher. This matters because Cloud is no longer just a growth story. It’s becoming a meaningful profit contributor at the same time as it absorbs a large share of AI-related capex. Other Bets remain a drag, with operating losses widening to $3.6bn, largely due to increased investment and a $2.1bn Waymo-related compensation charge. This isn’t new, but it does remind investors that Alphabet is still willing to fund long-dated optionality even when it weighs on near-term profitability.

Cash generation remained strong, with $52.4bn of operating cash flow in the quarter and $73.3bn of trailing free cash flow, even as capex stepped up meaningfully to support AI infrastructure. The balance sheet remains a fortress, with $126.8bn of cash and marketable securities, giving Google ample flexibility to continue investing aggressively without putting pressure on returns. Capex was the headline, with Google lifting its 2026 capital expenditure guidance to $175–185bn, far above consensus expectations of around $120bn, which spooked investors.

Price Action

Shares fell ~6% before buyers stepped right in. To me, this reiterated the confidence that investors have in Google. There are plenty of dip buyers, both retail and institutional, ready to nibble on any pullbacks. Myself being one of them. I will continue to add on any weakness as I think Google is extremely well placed, has, in my humble opinion, the best research and engineering team and will continue to do extremely well.

Outlook

Management didn’t provide explicit revenue guidance, but reiterated expectations for continued double-digit revenue growth in 2026. Advertising demand was described as stable, with no deterioration in Search trends, and management stated they expect Search growth to remain resilient, supported by AI-driven increases in query volume and engagement. Cloud visibility was described as improving, with management signalling sustained strong growth into 2026, albeit at a more normalised pace versus the 48% growth seen this quarter. On the call, management was clear that AI is not dilutive to Search economics, pushing back directly on concerns around cannibalisation, and instead framed AI as an incremental driver of usage and monetisation. Overall, Google still looks like a solid “bet”.

Amazon

As I have always done, I will stick to the tech side of Amazon rather than the entire business. It was a shaky start to the year for Amazon, led by our common theme of the week, capex.

Financials

Amazon reported a strong quarter overall, with total revenue of $213.4bn, up 14% YoY. North America revenue came in at $127.1bn and International at $50.7bn, up 10% and 17% YoY, respectively. Operating income for the group rose 18% YoY to $25.0bn, despite several one-off charges taken in the quarter.

AWS remained the core driver of profitability. Revenue grew 24% YoY to $35.6bn, marking a clear reacceleration, while operating income rose to $12.5bn, taking segment margins to roughly 35%. Growth was driven by AI workloads across both training and inference, with management noting that capacity constraints persist in parts of the business, indicating demand continues to exceed available supply.

Advertising was another bright spot. Ad revenue increased 22% YoY to $21.3bn, continuing to scale as a high-margin, tech-enabled business embedded within Amazon’s ecosystem and reinforcing the diversification of earnings beyond AWS alone.

Capex was the dominant talking point. Amazon plans to spend around $200bn in 2026, up more than 50% from roughly $130bn in 2025 and well ahead of market expectations closer to $150bn. Q4 capex alone came in at ~$38bn, materially above consensus, reflecting aggressive investment in data centres, custom AI chips, robotics, and satellite infrastructure (Blue Origin). Management was explicit that spending is being pulled forward to meet surging demand, with AWS having added 4 gigawatts of data centre capacity over the past year.

That investment intensity is flowing through cash metrics. Trailing twelve-month operating cash flow reached $139.5bn, but free cash flow fell to $11.2bn as spending ramped sharply. Amazon exited the quarter with a strong liquidity position, giving it the balance sheet capacity to absorb the build-out.

Price Action

fell by ~10% after earnings, but similar to Google, investors gobbled up part of the dip. Regardless, it is down ~15% for the year. I’ve not held Amazon as a position, and I struggle to see myself holding it, given that the bulk of the business is retail and not tech. But nevertheless, the AWS segment remains an important read-through for the broader cloud and AI infrastructure cycle, making Amazon a useful barometer even if it isn’t a core holding. Having bounced up off the $200 level, any break below could be interesting if I did decide to start a position in Amazon, which I won’t be doing.

Outlook

Out of all the hyperscalers, Amazon’s capex somewhat dwarfed the others. Management was explicit that demand for AWS capacity continues to outstrip supply, with AI workloads a key driver, and reiterated confidence that current investment levels are justified by long-term customer commitments. For Q1, Amazon guided group revenue of $173.5bn–$178.5bn, with the midpoint slightly above consensus, but flagged lower operating income as investment ramps. On AWS specifically, management pointed to a $244bn backlog, up from $200bn last quarter, and highlighted continued momentum across both training and inference workloads, while acknowledging that capacity constraints will persist in the near term. The message from the call was clear. Growth remains demand-led, but profitability will be secondary until capacity catches up.

Final Thoughts

Stepping back, this week felt less like a change in direction and more like a pause to reassess. The sell-offs weren’t driven by weakening demand or broken business models, but by the scale and timing of investment at a point where expectations had run hot. AI remains the dominant force shaping tech strategy, yet the path from spending to return is unlikely to be smooth or linear. For now, the market seems to be recalibrating patience, and that creates both noise and opportunity. I remain constructive on the long-term trajectory, but increasingly mindful that selectivity, timing, and discipline will matter more as this cycle matures.

Thank you for reading this article. I hope you found it insightful and gained value from it.

Disclaimer: This article is for informational purposes only and should not be considered financial or investment advice. The views expressed are my own and based on publicly available information, market trends, and personal analysis. I own GOOG shares, which may change at any time. Readers should conduct their own research and consult a financial professional before making any investment decisions.

Getting back to tech, I feel that the results so far have been strong, and there are three key names that I think are important to highlight. Let’s get into it.

ASML (4Q25)

For a recap of where we sit with ASML, check out my previous post. Shares have smashed through ATHs, reaching +€1,200. I think a lot of the growth is now baked into the share price, and the reaction post results did highlight this.

Financials

It was another strong quarter for ASML, with one KPI blowing expectations out of the park. But before getting to that, let’s go through the fundamental figures. Total net sales reached €7.5bn, down slightly QoQ but still up 16% YoY and broadly in line with guidance. System sales contributed €5.4bn, while the Installed Base Management segment delivered €2.1bn, continuing to provide a stable and increasingly important revenue base.

Gross margin came in at 51.6%, down QoQ from 53.7%, largely reflecting dilution from early High-NA EUV shipments, which continue to carry higher initial production costs. Operating margin was 32.8%, broadly unchanged YoY, while net income of €2.1bn highlighted the strength of the underlying operating model despite softer sequential sales.

The standout this quarter was bookings. Net bookings reached €13.2bn, an all-time high, driven by a sharp acceleration in EUV demand, which accounted for €7.4bn of the total. Demand looks strong across the board, with multiple drivers contributing. On the logic side, TSMC is adding incremental 3nm capacity while 2nm production ramps, increasing EUV intensity. Interestingly, memory was the key swing factor, accounting for more than half of the bookings, as the DRAM market remains tight on capacity. This supports expectations for additional fabs and tool orders over the next two years, while memory makers continue to migrate more DRAM layers to EUV to support high-bandwidth memory for AI workloads.

Backlog increased to €38.8bn, heavily skewed towards EUV, providing strong multi-year revenue visibility. The strength in orders reflects rising confidence in the durability of AI demand, increasing lithography intensity at advanced logic and memory nodes, and customers pulling orders forward to secure scarce EUV capacity rather than a short-term capex surge.

Management also announced a reduction of around 1,700 net roles, aimed at increasing the proportion of engineers while streamlining management layers. Large organisations often lose efficiency to lengthy command chains and internal process overhead, so a leaner structure should allow teams to spend more time on engineering and execution rather than internal coordination. Overall, the move reinforces ASML’s long-term focus on engineering excellence as the core driver of innovation and competitiveness.

Price Action

Shares peaked ~8% post the report but closed flat for the day. Currently trading at ~48x P/E, the highest it’s ever been, I do think entering a position here would be overpaying. Having added to my position on the breakout above €700, I don’t feel a need to add more at these levels. I think in the long run ASML will trade higher than this point, but I see no reason to increase my average entry price, so I will be waiting for a pullback to add. The business looks incredibly strong, and I think the memory squeeze will serve as a further tailwind. I’m confident they will be able to meet the level of demand implied by these results.

Outlook

Management guided to Q1 2026 revenue of €8.2bn–€8.9bn with gross margins of 51–53%, and full-year 2026 revenue of €34bn–€39bn, implying continued momentum into next year. The outlook is supported by increasing customer confidence in the sustainability of AI demand, which is driving firmer medium-term capacity planning rather than short-term ordering. Management also highlighted rising EUV layer intensity across advanced logic and DRAM, ongoing HBM-driven investment in memory, and strong demand for installed base upgrades as the fastest way for customers to add output, while High-NA systems continue to progress well through qualification.

While ASML’s results and guidance reinforced confidence in infrastructure-level AI investment, Microsoft’s earnings told a more cautious story.

Microsoft (2Q26)

For a recap on Microsoft & Meta, check out this post. Investors were rather unhappy with Microsoft’s latest results, and in fact, it was reminiscent of Meta's earnings from last year when they announced capex higher than expectations, which led to a +10% sell off.

Financials

Microsoft reported Q2 FY26 revenue of $81.3bn, up 17% YoY, driven by Intelligent Cloud and Productivity and Business Processes, which more than offset continued weakness in More Personal Computing. Intelligent Cloud revenue grew 29% YoY to $32.9bn, supported by strong Azure and AI-related demand, while Productivity and Business Processes increased 16% YoY to $34.1bn.

Margins came under pressure from AI investment. Gross margin percentage declined 1ppt to 68%, reflecting higher infrastructure costs and rising AI usage, while operating expenses increased 5% YoY on higher R&D spend and Gaming-related impairments. Operating income grew 21% YoY to $38.3bn, with operating margins expanding to 47%, though this was increasingly capital-intensive.

Below the line, results included a $10bn OpenAI-related gain, lifting GAAP EPS to $5.16. On a non-GAAP basis, EPS was $4.14, up 24% YoY, offering a clearer view of underlying earnings growth.

Cash flow was the main weak spot. Capital expenditure surged 66% YoY to $37.5bn, while free cash flow fell 9% YoY to $5.9bn, reflecting the scale of AI infrastructure investment. Microsoft returned $12.7bn to shareholders during the quarter, but investor focus remained on capex intensity, margin dilution, and the timing of AI monetisation.

Price Action

Has sold off ~10% after results, largely driven by concerns that capex is growing far faster than revenues, putting pressure on margins and free cash flow. I get that point, and it’s valid, but the results themselves weren’t that bad in my opinion. A lot of the negativity is centred on near-term cash flow dilution, slowing Azure growth QoQ, and uncertainty around the timing of AI monetisation. To me, these risks feel increasingly well understood and largely priced in.

What gives me comfort is how broad Microsoft’s business is. Even if AI doesn’t come through exactly as planned or takes longer to monetise, that compute isn’t stranded and can be redeployed across Azure, enterprise workloads, and the wider cloud stack. I haven’t held a position in Microsoft for a while, but with the stock now around 30% off ATHs, I’m strongly considering starting one here. The business is still performing well, and the sell-off feels more driven by sentiment and capex fears than a genuine deterioration in fundamentals.

Outlook

Management guided to Q3 FY26 revenue of $82–$85bn, implying mid-teens YoY growth, with Intelligent Cloud growth remaining in the high-20% range and Azure growth continuing to decelerate slightly QoQ as capacity constraints persist. Gross margin is expected to remain under pressure, with management flagging a low-to-mid-60s cloud gross margin profile as AI usage scales. Capital expenditure is set to remain elevated, with FY26 capex now expected to run well north of $100bn, driven by continued investment in data centres, GPUs and networking, keeping free cash flow suppressed in the near term. Management reiterated that AI demand remains strong, but revenue and margin capture will lag usage, reinforcing that this is a front-loaded investment cycle with returns pushed further out.

The main thing that worries me about Microsoft is OpenAI, but most hyperscalers/chipmakers have the same common denominator when it comes to exposure risk.

Now moving onto Meta.

Meta (4Q25)

Meta seems to always be in the headlines for negative reasons and is hatewatched by many, and to a point, I kind of get it. But regardless, it remains a portfolio position, so I want Meta to do well. I personally am more of a fan of the VR/AR & hardware products that they are building than the advertising side of things, but anyway. Investors were very happy with the latest set of results.

Financials

Meta delivered another strong quarter, with revenue up 24% YoY, driven almost entirely by advertising. Engagement across Reels continues to improve, and ad pricing moved higher again, showing that Meta’s core ad engine remains in good shape despite ongoing regulatory noise. Reality Labs revenue remains small but continues to benefit from growing interest in AI-enabled Ray-Ban glasses.

Operating margins remained robust at around 42%, with strong operating leverage in the Family of Apps offsetting rising AI and infrastructure costs. Unlike some peers, Meta has so far been able to absorb higher investment without materially diluting profitability, which helped drive the positive market reaction. Reality Labs contributed just under $1bn of revenue but remains loss-making, with quarterly losses of roughly $4.5bn. The hardware roadmap looks more credible than in the past, though it remains a long-dated and expensive bet.

Capex stayed elevated at just under $20bn for the quarter as Meta continues to front-load AI infrastructure investment. Management argues that this spend can be redeployed back into the core business if AI progress slows, but that assumption is where my concern lies. Meta is ultimately an advertising business, not a cloud provider, and it’s not obvious how all of that excess compute gets used if monetisation takes longer than expected. However, an important nuance is that not all of this capex is going into Nvidia GPUs. A growing share is being directed toward Meta’s custom silicon roadmap, MTIA, which management highlighted is now being expanded to support core ad ranking and recommendation model training. Moving these workloads off external GPUs and onto in-house silicon could drive meaningful cost savings over time, with MTIA chips estimated to be over 40% cheaper for internal workloads, partially offsetting the longer-term cost of Meta’s aggressive investment cycle.

Price Action

Jumped up by ~10% post results. I’m happy to hold here and perhaps add a bit more if the gap fills, as I can see Meta performing well this year. They have many important rollouts this year, such as the planned launches of next-generation AI models like “Avocado” and “Mango,” designed to advance text, image and video capabilities across Meta’s platforms and agentic commerce tools that aim to automate shopping and conversion workflows using AI agents, which could materially enhance both engagement and monetisation. And given the amount they’re paying the AI team, they'd better deliver strongly.

Outlook

Management struck a confident tone on the outlook, guiding to Q1 revenue of $53.5–$56.5bn, ahead of expectations, and pointing to continued AI-driven gains in engagement and ad performance. Meta now expects 2026 capex of $115–$135bn, nearly double 2025’s $72bn, as it accelerates spending on AI data centres, custom silicon and talent, alongside total expenses of $162–$169bn. Zuck framed this as a deliberate front-loading of infrastructure to support faster AI progress, with new models, products and monetisation avenues expected to roll out over the coming months. While the scale of spend is clearly aggressive, the market reaction suggests investors are currently willing to tolerate it, supported by record revenues, strong margins in the core ad business, and early signs that AI is already improving performance across Meta’s platforms.

Final Thoughts

Overall, this week’s earnings showed that AI demand is still very real, but the market is becoming much more sensitive to how companies are spending to capture it. ASML continues to sit in the strongest position structurally, with demand visibility improving and memory finally turning into a tailwind, though valuation now leaves little room for error. Microsoft feels like it’s being punished more for the size and timing of its investment than for any real deterioration in the business, which is why I’m becoming more constructive at these levels. Meta Platforms, on the other hand, is showing that heavy AI spend can still coexist with strong margins and cash generation, which explains why investors have been more forgiving.

Thank you for reading this article. I hope you found it insightful and gained value from it. Please share it with someone who may find it interesting!

Share

Disclaimer: This article is for informational purposes only and should not be considered financial or investment advice. The views expressed are my own and based on publicly available information, market trends, and personal analysis. I own both META and ASML shares, which may change at any time. Readers should conduct their own research and consult a financial professional before making any investment decisions.

In other news, earnings season has begun with most heavy hitters yet to report. But the two largest foundries have reported, with one of them causing a sell-off in the respective shares, whilst the other one continues to produce fine results. I don’t think it’s hard for you to guess which one’s which.

TSMC

For a recap of where TSMC is in its journey, check out this article. There weren’t any major surprises in the Q4 report, and the market reaction verifies that point, as there wasn’t much movement post-earnings. Nevertheless, let’s see where things are shaking out.

Financials

Total revenue came in at $33.3bn, up 5.7% QoQ and 20.5% YoY. Sequential growth was achieved despite wafer shipments declining 3% QoQ, underscoring that revenue growth was driven by richer node mix and improved pricing rather than higher unit volumes. On a full-year basis, revenue reached $122.4bn, up 35.9% YoY, driven by sustained AI-related demand at advanced nodes, with HPC workloads increasingly dictating both capacity allocation and pricing. Gross margin increased to 62.3%, up 280bps QoQ and 330bps YoY, coming in above guidance. The sequential improvement was primarily driven by higher utilisation at 3nm and 5nm, alongside ongoing cost optimisation efforts and favourable FX. These tailwinds were partially offset by margin dilution from overseas fabs, particularly in Arizona and Japan, which continue to ramp and remain structurally lower margin at this stage. On a full-year basis, gross margin averaged 59.9%, up 380bps YoY, reflecting both mix and scale benefits.

Cash generation remained robust. Operating cash flow totalled $23.4bn in Q4, while capex reached $11.5bn, resulting in free cash flow of $11.9bn for the quarter. For the full year, free cash flow reached $32.2bn, up 15% YoY, even as capex increased 33% YoY to $40.9bn, underscoring the durability of cash generation despite aggressive reinvestment.

By platform, HPC remained the largest contributor at 55% of revenue, down modestly from 57% in Q3. HPC revenue grew 4% QoQ, supported by sustained demand for AI accelerators and custom silicon, with the mix shift reflecting stronger seasonal smartphone demand rather than any softening in AI-related orders. Smartphone revenue increased to 32% of sales, up from 30% in Q3, driven by an 11% QoQ seasonal ramp in flagship processors on advanced nodes, particularly 3nm, ahead of year-end and early-2026 launches. As demand remains cyclical, the transition from 5nm to 3nm continues to support higher dollar content per device. IoT and Automotive each contributed 5% of revenue. IoT grew 3% QoQ as inventories stabilised, while Automotive declined 1% QoQ amid continued customer caution. Digital Consumer Electronics was the weakest segment, falling 22% QoQ, reflecting ongoing softness in mature consumer devices and legacy nodes.

From a technology perspective, advanced nodes continued to drive the mix shift. 3nm accounted for 28% of wafer revenue, up from 23% in Q3, reflecting improving yields and broader customer adoption. 5nm and 7nm contributed 35% and 14%, respectively, bringing total advanced-node contribution (7nm and below) to 77% of wafer revenue, up from 74% in the prior quarter. This shift toward leading-edge production remains the key structural driver of revenue growth and margin resilience.

So overall, solid growth across the board, but a lot of the good news/growth seems to already be baked into the price.

Price Action

Shares were up ~4% or so before closing the week flat. It seems to have been a pretty routine earnings report with nothing out of the ordinary announced. I think that TSMC will continue to benefit from the AI buildout. However, I am starting to grow concerned about geopolitical tensions, which I won’t go into too much. But I do believe that within the next few years, China will attempt to invade Taiwan, and although trying to figure out the timing of this is virtually impossible, unless you have insider info, I don’t want to give back my profits when/if this happens. But this is more of a big picture view and doesn’t affect my near-term investments into TSMC, therefore, I will continue to add on any weakness.

Outlook

TSMC guided Q1 FY26 revenue of $34.6–35.8bn, with gross margin of 63–65% and operating margin of 54–56%, pointing to continued strength in advanced-node utilisation into the start of the year. Management reiterated expectations for close to 30% revenue growth in 2026 and raised full-year capex guidance to $52–56bn, reflecting clearer demand signals at the leading edge and continued build-out of advanced packaging and backend capacity.

At the same time, the tone on the call was deliberately cautious. TSMC CEO C.C. Wei said he is “nervous” about AI, cautioning against extrapolating near-term momentum too aggressively. While customer commitments and tape-outs remain firm, Wei emphasised that capex remains tightly linked to confirmed demand rather than speculative expansion, underscoring management’s awareness of cyclicality even as AI continues to drive capacity allocation and pricing. This reassures me, as an investor, that they are not raising capex to meet speculative demand.

Intel

It was a much more eventful earnings report for Intel, for the worse. I think the market reaction was the result of overly optimistic expectations around Intel’s turnaround. I personally don’t think the results were bad enough to merit a ~15% sell-off, but because of the enthusiasm surrounding Intel, the poor outlook somewhat brought investors back to the reality that Intel is still in a turnaround, not out of it, just yet.

Financials

Intel’s Q4 came in modestly ahead of both Street expectations and its October guidance, though the quarter's quality was shaped as much by supply decisions as by end demand. Revenue came in at $13.7bn, down 4.1% YoY, with non-GAAP gross margin of 37.9% and non-GAAP EPS of $0.15, all slightly ahead of expectations. Management was clear that demand exceeded what Intel could physically ship during the quarter, particularly in Data Center, with internal wafer supply prioritised toward servers and a greater reliance on externally sourced wafers in Client to balance the network.

That trade-off showed up clearly in the segment mix. Client Computing Group revenue was $8.2bn, down ~3.5% despite AI PC unit volumes growing 16%, indicating that supply rather than demand was the limiting factor. Data Center and AI was the standout, with revenue of $4.7bn, up 15% QoQ, which Intel described as its fastest sequential growth in over a decade. Management said results would have been higher with more available supply, and highlighted continued momentum in custom silicon, with custom ASIC revenue reaching an annualised run rate above $1bn.

Intel Foundry revenue rose to $4.5bn, up 6.4% QoQ, driven by a richer EUV wafer mix. However, operating losses widened to $2.5bn as Intel absorbed the cost of the early 18A ramp. Management framed this as expected ramp inefficiency rather than demand weakness and highlighted that Intel is now shipping revenue generating products using gate all around transistors and backside power delivery. While margin dilutive in the near term, this represents a meaningful execution milestone. For those not familiar, Gate-all-around transistors are a new chip design that lets Intel control electricity more precisely as chips get smaller, while backside power delivery feeds power from underneath the chip instead of the top, making it faster and more energy-efficient.

Intel generated $4.3bn of operating cash flow in Q4 and spent $4bn on gross capex, resulting in $2.2bn of adjusted free cash flow once government incentives and partner funding were included. Net income was $1.6bn for the quarter, though profitability remains constrained by margin pressure and restructuring charges.

Price Action

Dropped ~15% post results and it was mainly due to the weak outlook provided, which I will get to shortly. I’m happy to hold Intel here despite the execution troubles and ongoing turnaround. My reasons are mainly big picture-based - moving chip manufacturing back to the US, which I’m sure President Trump will put as much pressure as he can to make this happen. It’s not a large position in my portfolio, and I think at current prices it is slightly overvalued, therefore I will only be adding sub $40.

Outlook

The big issue in the eyes of investors is the outlook. Intel guided Q1 revenue to $11.7–12.7bn, below Street expectations, reflecting supply constraints and ongoing challenges ramping its 18A manufacturing process. Management said late-2025 demand exceeded what Intel could ship, but constrained yields, meaning a lower proportion of functional chips per wafer, are pushing up unit costs and limiting near-term upside.

Against that backdrop, Intel highlighted the launch of Panther Lake, its first major PC processor built on 18A, as an important technical milestone. The chip is focused on power efficiency and on-device AI for thin-and-light laptops, but its significance is strategic rather than financial, serving as early validation of the process. Management was clear that meaningful yield improvement remains the gating factor for margin recovery and for the credibility of Intel’s manufacturing turnaround as it looks ahead to 14A.

Final Thoughts

These two earnings reports underline how bifurcated the foundry landscape has become. TSMC continues to execute from a position of strength, with pricing power and demand visibility at the leading edge, albeit with much of the upside already reflected in the share price and an unavoidable geopolitical overhang. Intel, by contrast, remains in rebuild mode, where progress is visible but execution risk, particularly around yields and manufacturing economics, still dominates. From a portfolio perspective, TSMC offers quality and consistency, while Intel functions more as a long-dated hedge on US chip manufacturing and supply chain sovereignty, a theme likely to gain further political backing under Donald Trump, even if the path there remains volatile.

As always, thank you for taking the time to read this article. I hope you gained value from it. Please share this article with anyone who may be interested, as it helps the page grow.

Share

Disclaimer: This article is for informational purposes only and should not be considered financial or investment advice. The views expressed are my own and based on publicly available information, market trends, and personal analysis. I own both Intel and TSMC shares, which may change at any time. Readers should conduct their own research and consult a financial professional before making any investment decisions.

This article, rather than being a deep dive on a single stock, will be a higher-level look at two companies within the Russell 2000. For those unfamiliar, the Russell 2000 tracks 2,000 US small- and mid-cap companies that tend to be earlier in their growth cycles and more sensitive to changes in financial conditions. I find the index interesting amid this rate-cutting cycle for two reasons. Firstly, lower rates = lower borrowing costs. These companies can access cheaper financing to supplement growth. Secondly, I’m looking further down the market-cap spectrum because the AI ecosystem is still evolving. As systems scale, new and more specific technical constraints emerge, and many of the solutions to these problems are being developed by smaller, more specialised companies. Without further ado, let’s begin!

SkyWater Tech

SkyWater Technology was founded in 2017 following the carve-out of Cypress Semiconductor’s Minnesota fab, a facility with roots going back to the 1980s. The business was formed by Abbas Khan, alongside a group of industry veterans, with the aim of creating a US-based, trusted semiconductor foundry focused on specialised manufacturing rather than leading-edge logic.

SkyWater specialises in advanced technology services and specialised process nodes, including radiation-hardened chips, secure manufacturing, MEMS, mixed-signal technologies, and quantum computing devices. Its model is very different from large foundries like TSMC. Rather than chasing scale or the most advanced nodes, SkyWater focuses on low-volume, high-complexity production for customers in aerospace, defence, automotive, medical, and industrial end markets.

On the quantum side, SkyWater manufactures superconducting and advanced materials-based chips used in quantum research and early commercial systems, positioning it as a domestic fabrication partner for next-generation computing technologies that require tight process control rather than extreme scale.

SkyWater IPO’d in April 2021, listing on Nasdaq during a period of heightened focus on supply-chain resilience and US semiconductor sovereignty. Since then, SkyWater has positioned itself as a growing domestic manufacturing partner, particularly as governments and enterprises look to reduce reliance on overseas fabs for sensitive or mission-critical applications.

What stands out to me is SkyWater’s positioning as it sits at the intersection of reshoring semiconductor supply chains and more specialised technologies such as secure manufacturing and quantum computing, areas where scale matters less than process control.

Products

SkyWater’s offering centres on semiconductor development and manufacturing, rather than mass-market end products. Its Technology-as-a-Service model combines process development with in-house manufacturing, allowing customers to move from design to production within a single, secure domestic foundry. Advanced Technology Services focus on specialised applications such as radiation-hardened defence chips, superconducting devices for quantum computing and sensing, MEMS, integrated photonics, and advanced imaging technologies. Radiation-hardened chips are built to function in extreme environments where radiation would disrupt standard semiconductors, while superconducting devices used in quantum computing operate at cryogenic temperatures and require highly precise fabrication.

SkyWater operates through two reportable segments. Legacy SkyWater covers core technology foundry operations, providing Advanced Technology Services and Wafer Services from the Minnesota fab, alongside advanced packaging from the Florida facility. SkyWater Texas is a higher-volume manufacturing operation following the acquisition of Fab 25 in Austin. This segment focuses on 200mm wafer manufacturing, with capabilities in high-voltage, copper processing, and mature-node production, primarily serving automotive and industrial end markets.

Financials

SkyWater reported $150.7m of revenue in the third quarter of 2025, up from $93.8m in the prior-year period, reflecting the contribution from the Texas fab acquisition, which was completed on June 30, 2025. Gross profit was $36.2m, implying a gross margin of 19.9%, consistent with a mix weighted toward development-led ATS work and an early-stage ramp in higher-volume manufacturing. Total assets increased to $787.0m as of September 28, 2025, driven primarily by property, plant and equipment. Liquidity stood at $30.9m of cash and cash equivalents, with $65.0m of availability under a $350m revolving credit facility.

On cash flows, SkyWater generated a $12.1m increase in cash over the first nine months of 2025. Financing activity was the primary driver, with $111.9m of net cash provided by financing, reflecting the completion of the Fab 25 acquisition on June 30, 2025, revolver usage, and equity-related inflows. Capital intensity remained elevated, with $20.5m of capital expenditure in Q3 and $10.0m of capital expenditures incurred but not yet paid at period end, largely tied to fab upgrades and advanced packaging investments.

Price Action

As the chart shows, post-IPO enthusiasm faded quickly, with the stock trading as low as $4.3 before spending roughly two and a half years range-bound. Recently, has broken out of this range, suggesting renewed interest which is why I added a small position in November. A move back toward prior highs would require continued execution and improving sentiment toward US semiconductor manufacturing.

Conclusion

SkyWater is well-positioned to benefit from increased investment in US semiconductor manufacturing, particularly as lower interest rates improve funding conditions for capital-intensive growth and supply chains continue to shift onshore. As geopolitical tensions around Taiwan continue to build over time, the likelihood of a more forceful US push toward domestic fab capacity increases, which would materially accelerate investment into the sector. That said, SkyWater remains a small operator today with a relatively modest revenue base, and the investment case still rests on execution as capacity and demand scale.

Credo Tech

Credo was founded in 2008 by Bill Brennan and Paul Master, with a focus on solving signal integrity challenges in high-speed data transmission. The business was built around a simple but increasingly important problem: as data rates rise, moving data reliably and efficiently across copper and optical links becomes harder, more power-intensive, and more expensive.

Credo specialises in designing high-speed connectivity silicon and intellectual property for data centre and cloud infrastructure. As AI clusters grow and bandwidth requirements increase, traditional electrical connections struggle to maintain performance efficiently. Credo’s designs improve signal integrity and reduce power consumption across copper and optical links, enabling higher bandwidth without a proportional increase in energy use.

Credo IPO’d in January 2022, listing on Nasdaq during a period of accelerating cloud and AI infrastructure investment. Since then, Credo has increasingly been positioned as an enabler of next-generation data centre architectures, particularly as hyperscalers seek to scale bandwidth within fixed power envelopes.

I like Credo because it addresses a growing bottleneck in AI infrastructure: moving data efficiently at scale, where connectivity and power efficiency increasingly matter as much as compute.

Products

Credo’s products focus on high-speed serial connectivity, spanning optical DSPs, retimers, Active Electrical Cable (AEC) solutions, and SerDes IP. These designs sit between compute, memory, and networking hardware, supporting reliable data movement across servers and racks at very high speeds.

At a basic level, electrical signals weaken and distort as data rates increase and signals travel through cables. Credo’s technology addresses this by using digital signal processing to clean and reshape signals in real time, allowing higher speeds and longer reach without errors. SerDes designs convert data into efficient high-speed serial streams and back again, enabling reliable communication while keeping power consumption lower than alternative approaches.

Optical DSPs are deployed in AECs and optical modules inside data centres, while SerDes IP is integrated directly into custom silicon designed by hyperscalers. In this context, optical refers to data being transmitted using light rather than electrical signals, typically through fibre-optic cables, embedding Credo’s technology deeply into customer platforms through both silicon and IP.

Credo reports revenue across two segments. Product revenue consists of sales of connectivity silicon, including optical DSPs used in Active Electrical Cables (AECs) and optical modules deployed in data centres. IP revenue consists of licensing SerDes intellectual property into customer-designed ASICs, where Credo’s designs are integrated directly into hyperscaler silicon.

Financials

Credo Technology Group reported $268m of revenue for the quarter ending November 2025, up from $72.0m in the prior-year period, driven overwhelmingly by growth in Product revenue, which reached $261.3m versus $69m a year earlier. This reflects the accelerating deployment of Credo’s connectivity silicon across hyperscale data centre networks. IP revenue contributed $6.7m, broadly stable year-on-year, as SerDes licensing remains a smaller but higher-margin component of the mix.

Gross profit for the quarter was $181.0m, implying a gross margin of 67.5%, benefiting from scale and a product mix increasingly weighted toward higher-volume deployments. Operating income reached $78.8m, marking a clear inflection from operating losses in the prior year and demonstrating operating leverage as revenue ramps.

The balance sheet remains strong. Cash and cash equivalents stood at $567.6m, alongside $246.0m of short-term investments, taking total liquidity above $800m, with no outstanding debt. On cash flows, operating cash flow reached $115.8m over the six-month period, while cash balances increased by $331.2m, supported by improved profitability and equity issuance. Overall, the financial profile reflects a capital-light model with strong margins and significant balance-sheet flexibility as data centre investment continues to scale.

Price Action

performed strongly during 2025 but has since retraced roughly 30% from its highs of just under $215. I’m not currently in a position, but it remains on my watchlist, especially if the stock were to trade below $125. Despite being a small-cap name, Credo has a strong balance sheet and rapidly growing revenues. With data centre capacity set to expand materially over the next five years, and Credo’s products addressing a core connectivity bottleneck, continued growth looks achievable provided demand remains intact.

Conclusion

Credo stands out as a focused play on the scaling challenges inside modern data centres, where connectivity and power efficiency are becoming binding constraints rather than secondary considerations. With a capital-light model, strong margins, and deep integration into hyperscaler platforms through both silicon and IP, Credo is well-positioned to benefit as AI infrastructure continues to expand.

Final Thoughts

SkyWater and Credo offer two very different but complementary ways to gain exposure to the next phase of semiconductor and AI infrastructure investment. SkyWater provides leverage to reshoring, trusted domestic manufacturing, and specialised technologies where process control matters more than scale. Credo, by contrast, is a capital-light play on the growing connectivity bottlenecks inside data centres, with strong margins and balance-sheet flexibility. Both sit further down the market-cap spectrum, where execution matters, but where the payoff can be meaningful if these structural trends continue to play out.

TSMC report earnings next week, kicking off a busy earnings season. Stay tuned for my thoughts and subscribe if you haven’t already!

Disclaimer: This article is for informational purposes only and should not be considered financial or investment advice. The views expressed are my own and based on publicly available information, market trends, and personal analysis. I own SKYT shares, which may change at any time. Readers should conduct their own research and consult a financial professional before making any investment decisions.

I’m not going to dwell much on what has happened in 2025, as it’s in the past now, but to know where we go next, it’s important to have a recap of what got us to this point.

2025 Review

The first major shock of 2025 came from DeepSeek. Its release showed that competitive reasoning models could be trained at a fraction of the cost markets had assumed, undermining the belief that AI leadership required ever-rising capital intensity. That single event triggered a sharp sell-off across the AI supply chain as investors reassessed returns tied purely to training spend. The implication wasn’t weaker AI demand, but a shift in where value sits, with efficiency, deployment and margins becoming more important than brute-force compute.

Macro pressure then reinforced the move. In April, Liberation Day tariffs reintroduced uncertainty around global supply chains, China-linked revenues and capex just as AI investment was accelerating. This gave rise to the TACO (Trump Always Chickens Out) trade, as repeated tariff threats were followed by pauses across several countries in the months that followed. That dynamic allowed markets to recover and extend the existing bull run. At the same time, scale became visible elsewhere. OpenAI moved decisively into industrial mode, committing to tens of billions of dollars per year in compute and infrastructure during 2025, with long-dated contracts running into the low hundreds of billions across Microsoft, Oracle and data centre partners. This locked in large, long-dated commitments, but also raised concerns around concentration risk, given how dependent the ecosystem has become on OpenAI despite its revenues being tiny relative to the scale of those contracts. Technologically, the focus also shifted from training to inference, where cost per query, power efficiency and latency dominate, broadening the opportunity set beyond GPUs into TPUs and custom accelerators.

All in all, it was a great year for equities, mainly because it was a great year for tech stocks.

Key Index Performances (rounded flat):

• NASDAQ 100: +20%

• S&P 500: +16%

• DJ30: +13%

• Russell 2000: +11%

2026 Outlook

Earnings

Earnings this year will be very interesting and important, given the high bar set last year alongside the large amount of capex being spent. I think it will be a volatile year on that front with some major misses within the sector but also major beats. In terms of which companies will provide these, I wish I had a crystal ball but I don’t so I can’t say. I am confident that Nvidia, TSMC, Alphabet and Microsoft will continue to beat estimates and deliver significant YoY growth. I’m not so sure about Meta and Oracle doing the same. The universe is, of course, broader than these names, but they are a good proxy for demand.

Revenue growth will remain the primary driver of earnings outcomes in 2026, particularly given how central AI demand has become to expectations across the sector. However, the quality of that growth will matter more than it has in previous years. With capex having ramped meaningfully only recently, depreciation isn’t likely to rise sharply straight away, but it will begin to feed through over the coming quarters as new infrastructure is brought into service. That puts more emphasis on whether revenue growth can scale alongside the expanding asset base. Margins and operating leverage will therefore play a larger role in how results are received, not because costs are about to surge, but because expectations already assume smooth execution. In that environment, even relatively small shortfalls on growth, margins or guidance can lead to outsized market reactions, which is why volatility around earnings is likely to remain elevated.

Beyond the headline numbers, guidance will likely matter more than usual. With expectations already stretched, markets will be particularly sensitive to any commentary around incremental capex, timelines or returns on investment. We saw this clearly in 2025 with Meta, where strong revenue and earnings results were overshadowed by guidance pointing to materially higher AI and infrastructure spend. When Meta raised its full-year capex outlook to roughly $70–72bn and signalled that investment intensity would remain elevated, the stock sold off sharply despite the underlying business continuing to perform well. That reaction came against a backdrop where the hyperscalers had already committed enormous sums, with Microsoft, Alphabet and Amazon collectively running well over $150bn of annual capex in 2025, largely tied to AI and cloud infrastructure. In that context, tolerance for any further escalation in spend remains extremely low going into 2026. Unless revenues have grown materially by that point.

ASICs

ASICs are becoming more relevant as AI workloads shift toward inference, where cost per token, power efficiency and serving latency matter more than flexibility. As a reminder, inference refers to running trained models in production to serve real-world queries, rather than training new models. Hyperscalers aren’t developing custom silicon because GPUs are inadequate, but because predictable, repeatable inference workloads are better served by chips designed for a narrower set of operations. Google’s TPUs and Amazon’s Trainium and Inferentia allow these companies to internalise a growing share of AI demand, reduce reliance on external suppliers, and lower long-run compute costs by controlling the full software and hardware stack. This does pose a threat to GPU market share at the margin. As more inference is served in-house on ASICs, incremental demand that would previously have translated into additional GPU orders increasingly doesn’t. That said, GPUs remain critical for frontier training and for workloads where flexibility and a mature ecosystem still matter.

While hyperscalers design and deploy ASICs internally, much of the economic upside sits with the suppliers that enable these systems at scale. Broadcom is the clearest beneficiary, acting as a key partner in custom AI accelerator and networking silicon for hyperscalers, monetising the shift toward ASICs without competing directly with GPUs. Marvell plays a similar role through custom compute, high-speed networking and optical interconnects, areas that become more critical as inference workloads scale and data movement, rather than raw compute, turns into the bottleneck. This is reinforced further up the stack by networking vendors like Arista Networks, which benefit from rising bandwidth requirements regardless of the underlying accelerator, and by memory suppliers such as Micron and SK Hynix as inference and custom silicon continue to drive demand for high-bandwidth memory. At the manufacturing layer, TSMC remains a structural winner across GPUs and ASICs alike, while EDA providers such as Synopsys and Cadence quietly benefit from rising custom silicon complexity. The net effect is that the shift toward ASICs doesn’t eliminate spending, it reallocates it, away from off-the-shelf accelerators and toward a concentrated group of companies that design, connect, memory-enable and manufacture the AI stack. The more likely outcome for 2026 isn’t displacement, but a split, with GPUs retaining dominance in training while ASICs steadily absorb a larger share of hyperscaler inference, shaping where incremental AI spend ultimately flows.

My Watchlist

Quantum

Quantum computing remains a long-term theme, with Google and IBM clearly leading on research and infrastructure. However, from an investment perspective, I’m increasingly focused on the smaller, pure-play names such as IonQ and Rigetti, where valuations are far more sensitive to genuine technical progress. These companies carry higher risk and limited near-term revenues, but they offer more asymmetric upside if milestones around qubit stability, error correction and early enterprise use cases continue to be met. For 2026, quantum sits outside the core earnings debate, but remains an area of long-term optionality worth monitoring. Below are names on my watchlist:

Semi Manufacturers & Equipment

Despite the strong rally across semiconductors and equipment this year, this remains an area where I’m comfortable adding on weakness. Demand for leading-edge chips continues to be structurally strong, driven not just by AI, but by the broader push toward more advanced logic architectures and manufacturing nodes. The transition toward next-generation process technology, including continued FinFET optimisation and the move to 2nm-class nodes, materially raises complexity, cost and tooling intensity. That benefits both chip manufacturers and the equipment suppliers that enable these transitions. Valuations are no longer cheap, but any cyclical pullback would be an opportunity to build exposure, given how central advanced semiconductors remain to AI, cloud and compute roadmaps over the coming years.

Watchlist

Final Thoughts

Overall, 2025 was a strong year for the Tech Investor portfolio, reflecting both the strength of the underlying themes and a favourable backdrop for tech equities. Looking ahead, I’m comfortable adding to positions in Google, Applied Materials and Broadcom on any meaningful weakness, where long-term fundamentals remain intact despite near-term volatility. I remain more cautious on Meta and Oracle (which isn’t a position anyway), given elevated expectations and sensitivity to further capex escalation. While hedging remains important, being outright short can be dangerous in structurally strong markets, so I prefer dynamic protection through exposure to more defensive assets such as metals and bonds. At present, I’m holding a higher cash balance, not to time the market, but to retain flexibility to deploy capital during periods of dislocation rather than adding aggressively at current levels. 2026 will almost certainly be another volatile year, but that volatility is where opportunity tends to emerge. Bring on 2026!

Once again, thank you for your support, it’s really appreciated! Feel free to reach out with any questions, and please continue to support my work by sharing The Tech Investor with colleagues, friends and family!

Disclaimer: This article is for informational purposes only and should not be considered financial or investment advice. The views expressed are my own and based on publicly available information, market trends, and personal analysis. Readers should conduct their own research and consult a financial professional before making any investment decisions.

Marvell Intro

Marvell was founded in 1995 and spent most of its early life as a broad communications and storage silicon supplier, serving hard-drive controllers, consumer networking and early Ethernet devices. The business evolved meaningfully over the past decade as management repositioned its portfolio toward data-centre infrastructure, moving into optical connectivity, switching silicon and eventually custom compute. Several acquisitions shaped this transition, most notably Cavium in 2018 and Inphi in 2021, which gave Marvell a leading position in PAM4 optics that now sits at the heart of high-performance AI networks. PAM4 is a way of sending data that uses four signal levels instead of two, which lets far more information travel through the same connection. It has become important because AI clusters need much faster links between chips.

Marvell designs the connective and acceleration layers used in modern data centres. Its optical DSPs (Digital Signal Processor) and retimers move data between GPUs and CPUs at very high bandwidths, its merchant switching silicon underpins large training fabrics and its PCIe and CXL connectivity products allow memory, accelerators and compute nodes to communicate efficiently. Marvell also co-develops custom ASICs with hyperscalers for their in-house accelerators. These designs are used at AWS, Microsoft, Google and Meta, whose cluster roadmaps require increasing levels of bandwidth and specialised compute as model sizes grow.

Marvell’s relationship with Amazon has become one of its most important strategic anchors. Amazon has been investing heavily in its own in-house silicon, including Trainium for training and Inferentia for inference, and Marvell plays a key role in enabling these platforms. Marvell supplies custom silicon that supports Amazon’s accelerator roadmap, along with high-speed optical and electrical interconnect that links these chips together inside the data centre. As Amazon expands its training clusters and moves toward larger, more power-efficient architectures, the demand for richer bandwidth, tighter latency control and specialised connectivity continues to grow, all of which align directly with Marvell’s strengths.

Despite this, it’s been a rollercoaster ride for the stock this year as shares peaked earlier in the cycle and then experienced a substantial sell-off as expectations proved too ambitious for the pace of execution. AI demand remained strong, but several non-AI markets weakened at the same time, including enterprise networking and carrier infrastructure, which exposed how dependent overall growth still was on end markets that were digesting excess inventory. Earlier quarters showed that AI revenue was offsetting softness elsewhere rather than creating a clean acceleration across the portfolio. Guidance disappointments during this period also hurt sentiment and led investors to reconsider how much premium AI optionality Marvell deserved. Broadcom’s steadier results across similar optical and ASIC categories made Marvell’s volatility more apparent and contributed to the derating.

Marvell now reports two primary segments. Data Centre has become the growth engine and includes optics, switching silicon, PCIe and CXL connectivity, storage controllers and custom accelerators. This segment is tied directly to hyperscaler spending on AI clusters and is the main driver of revenue and margin trajectory. Communications and Other encompasses enterprise networking, carrier infrastructure, consumer and industrial. Now, let’s take a look at the latest results.

Financials

Marvell delivered a solid Q3 FY26 with record revenue of $2.075bn, up 37% YoY and 3% QoQ, reflecting a clearer recovery across both AI and non-AI markets. Data Centre remained the core driver with revenue of $1.518bn, up 38% YoY, supported by stronger optical shipments, improving switching demand and sustained adoption of PCIe and CXL connectivity as hyperscalers expanded training capacity. PCIe and CXL are high-speed links inside the data centre that allow GPUs, CPUs and memory to communicate efficiently, helping hyperscalers move larger AI models through their systems without running into bandwidth bottlenecks. Custom silicon activity also remained healthy, with more than twenty active design programmes feeding into future cycles. The Communications and Other segment generated $557mn, up 34% YoY, helped by stabilising enterprise and carrier demand as customers rebuilt inventories after a prolonged digestion phase. Automotive and industrial revenue declined after the divestiture of the Automotive Ethernet business, but the broader mix improved sequentially.

Margins also moved in the right direction. Non-GAAP gross margin increased to 59.7%, up 30bps QoQ, while operating margin rose to 36.3%, up 150bps QoQ. This was encouraging given the heavier mix of optical products, which typically pressure margins. The improvement reflected firmer pricing, a richer contribution from advanced optics and higher value switching and custom compute. Non-GAAP EPS reached $0.76, up 13% QoQ and 77% YoY. Operating cash flow came in at a record $582mn, lifting cash on hand to $2.7bn, while inventories declined slightly to $1.015bn, signalling healthier supply chain balance and more disciplined internal planning. Non-GAAP net income reached $655mn, up 76% YoY, reflecting stronger margins and healthier operating leverage as both demand overall improved through the quarter.

Overall, the quarter showed clear progress in the areas that matter most. AI infrastructure demand continued to strengthen, non-AI end markets stabilised meaningfully and margins expanded despite mix headwinds. After a year shaped by volatility, Q3 FY26 represented a step toward a more consistent and investable profile, with a backdrop that now looks firmer heading into FY27.

Marvell also announced the acquisition of Celestial AI for $3.25bn, funded through $1bn in cash and roughly 27.2mn Marvell shares. Celestial AI develops on-chip and near-chip photonic interconnect that moves data using light rather than electrical signalling, which directly targets the bandwidth and power bottlenecks inside advanced AI systems. The deal includes an earn-out that could lift the total value to roughly $5.5bn if Celestial AI meets revenue milestones through FY29, a structure that ties most of the upside to actual commercial progress. Marvell expects the technology to begin contributing meaningfully toward the end of FY28, with a projected annualised revenue run rate of about $500mn, rising toward $1bn in FY29 as hyperscalers evaluate architectures that pull optical connectivity closer to the accelerator. The acquisition is large, but it strengthens Marvell’s roadmap in next-generation interconnect and aligns with the clear direction of travel as AI clusters push deeper into optical solutions to overcome scale and power limitations.

Price Action

Was up ~8% post earnings as the upbeat outlook (which I am yet to touch on) and steady performance satisfied investors. There’s another ~30% left before getting to all-time highs again, which I think is achievable within the next few months. I don’t currently own Marvell as I’ve always opted to hold Broadcom instead. However, as ASICs, in my opinion, will be the theme heading into next year, I’m considering adding to my exposure in this area, and Marvell is on my watchlist. It’s similar to AMD and Nvidia. There’s no way Marvell will dethrone Broadcom, but even if it can take a piece of its market share, the incremental revenue and margin uplift would be meaningful given the size of the opportunity.

Outlook

Management guided for Q4 revenue of roughly $2.2bn with non-GAAP gross margin of 58.5% to 59.5%, and the tone on the call was noticeably more confident than in previous quarters. They expect data centre revenue to grow more than 25% YoY in FY27, supported by firmer hyperscaler demand across optics, switching and PCIe or CXL connectivity. Visibility has also improved, with Marvell already holding purchase orders that cover the full FY27 forecast for its lead hyperscaler custom-silicon programme. Management reiterated that revenue should grow sequentially every quarter next fiscal year with a stronger exit rate into FY27, helped by the recovery in enterprise and carrier markets, which should lift Communications and Other by roughly 10% YoY after a year of digestion.

On the product side, management expects interconnect to outgrow cloud spending again next year as 800G and 1.6T optics ramp. Switching is also inflecting, with revenue set to exceed $500mn in FY27 as 51.2T adoption broadens and early work begins on 115Tbps UALink switches, which are planned to sample in the second half of CY26. Custom silicon should grow at least 20% YoY, with several new designs entering production in the back half of FY27 and a sharper acceleration expected in FY28 when higher-volume attached products ramp. Taken together, management sees a clear path toward more than 40% YoY growth in data centre revenue in FY28, supported by stronger switching, deeper custom-silicon penetration and the early contribution from photonics following the Celestial AI acquisition.

Final Thoughts

Marvell is not the finished article, but the direction of travel is clearly improving. The business has spent most of the year working through volatility in its non-AI markets while trying to prove that its AI exposure can deliver consistent growth rather than one-off surges. Q3 finally showed signs of that shift, with a stronger outlook, firmer visibility and a more balanced demand backdrop. Marvell will not challenge Broadcom at the top of the stack, but it doesn’t need to. Even modest share gains in ASICs, switching and advanced interconnect can translate into meaningful revenue and margin expansion given the scale of the markets it participates in. With the digestion phase now behind it and multiple growth drivers aligning into FY27 and FY28, Marvell looks far better positioned heading into next year than it did for most of this one, and that alone makes it a name worth watching carefully as the AI infrastructure cycle enters its next phase.

Thank you for taking the time to read this article. I hope you gained value from it. Until next time!

Disclaimer: This article is for informational purposes only and should not be considered financial or investment advice. The views expressed are my own and based on publicly available information, market trends, and personal analysis. Readers should conduct their own research and consult a financial professional before making any investment decisions.

Financials

Nvidia reported $57bn in revenue for the quarter, an increase of 62% YoY. Gross profit reached $41.8bn, with gross margin at 73.4% compared with 74.6% a year earlier. Cost of revenue rose to $15.2bn, up from $8.9bn in the prior-year quarter. Operating expenses were $5.8bn, up 36% YoY, driven by higher compute and infrastructure costs, increased compensation and benefits, and engineering development for new product introductions. Operating income came in at $36.0bn compared with $21.9bn a year earlier, resulting in an operating margin of 63.1%. Net income reached $31.9bn, rising from $19.3bn in the prior-year period

Data Center performance remained the key driver of the quarter. Compute revenue reached $43.0bn, increasing 56% YoY and 27% sequentially. Networking revenue was $8.2bn, up 162% YoY and 13% QoQ on continued growth in NVLink, XDR InfiniBand and Ethernet for AI systems. At the segment level, Compute & Networking generated $50.9bn, up 64% YoY, while Graphics delivered $6.1bn, up 51% YoY.

The remainder of the portfolio also showed solid YoY growth. Gaming revenue increased 30% YoY and declined 1% sequentially as channel inventories normalised. Professional Visualization grew 56% YoY and 26% QoQ, supported by the launch of new systems and Blackwell-related demand. Automotive revenue rose 32% YoY and 1% sequentially on continued adoption of Nvidia’s self-driving platforms. Inventory rose 32% QoQ, moving from $15.0bn in Q2 to $19.8bn in Q3. Raw materials more than doubled, and work-in-process increased from $4.4bn to $8.7bn as Nvidia ramped production of full Blackwell systems, which carry longer assembly and integration cycles. Finished goods fell from $8.7bn to $6.8bn, indicating that completed systems are being delivered while the new product cycle pushes a larger share of inventory into earlier build stages. Free cash flow remained strong. Nvidia generated approximately $11.3bn of operating cash flow in the quarter and spent around $0.4bn on capital expenditures, resulting in quarterly free cash flow of roughly $10.9bn. The margin profile reflected the early phase of the Blackwell cycle. Gross margin decreased YoY as the business shifted from Hopper HGX systems to full-scale Blackwell data center solutions, while sequential margin improved as Blackwell ramped and product mix shifted. Operating expense growth continued to track infrastructure investments, compensation increases and engineering costs associated with new product development.

Q3 Updates

The most significant update this quarter, in my opinion, was Nvidia’s plan to invest up to $10bn in Anthropic. While the detailed terms aren’t disclosed, the strategic intent is clear. Anthropic is one of the most compute-intensive model (Claude) developers in the market and is scaling training volumes rapidly. By taking a stake, Nvidia strengthens its position as Anthropic’s preferred hardware partner and gains earlier visibility into how its future models will be trained. That insight helps Nvidia fine-tune its roadmap for Rubin and later architectures, while also reinforcing long-term demand from a customer that is likely to consume large volumes of high-end accelerators over several product cycles.

Nvidia also signed a letter of intent to invest up to $100bn in OpenAI, tied to a strategic plan to deploy at least 10GW of Nvidia systems for OpenAI’s next generation of training infrastructure. The structure is a non-controlling stake, but the scale of the commitment gives Nvidia direct visibility into how OpenAI plans to train and scale its frontier models over the coming years. With the first 1GW of Rubin-class systems expected to be deployed from 2026, this agreement effectively aligns Nvidia’s future architectures with one of the largest and most compute-intensive model developers in the market, securing multi-year demand for Blackwell, Rubin and the platforms that follow. This is slightly less exciting news as I don’t see how OpenAI will be able to deploy that much in the medium-term future but I won’t get into that.

Nvidia also announced a $1bn investment in Nokia to accelerate AI-RAN development and support the industry’s transition from 5G to 6G. The partnership will see Nokia build new AI-RAN products on Nvidia’s platform, with Nvidia supplying its Arc Aerial RAN Computer as the core of Nokia’s next-generation wireless systems. The collaboration brings Nvidia directly into the radio access network stack at a time when mobile AI traffic is set to grow sharply, with operators like T-Mobile already testing these systems for their future 6G deployments. For Nvidia, the deal extends its reach beyond data centers into the telecom infrastructure that will ultimately carry AI workloads, giving it a foothold in a part of the ecosystem that is becoming increasingly compute-intensive, which is a smart move.

Price Action

Shares were up roughly 6% at their highs post earnings but erased those gains throughout the trading session. It’s inevitable that the growth will slow at one point. I just don’t think it’ll be in any of the next few quarters. Revenues are only getting stronger, margins continue to remain high, the cash is flowing, and the moat is only growing deeper than it already is. For these reasons, I’m very much looking to add to my position around the $170 mark. Sentiment has been very risk off however, so if equities continue to head lower, I will look to add at a better price. Trading at a 44.74 P/E multiple is pretty standard for Nvidia as they’ve grown into their valuation so I don’t have any concerns on the valuation front.

Outlook

Looking ahead, Nvidia still appears to be in the middle of a multi-year AI infrastructure cycle rather than anywhere near the end of it. The company guided Q4 FY26 (which ends in Jan) revenue to be $65bn, which, even on a base this large, suggests demand remains broad and well supported across hyperscalers, model developers and enterprise customers. The Blackwell ramp is clearly tracking ahead of expectations, and the mix of full-rack systems, networking and software continues to deepen Nvidia’s footprint inside customer deployments. The biggest unknowns sit further out rather than in the next few quarters. AI infrastructure spending is still being driven by an unusually small group of buyers, and there will inevitably be a point where this pace settles into something more normal. But based on the numbers this quarter, that point doesn’t feel close. The scale of cloud commitments, the acceleration in networking, and the strategic links Nvidia is building with Anthropic and OpenAI all suggest the demand environment remains robust and likely to hold through the early phases of Rubin. In the near term, the story remains unchanged. Nvidia continues to benefit from the strongest tailwind in technology, and there is still plenty of runway left in this cycle. The long-term debate will eventually shift to how sustainable this rate of investment is, but nothing from this quarter points to a slowdown. If anything, visibility has improved.

Thank you for taking the time to read this article. I hope you gained value from it.

Disclaimer: This article is for informational purposes only and should not be considered financial or investment advice. The views expressed are my own and based on publicly available information, market trends, and personal analysis. I own NVDA shares, which may change at any time. Readers should conduct their own research and consult a financial professional before making any investment decisions.

• CoreWeave & AMAT Q2: Leveraged Gamble vs. Cash Generator

• AMD, Arista & D-Wave Q2: AI Momentum, Networking Growth, Quantum Progress

Financials

AMD delivered a record quarter with revenue of $9.2bn, up 36% YoY, driven by broad strength across Data Centre AI, EPYC servers and a sharp rebound in Client and Gaming. Data Centre revenue reached $4.3bn, up 22% YoY, as MI300 continued to ramp and EPYC volumes increased across cloud and enterprise. Client and Gaming posted $4.0bn, up 73% YoY, supported by record Ryzen demand, stronger semi-custom sales and improving Radeon momentum. Embedded declined to $857m, down 8% YoY, although sequential growth returned. Operating performance across the segments was solid. Data Centre generated $1.07bn of operating income, giving the business a 24.7% operating margin, down from 29.3% a year ago as higher AI-focused R&D weighed on profitability. Client and Gaming delivered $867m of operating income, translating to a 21.4% margin, up sharply from 12.3% last year as Ryzen mix improved and semi-custom volumes strengthened. Embedded reported $283m of operating income, a 33.0% margin compared with 40.1% a year ago, which remains the highest across the portfolio despite lower revenue.

Whilst most of the hype is around the Data Centre segment, it’s important to note that Client and Gaming is also doing very well and outperformed for clear reasons. The PC business delivered its strongest quarter in years, with record Ryzen sales across desktops and notebooks, as Ryzen 9000 drove all-time-high desktop sell-in and strong demand from gamers, creators, and commercial users. Notebook sell-through accelerated, and enterprise adoption rose more than 30% YoY as Fortune 500 customers standardised more of their fleets on AMD systems, improving mix and margins. Gaming also rebounded sharply, with higher semi-custom revenue ahead of the holiday cycle and stronger Radeon performance helped by healthier channel conditions, uptake of the Radeon 9000 series and accelerating FSR4 adoption. Together, these trends created the best backdrop for the segment in years, supported by a richer Ryzen mix, rising notebook and desktop sell-through, enterprise upgrades and a clear recovery in Radeon.

The balance sheet remains solid with $7.2bn in cash and short-term investments and a record $1.5bn in free cash flow. Inventory increased slightly ahead of a heavier accelerator and CPU product cycle, while total debt stayed low at $3.2bn.

Two points worth highlighting are AMD’s expanding relationship with Oracle and its multi-year agreement with OpenAI, which sent shares flying +20% on announcement in October. OpenAI has committed to buying 6GW of MI450 compute, a scale that implies tens of billions of dollars in potential revenue for AMD over time. As part of the deal, AMD granted OpenAI warrants, which simply give OpenAI the right to buy up to 160mn AMD shares at $0.01 if it deploys AMD’s chips at scale and AMD hits certain share-price targets. In practice, this rewards OpenAI for becoming one of AMD’s biggest customers and tightly aligns both companies around MI450’s success. Oracle adds a second major source of visibility, with OCI set to deploy tens of thousands of MI450 GPUs through 2026 and 2027 and build on the large MI350 clusters already in production. The combination of OpenAI’s 6GW commitment and Oracle’s multi-year GPU rollouts gives AMD a clearer forward revenue pipeline than at any point in its history, which is enticing as an investor. But of course, buying when things are already going great can limit the risk-to-reward of a trade.

Price Action

AMD has done extremely well this year and rightly so. I don’t hold AMD shares in my portfolio, and I wouldn’t be looking to start a position at these levels either, but in the long run, I do see AMD as a strong holding to have in a tech portfolio. I’ll be looking to pick up AMD shares on pullbacks because the catalysts ahead are genuinely strong. Of course, this all depends on the overall strength of the AI build-out continuing. If projects start getting cancelled or delayed, I wouldn’t be keen to pick AMD up even at a steep discount. AMD currently trades at around 122x earnings, which is undeniably rich, but AMD has traded at elevated multiples ever since the start of the AI boom, so it’s not something that, in my opinion, needs to be overthought.

Outlook

At the Investor Day last week, AMD laid out one of the most ambitious multi-year roadmaps we’ve seen from them. Management expects the Data Centre business to grow at more than 60% CAGR over the next three to five years, with MI350 and MI450 at the core of that acceleration. The company reiterated that its Data Centre AI revenue is on track to reach the “tens of billions” range in 2027, supported by the 6GW commitment from OpenAI, Oracle’s plan to deploy tens of thousands of MI450 GPUs, and broader hyperscaler and sovereign AI demand. The introduction of the Helios rack-scale system also signals a shift toward full-stack deployments, allowing AMD to compete directly with Nvidia at the cluster level rather than only at the chip level. Combine that with a rapidly maturing ROCm software stack and deeper customer co-design wins, and AMD enters 2026 with more visibility across its AI roadmap than at any point in its history.

Turning to guidance, AMD expects Q4 revenue of roughly $9.6bn, plus or minus $300m, with non-GAAP gross margins around 54.5%. Management noted continued growth in Data Centre and solid momentum in Client and Gaming heading into year-end, while Embedded remains soft but stabilising. The key swing factor into 2026 will be the MI350 and MI450 ramps, where AMD already has multi-quarter visibility from hyperscalers and AI-native customers. If execution remains strong and the broader AI build-out continues at the current pace, AMD’s long-term revenue trajectory remains firmly intact. It’ll be interesting to hear AMD’s 2026 guidance next quarter.

CoreWeave

If you’ve been a long-time reader of The Tech Investor, you’ll know I have plenty of reservations about CoreWeave, and it’s not a stock that’ll ever be in my portfolio. That said, they do solve a very real problem in today’s environment, which is the sheer need for compute, and they’re becoming increasingly important in that ecosystem. Even so, investors weren’t pleased with the latest results, and the reaction reflected that.

Financials

Revenue came in at $1.4bn, up 134% YoY and comfortably ahead of expectations, supported by persistent AI infrastructure demand and improved fleet efficiency. Adjusted operating income came in at $217mn, above the $176mn consensus, translating to a 16% operating margin, helped by higher revenue and lower near-term operating costs as several third-party data centre deliveries slipped into next year. Capex was also significantly lower at $1.9bn versus the $4.0bn estimate, reflecting delays from CoreWeave’s external data centre development partners. One of its major developers failed to deliver a powered shell on time, pushing back the onboarding of a large contract, while several other sites ran into supply-chain bottlenecks that shifted infrastructure deployment into early 2026. Adjusted net income remained negative at –$41mn, reflecting the ongoing cost of scaling infrastructure and headcount alongside these delays.

Total RPO rose by $25.5bn in the quarter to $55.6bn, marking 271% YoY growth, which is impressive and highlights the demand out there. This included a $6.5bn OpenAI upsize, a $14.2bn Meta deal and roughly $4.8bn of additional hyperscaler contracts, including the sixth expansion of an agreement with a major cloud customer. Customer concentration, which was a major cause of concern for me, continues to improve, with the largest customer now representing ~35% of backlog, down from ~50% last quarter and ~85% at the start of the year. Notably, CoreWeave is still signing contracts for both new Blackwell GPUs and older H100s, with recontracted pricing only within 5% of original agreements, underscoring the long economic life of its GPU fleet.

CoreWeave ended the quarter with $3bn in cash, providing enough flexibility heading into a heavy build-out phase. Debt increased sequentially as the company continued to fund expansion, which is expected, given the scale of infrastructure required to meet demand. Using CoreWeave’s adjusted operating income of $217mn and subtracting the $1.9bn of capex spent in Q3, you arrive at a cash burn of just over $1.6bn, highlighting the imbalance between near-term profitability and the level of investment required to bring new capacity online. Capex will remain lumpy, but long-term contracts still give CoreWeave multi-year revenue visibility as these deployments come through, for as long as they do.

Price Action

$CRWV Shares were down ~9% post-earnings as the company cut guidance meaningfully, with delays in data centre build-outs pushing revenue and operating income into next year and forcing investors to re-price the near-term growth trajectory. I personally have no interest in adding CoreWeave to my portfolio even as it sells off. There’s not much history to the chart, therefore I have no levels drawn, although I expect a drop to the IPO price of $40 is much more likely than climbing back up to October highs around $150.

Outlook

Looking ahead, CoreWeave’s near-term trajectory is shaped almost entirely by the timing of its infrastructure deployments. After cutting full-year guidance by $150mn on revenue and $110mn on operating income, management made clear that the bulk of the downgrade reflects the pushout of several large contracts into early 2026 rather than any deterioration in demand. Full-year capex was lowered by $8.5bn and re-timed into 1Q26, which means next year’s capex will be more than double 2025 levels. Management also highlighted that the revised schedule will create a temporary margin headwind in Q4 and early 2026, as data centre costs are incurred before associated revenue begins to flow through. On the call, CoreWeave noted that several sites are being re-sequenced to avoid similar delays, and the company is accelerating its move to own more of its data centres to reduce reliance on external developers. Multiple large clusters are scheduled to go live in the first half of 2026, which should lift revenue as those deployments ramp.

Another point worth flagging from the quarter is CoreWeave’s failed attempt to acquire Core Scientific, which ultimately fell apart after Core Scientific rejected the offer and chose to continue expanding independently. While the deal would have given CoreWeave immediate access to additional power capacity and faster deployment timelines, the collapse reinforces a broader theme we’ve seen this quarter: CoreWeave is struggling to secure infrastructure fast enough to match demand, whether through partners or through acquisition.

Final Thoughts

AMD continues to execute well, and the business clearly has momentum across Data Centre, Client and Gaming. I like the trajectory, and while I wouldn’t buy it at today’s valuation, it’s a name I’d look to pick up on a proper pullback. CoreWeave is also performing well on the top line and signing huge deals, but the model is far more capital-intensive and execution-heavy than I’m comfortable with. It’s becoming important in the compute ecosystem, but it’s still not a stock I’d want to hold. For now, it’s one I’m happy to watch from the sidelines.

Thank you for reading this article. I hope you found it insightful! For more articles like this, subscribe! And if you think any friends/family/colleagues would find this article interesting, feel free to send it across.

Disclaimer: This article is for informational purposes only and should not be considered financial or investment advice. The views expressed are my own and based on publicly available information, market trends, and personal analysis. Readers should conduct their own research and consult a financial professional before making any investment decisions.

Having said that, there is some truth in what is being said, but hey, bubbles can go on for years. Imagine listening to the media pre-tariff chatter, you’d have missed out on a lot of juice. One name which didn’t close in the red was Datadog, which is the topic of today’s article.

Datadog

Datadog was founded in 2010 by Olivier Pomel and Alexis Lê-Quôc, both former engineers at Wireless Generation, with the goal of helping companies understand what is happening inside their technology systems in real time. When an application slows down, a server crashes, or a website goes offline, Datadog’s platform identifies the cause, whether it is a software bug, a cloud outage or a security issue. What started as a simple tool for tracking cloud performance has grown into a broad observability and security platform used by thousands of enterprises worldwide. Its technology brings together everything a company needs to monitor, from infrastructure and databases to AI models and user activity, into one live dashboard. In essence, Datadog acts as the central nervous system of digital operations, giving engineers and security teams the visibility they need to keep systems running smoothly, efficiently and securely. Its core products include infrastructure monitoring, application performance monitoring and log management, which together let companies track the health and performance of their digital systems in real time. The platform has expanded into security, user analytics and AI observability, giving customers a single place to manage performance and protection across their entire cloud environment.

Datadog operates a usage-based subscription model. Customers pay according to how much data they monitor and how many products they use, which means revenue rises naturally as their digital footprint expands. Most start with one product and quickly adopt more, driving a high level of recurring revenue.

Datadog went public in September 2019 on the NASDAQ under the ticker DDOG, pricing its IPO at $27 per share and valuing the company at roughly $10bn. The IPO was one of the most successful software listings of that year. Shares surged more than 30% on the first day of trading, reflecting strong investor confidence in the company’s early leadership in cloud observability.

At the time, Datadog was primarily known for its infrastructure and application monitoring tools, used by cloud-native companies to track performance across AWS, Azure and Google Cloud. Since going public, the business has expanded rapidly both horizontally and vertically. It added log management in 2020, followed by security monitoring, real user monitoring (RUM), synthetics, and, most recently, AI observability and application analytics. The strategy has been to evolve from a single-use monitoring product into a unified cloud operations platform, allowing customers to consolidate multiple vendors under Datadog’s suite.

It hasn’t been all smooth sailing since the IPO, however. Between 2021 and 2023, Datadog’s stock performance lagged sharply despite strong operational results. The company entered the public markets during the height of the software boom, trading at over 60x forward sales, but as interest rates began to rise and investors rotated out of high-multiple growth names, that valuation collapsed. At the same time, many of Datadog’s large enterprise customers started cutting cloud spending to protect margins after the pandemic surge in demand. Because Datadog’s pricing is usage-based, revenue growth slowed even though customer retention stayed high.

Now that we’ve covered the history and what Datadog is all about, let’s dive into why shares rallied this week.

Financials

Datadog delivered one of its cleanest quarters in recent memory, with strong growth across all metrics and evidence of a clear reacceleration in usage. Revenue rose 28% YoY to $886mn, beating expectations by roughly $30mn and marking a return to high-20s growth after several quarters of more muted expansion. Management highlighted that this performance was driven by both rising consumption from existing customers and continued traction with new AI-native clients, suggesting the cloud optimisation cycle that weighed on results through 2023 is now largely behind them. Importantly, growth was balanced across all major product areas. Infrastructure monitoring grew in the low-20% range, showing renewed strength in core DevOps and cloud observability workloads. APM and log management rose in the mid-30s as enterprises consolidated their tools onto Datadog’s platform, while security was the clear standout, as management confirmed it grew in the mid-50% YoY range, accelerating from the mid-40s last quarter and now accounting for roughly 10% of total revenue. This mix shift toward higher-value analytics and security products continues to underpin margin expansion.

Profitability reached record highs. Gross margin was 80%, supported by improved infrastructure efficiency and the growing contribution from high-margin software services. Operating income rose 35% YoY to $203mn, driving an operating margin of 22.9%, the highest in company history, while net income was $175mn, translating to EPS of $0.55 versus expectations of $0.46. Free cash flow was $214mn, a 24% FCF margin, keeping Datadog among the most cash-efficient names in SaaS. Cost discipline remained evident, with operating expenses growing roughly half as fast as revenue, reflecting strong leverage even as R&D investment continues at around 20% of sales.

Customer metrics remain equally impressive. Datadog ended the quarter with 4,060 customers generating over $100k in ARR, up 16% YoY, and 15 AI-native customers now spending over $1mn annually. Platform depth continues to increase, with 54% of customers using four or more products (up from 49% a year ago) and 31% using six or more (up from 26%), underscoring growing cross-sell momentum. CEO Olivier Pomel emphasised that usage growth from non-AI-native clients reached its highest level in over three years, while AI-native customers now exceed 500 in total, expanding rapidly as observability and security become essential to scaling their real-time workloads.

Security monitoring grew roughly 55% YoY, driven by demand for unified threat detection, workload protection and compliance tools within the Datadog platform. Management also highlighted early adoption of Bits AI, the company’s new AI assistant designed to automate issue detection and remediation, which has already gained strong traction in preview.

Price Action

$DDOG shares rallied roughly 20% post-earnings as investors digested the beat and confident guidance. Datadog isn’t a position in my portfolio, but it’s one I follow closely. The execution has been exceptional, though valuation remains full, the GAAP P/E north of 600x exaggerates the picture, but even on adjusted numbers, it’s hardly cheap. For now, it feels fairly valued rather than wildly overextended, so I’m happy to stay on the sidelines and watch how margins and AI-driven growth evolve into next year. Although if the selloff continues, I may be inclined to start a position, as I do find software to be an interesting sub-sector. In terms of where it trades next, it’s tough to call. If equities keep selling off, I doubt $DDOG breaks above $200, but with more positive sentiment, it could easily push past that level.

Outlook

Management guided Q4 revenue to $912–$916mn, well ahead of Wall Street’s estimate of $887mn, and lifted full-year revenue expectations to $3.39bn, alongside an improved EPS outlook of $2.00–$2.02. The tone of the call was confident. Pomel reiterated that “cloud migration and digital transformation remain strong” and that Datadog is increasingly critical to how enterprises monitor AI and cloud infrastructure in real time. Growth is broadening again as traditional enterprise workloads rebound while AI-native clients continue to scale, driving heavier and more sustained platform usage. As more AI applications move from development into production, the need for constant, real-time observability will only grow, an area where Datadog already holds a clear competitive lead. With over 500 AI-native customers, accelerating multi-product adoption and expanding margins, Datadog is set to close out 2025 with clear momentum and one of the strongest operating profiles in enterprise software.

Final Thoughts

Datadog’s results speak for themselves. The business has found a second wind as cloud spending rebounds and AI adoption deepens, showing that it’s more than just another software name riding the trend. Execution is solid, margins are expanding, and the platform is becoming indispensable for modern infrastructure. That said, the stock has already priced in a lot of this strength. It’s a company I’d want to own, just not at this price.

Thanks for taking the time to read. I hope you found this breakdown useful and maybe learned something new along the way. If you enjoyed it, consider sharing it with someone who might too, as it really helps the page grow.

Disclaimer: This article is for informational purposes only and should not be considered financial or investment advice. The views expressed are my own and based on publicly available information, market trends, and personal analysis. Readers should conduct their own research and consult a financial professional before making any investment decisions.

On a more positive note, all the names that I’ll be covering today beat expectations (excluding Meta’s tax charge), of course, but the reception received from investors was very different. Let’s begin with the least well-received.

Meta

Post Q2 earnings, shares were up ~10%. This time around, it’s the opposite with Meta selling off ~10% post Q3 earnings. The results themselves weren’t bad, but ultimately, given the hype and the scale of investment, ‘weren’t bad’ will no longer cut it for investors.

Financials

Meta reported revenue of $51.2bn, up 26% YoY, driven almost entirely by its ad business, which continues to benefit from higher engagement across Reels and improving ad pricing. Ad impressions rose 14% while the average price per ad increased 10%, marking another quarter of solid momentum despite regulatory pressures in Europe. Reality Labs contributed $470mn in sales, up 74% YoY, reflecting stronger demand for Quest headsets and the new AI-powered Ray-Ban glasses.

Operating income came in at $20.5bn with a 40% margin, slightly below last year as infrastructure and legal costs rose alongside a heavier investment cycle in AI data centres. The bottom line, however, was distorted by a one-off $15.9bn non-cash tax charge tied to the implementation of the new U.S. Corporate Alternative Minimum Tax. The change forced Meta to revalue its deferred tax assets, which are accounting credits representing future tax savings from items like stock-based compensation and past overpayments. Under the new tax law, large companies must pay at least 15% tax on their reported book income, meaning many of those old credits can no longer be fully used. Meta therefore reduced their value, booking an accounting loss that doesn’t affect cash flow. In fact, the company expects to pay less U.S. federal cash tax going forward, despite the headline charge. Excluding this adjustment, Meta would have reported roughly $18.6bn in net income, or $7.25 per share, making it one of its most profitable quarters ever.

The Family of Apps segment continues to show healthy operating performance, with advertising revenue up 26% YoY to $50.1bn, supported by double-digit growth in both impressions and pricing. Engagement across Reels remains high, and monetisation continues to improve as the format matures, contributing meaningfully to overall revenue. The introduction of new AI models has further enhanced ad targeting and recommendation accuracy, helping advertisers achieve better conversion rates and higher ROI. These trends show that Meta’s core business is holding up well even as spending rises sharply across AI and infrastructure.

Reality Labs, while still small in comparison, continues to represent Meta’s long-term bet on spatial computing. Losses narrowed slightly to $4.4bn but remain significant as Meta invests in AR, VR and mixed-reality hardware. The success of the new AI glasses signals genuine consumer interest beyond gaming, which is encouraging as the company looks to integrate its AI assistant more deeply across devices. The business is performing well, but back to my earlier point, at what cost?

CapEx reached $19.4bn in the quarter as Meta continues to scale its AI infrastructure, bringing year-to-date spending above $50bn. Meta expects investment to stay high through 2026 as it builds new data centres to support generative AI and recommendation systems. Most of this compute goes toward ad ranking, recommendation models and the training of Meta’s in-house AI systems like LLaMA. However, they are also building capacity for future use across products such as its AI assistant and AR glasses. Management said this will eventually make operations more efficient as Meta moves to its own AI training and inference platform that powers its apps.

What stood out to me, though, was Zuckerberg’s comment that they’re front-loading this buildout in case superintelligence comes sooner than expected. If it doesn’t, they’ll just slow construction and reallocate the compute to their core business. That’s the part I struggle with. Meta isn’t a cloud provider, it’s an advertising business, so I don’t really see where all that compute goes if AI progress takes longer. It feels like they’re building massive capacity mostly out of fear of falling behind rather than from a clear idea of how it directly drives revenue, which is a concern shared amongst investors.

Meta has also started raising debt to fund part of this CapEx cycle, which I don’t necessarily see as a bad thing. It has raised about $27bn via a private-credit joint venture with Blue Owl Capital to build its Hyperion data-centre campus, and this week it announced a bond offering of up to $30bn across six tranches to support its AI infrastructure push. The costs here are largely fixed, interest and principal repayments, while the upside, if this AI infrastructure actually pays off, belongs entirely to equity holders. In other words, the downside is limited to debt servicing, but the upside is uncapped if Meta’s AI investments meaningfully improve engagement, ad performance, or unlock new revenue streams. That’s attractive from an equity perspective. But on the other hand, the problem is what happens if the payoff doesn’t come. If AI progress slows or fails to translate into real monetisation, Meta would still take a financial hit as the company is pouring billions into infrastructure that won’t generate immediate returns. But it isn’t an existential risk. Meta remains highly cash generative and isn’t funding the entire buildout from its own balance sheet, so it can absorb some inefficiency. The real damage would come from the loss of narrative. Without AI as a credible growth driver, Meta would start to look less like a company reinventing itself and more like one that’s maturing without a clear next leg of growth.

Price Action

I feel like is in a bit of no man’s land right now, and it’s a bit too early to say where it’ll head in the near term. The business itself is still performing well, but the spending narrative is starting to damage that. I think right now the story around Meta is quite fragile, and any hint of negativity or bad news both in the AI space and the macro picture, would perhaps result in another outsized sell-off. So right now I’m sitting on my hands, not adding and not taking off, but it’s one I will keep a very close eye on as the possibility of taking some chips off the table is becoming more realistic.

Outlook

Management guided Q4 revenue to $46–49bn, reflecting continued ad demand but a slower growth rate versus the first half. Full-year CapEx was reaffirmed at $53–57bn, with similar levels expected in 2026 before easing. Meta said the current spending phase is focused on expanding its AI infrastructure to power products like its AI assistant, AI Studio and AI agents, which will be integrated more broadly across WhatsApp, Instagram and Facebook next year. Zuckerberg and CFO Susan Li both emphasised that the company is still early in monetising these products, but they view them as central to Meta’s long-term AI roadmap and a key part of how the company will evolve beyond pure advertising.

I think management’s message was clear. They see this as a foundational investment cycle rather than a near-term growth lever. The issue is that these AI products are still far from making any large contributions to revenue, relative to the spending. Until there’s a clearer link between AI adoption and financial returns, the market will continue to treat this as a spending story rather than a growth one.

Alphabet

Now, onto a more positive story, Google blew it out of the park this quarter.

Financials

Google reported revenue of $102.3bn, up 16% YoY, marking its first-ever $100bn quarter and driven by broad-based strength across Search, YouTube and Cloud.

Search and Other revenue grew 15% YoY to $56.6bn, led by Retail and Financial Services. AI continues to play a larger role here, with AI Overviews and AI Mode driving higher query and click growth. Management said commercial queries are expanding faster than total queries, meaning the mix is becoming more monetisable. AI Max, Google’s new AI-powered ad format, has already been adopted by hundreds of thousands of advertisers since launching in September, unlocking billions of incremental ad impressions. Encouragingly, monetisation rates within AI Overviews remain roughly in line with traditional search results, easing concerns that AI integration could dilute ad yield. Gemini usage continues to rise, with daily active users up over 30% sequentially as Google expands access across Gmail, Docs, Android and Chrome.

YouTube advertising revenue rose 15% YoY to $10.3bn, supported by Shorts and connected TV. Shorts now earns more revenue per watch hour than traditional in-stream ads, while connected TV continues to grow quickly, with interactive ads reaching a $1bn annual run rate. Subscription revenue from YouTube Premium, Music and YouTube TV increased 21% YoY, highlighting continued progress in diversifying beyond ads.

Cloud was the standout once again, up 34% YoY to $15.2bn, marking its fastest growth since 2022. Operating margins improved from 17% to nearly 24% as demand for AI infrastructure lifted profitability. Enterprise adoption of Gemini and Vertex AI continues to accelerate, with revenue from generative AI products tripling YoY. Google also announced the rollout of its sixth-generation TPU chips, designed to train larger multimodal models at lower cost, improving both compute efficiency and energy use across data centres. TPUs, or Tensor Processing Units, are Google’s in-house AI chips built specifically for machine learning workloads, giving the company tighter control over performance and cost compared to third-party GPUs. Cloud backlog climbed 46% sequentially to $155bn, with Google signing more billion-dollar deals in nine months than in the previous two years combined.

Operating income rose to $31.2bn, while margins came in at 30.5%, or closer to 34% excluding the $3.5bn European Commission fine. Net income was $35bn, up 33% YoY, and EPS came in at $2.87. Strong operating leverage across Cloud and Services offset higher infrastructure costs, and gross margins expanded modestly despite heavy investment. Chrome also saw meaningful AI upgrades this quarter, including on-device Gemini integration and faster page rendering through new optimisation tools, showing how Google continues to embed AI deeper across its ecosystem.

Free cash flow reached $24.5bn in the quarter, bringing trailing twelve-month FCF to $73.6bn, up 32% YoY. CapEx rose sharply to $24bn, up 83% YoY, as Google continued building out AI infrastructure. CFO Anat Ashkenazi raised full-year CapEx guidance to $91–93bn and said spending will rise again in 2026, with roughly 60% directed toward servers and 40% toward data centres and networking. While the spending is heavy, Google’s balance sheet gives it ample flexibility, and unlike Meta, these investments are already feeding into products with clear monetisation potential.

Price Action

Shares in were up ~2.5% post earnings, nothing extremely exciting, but better than being down 10%. Shares have really rallied since September, and I did add to my existing position in and around then. Because of that, I’m currently not looking to do anything here. I don’t buy at all-time highs, and I don’t think there’s much higher to go at this stage without having a meaningful-ish pullback. However, Google is, in my opinion, really well positioned as it has reach on both the advertising and cloud servicing side, which diversifies its revenue streams.

Valuation-wise, Google now trades at around 28x earnings and 23x cash flow, both up meaningfully from earlier in the year when multiples were in the low-20s. The rally reflects stronger execution and renewed confidence in AI monetisation, but it also means expectations are high heading into 2026. On a relative basis, the stock still screens reasonably against peers given its growth profile and cash generation, though it’s no longer cheap.

Outlook

Management expects continued growth into Q4 and 2026, with steady momentum across Search, YouTube and Cloud. Sundar Pichai said Google is entering a “multi-year AI expansion phase” and that Gemini’s integration across Android, Chrome and Workspace will be a major focus over the next year.

On the infrastructure side, Google plans to ramp TPU capacity further with the launch of its seventh-generation chip, Ironwood, which will soon be generally available. Management noted that demand from enterprise customers remains exceptionally strong, with capacity still lagging behind demand in some regions. CFO Anat Ashkenazi reaffirmed full-year CapEx guidance of $91–93bn and signalled another increase in 2026 as Google expands data centre and AI infrastructure to support Gemini and Vertex AI growth. Despite the heavy investment, management expects operating margins to remain broadly stable, given strong profitability in Cloud.

I was very happy with the results for this quarter and continue to be a happy shareholder due to the returns this position has given me. As mentioned, I think Google is very well placed and will continue to do well.

For the next section, I will only be focusing on the cloud businesses that overshadow Google’s cloud operations.

Cloud Titans

Microsoft

Microsoft’s Intelligent Cloud segment, which includes Azure, server products like Windows Server and SQL Server, and enterprise services, continues to dominate, with revenue up 26% YoY to $29.9bn and Azure growth accelerating to 39%. Demand for AI compute remains exceptional, and management again highlighted that capacity is still lagging behind demand. Amy Hood said that “utilisation is well ahead of plan,” with much of the new build already booked by existing customers, implying that growth is being limited by supply rather than demand.

A major development this quarter was the restructuring of OpenAI, which gives Microsoft a 27% stake in the newly formed for-profit entity, valued at roughly $135bn. The partnership further tightens the relationship between the two companies and secures a long-term pipeline of AI workloads for Azure. Management also mentioned that the compute infrastructure supporting OpenAI and other large customers can be repurposed across Microsoft’s wider business, ensuring that capacity built for AI isn’t tied to a single client but instead enhances scalability across the platform.

RPOs climbed sharply to $368bn, up 37% YoY, reflecting strong, long-term contractual commitments from enterprise customers. The company said AI-related bookings now make up a growing share of those commitments, as more clients integrate Azure OpenAI and Copilot into their systems. Management reiterated that demand continues to exceed supply and that Azure revenue could be even higher if not for ongoing capacity constraints.

I thought this was another strong quarter for Microsoft’s cloud business, but the bar is getting higher. Demand is impressive and the pipeline looks solid, yet the scale of investment needed to sustain that growth is significant. Azure remains well-positioned, but as the AI race intensifies, execution and cost discipline will matter more than just capacity expansion. Looking at the chart, has performed extremely well, and I don’t expect any big swings any time soon. I don’t hold a position in Microsoft anymore, but I would be open to picking up a few shares on any weakness.

Amazon

AWS reported revenue of $33bn, up 20% YoY, its fastest growth in over two years and well ahead of expectations. Operating income came in at $10.8bn, giving the business an operating margin of 33%, down slightly from last quarter as spending on data centre expansion continued. Growth was broad-based across both enterprise and AI workloads, with CEO Andy Jassy noting that AWS is now seeing “the strongest pipeline of cloud migrations and AI projects” in the company’s history.

Demand for AI infrastructure remains a clear theme. Management said AWS added over 3.8 GW of compute capacity over the past year, largely to support training clusters and inference workloads for customers building on Bedrock and Trainium. AWS also launched new Graviton 5 and Trainium 2 chips this quarter, designed to lower the cost per training cycle and improve efficiency, a direct response to Microsoft and Google’s push into custom silicon. Jassy made it clear that AWS will “keep building” to meet AI demand and sees no signs of slowing investment.

That investment, however, comes at a cost. CapEx for 2025 is now expected to reach $125bn, up sharply from last year, and management hinted at another step-up in 2026 as global data centre buildouts continue. CFO Brian Olsavsky acknowledged that while these investments will pressure free cash flow in the near term, utilisation across AWS infrastructure remains high, giving them confidence in future returns.

What stood out to me is that AWS feels like it’s back on the front foot. Growth has reaccelerated, the AI product set is expanding, and customer engagement looks stronger than it has in several quarters. But unlike Microsoft, the margin story here is less clear. AWS still has a huge lead in scale, but it’s spending heavily to defend it, and the payoff from that spend isn’t as visible yet.

Shares were up roughly 8% post earnings as investors welcomed the sharp reacceleration in AWS growth and better-than-expected profitability across the group. Amazon has never been in my portfolio, and at this stage, I’m not interested in adding it, but that’s purely down to personal preference.

Final Thoughts

This set of results left me more conflicted than usual. The numbers were strong across the board, but the spending is starting to feel excessive, and I can’t shake the sense that risk is quietly building beneath the surface. The fact that so many of these companies are tapping debt markets (which isn’t necessarily a bad thing) to fund AI infrastructure shows just how far and fast this cycle is moving. I’m not bearish, far from it, but some of that blind optimism I had earlier in the year is fading. My gut tells me we’re reaching a point where execution and balance sheet strength will start to matter a lot more than vision.

Thank you for reading this article. I hope you found it insightful and gained value from it.

Disclaimer: This article is for informational purposes only and should not be considered financial or investment advice. The views expressed are my own and based on publicly available information, market trends, and personal analysis. I own both META and GOOG shares, which may change at any time. Readers should conduct their own research and consult a financial professional before making any investment decisions.

ISRG

I’ve been discussing ISRG since I first started the Tech Investor so if you aren’t already familiar, I’d recommend having a quick look at my deep dive and Q2 earnings articles. Having started off the year so well, ISRG’s stock has been selling off heavily, and I’d like to think that it’s mainly due to the way management has been talking about tariffs having an impact amidst growing competition in China. But regardless, I’ve always felt that Intuitive’s moat is so wide that tariffs wouldn’t have an outsized impact, which so far seems to be correct. Shares rocketed post Q3 earnings, so let’s take a look at why.

As a quick reminder, the Ion systems are used for robotic-assisted lung biopsies, enabling physicians to navigate deep into the lung to sample small, hard-to-reach nodules using AI-guided imaging, while the da Vinci systems are Intuitive’s core surgical robots designed for minimally invasive procedures across general surgery, urology, gynaecology and thoracic care.

Financials

Revenue rose 23% YoY to $2.51bn, comfortably ahead of expectations, driven by robust procedure growth and broad adoption of Intuitive’s newer platforms. Global procedure volumes increased roughly 20%, with da Vinci up 19% and Ion up 52%, showing that growth remains well balanced between mature and emerging systems. Instruments and accessories revenue grew 20% YoY to $1.52bn, tracking procedure growth closely and confirming that utilisation across the installed base remains strong. Growth was broad-based across benign general surgery and gynaecology in the US, while internationally it was led by India, Korea and Brazil. Ion also contributed meaningfully, with procedure mix shifting toward higher-yield cases that supported stable average revenue per procedure of around $1,800 for da Vinci and $2,200 for Ion.

System revenue climbed 33% YoY to $590mn as placements rose to 427 from 379 last year. The da Vinci 5 accounted for over half of total placements, reflecting accelerating upgrades and continued hospital standardisation around the new platform. Within that, 204 units were da Vinci 5 systems, including 12 placed internationally following new clearances in Japan and Europe. The shift to usage-based operating leases continued, representing 54% of placements, as hospitals prioritised flexibility in capital budgets. Leasing revenue rose 33% YoY, supported by an 18% increase in the installed base under lease and a 10% rise in lease revenue per system, driven by a higher mix of da Vinci 5 and dual-console configurations. The average selling price for purchased systems also rose to $1.6mn from $1.5mn last year, partly offset by higher trade-ins as hospitals redeployed older Xi units across networks.

Service revenue grew 20% YoY to $396mn, reflecting a 13% increase in the da Vinci installed base to 10,763 systems and a 30% expansion in Ion systems to 954. Service revenue per da Vinci system increased 5% YoY as da Vinci 5 placements lifted maintenance and software-related income. This high-margin segment continues to compound steadily, now supported by an expanding base of usage-based contracts that secure recurring revenue visibility.

Gross profit rose to $1.66bn, with margins at 68% on a non-GAAP basis versus 69.1% a year earlier, primarily due to a 90bps tariff headwind, higher facility costs and a greater mix of lower-margin da Vinci 5 and Ion systems. Operating discipline remained intact, with operating expenses up 11% YoY, well below revenue growth, reflecting higher R&D prototype costs and headcount additions, partly offset by reduced legal spend. GAAP operating income reached $760mn and non-GAAP operating income climbed to $976mn, up 29% YoY. Non-GAAP EPS came in at $2.40 vs $1.84, marking roughly 24% underlying growth.

ISRG also announced the release of an AI-powered software upgrade for the Ion Endoluminal System, which helps doctors locate and reach small growths in the lung, known as nodules, more accurately during biopsies. These nodules can sometimes indicate early-stage lung cancer, but because the lungs move as patients breathe, their position can shift between the CT scan and the procedure, making them difficult to target. The new software uses AI to correct for this in real time and adds 3D imaging to give doctors a clearer view, improving accuracy and efficiency during the biopsy. While this upgrade isn’t expected to materially boost revenue, as most customers receive software updates through existing service contracts, it’s a clever piece of technology that enhances the Ion platform’s precision and makes it more appealing for future hospital adoption.

Price Action

Following the earnings beat, gapped up ~17% before closing lower. After having grinded lower since May, it was about time there was some relief! With management’s tone this year being on the negative/cautious side, this earnings beat came as a surprise and clearly highlighted that Intuitive is still going strong. The da Vinci 5 roll out is going strong especially as more countries outside the US begin to adopt these machines. Ion machine has been growing massively and continues to do so. Ultimately, in my view, trading in the 400s was cheap for a company with such a strong position.

At this stage, I wouldn’t be surprised if shares continued to creep higher, breaking the $560 level, and I’d be comfortable adding to my existing position at these levels. If I didn’t have a position I also don’t think that where the price is now would be a bad place to start one. ISRG has always, understandably, commanded a premium valuation. Still, its P/E multiple of 59.2x as of Q3 is notably lower than in prior quarters, which is cheap relative to historical performance but still above its peers.

Outlook

All in all, Intuitive enters Q4 with solid momentum. Management raised full-year procedure growth guidance to 17–17.5% as utilisation remains strong and adoption of the da Vinci 5 continues to accelerate, driving replacement demand and hospital standardisation. The expanding installed base and rising share of lease placements provide a stable base of recurring revenue. Ion is still early in its growth curve, but it delivered over 50% procedure growth, and the new AI-driven imaging upgrade adds meaningful clinical value ahead of broader rollout next year. Margins should hold near the guided 67–67.5% range, with management noting on the call that tariffs are expected to remain a modest headwind, roughly a 70–80bps impact to gross margin. They emphasised that this pressure is being offset by higher production efficiency and mix improvements as da Vinci 5 scales. Overall, Intuitive remains well-positioned for steady double-digit growth as its ecosystem expands across surgical and diagnostic robotics.

Intel

I haven’t covered Intel before, but it’s a story that I’ve kept a close track of, as it’s not often you see such an important player within an ecosystem, “fall off”. Intel designs and manufactures chips that power personal computers, data centres, and networking infrastructure, historically dominating the CPU market through its x86 architecture. For decades, it set the pace for the semiconductor industry, acting as both designer and manufacturer, a model that gave it unmatched scale and control over performance and cost.

However, its decline came from a mix of technological stumbles and strategic complacency that compounded over time. Intel’s long-delayed transition to 10nm broke the manufacturing cadence that had underpinned its dominance for decades, eroding both its cost and performance advantage. While TSMC and AMD advanced to smaller, more efficient nodes, Intel was forced to stretch older designs, widening the performance gap in both PCs and data centres. This lag meant Intel missed the industry’s shift toward systems built around multiple specialised chips, leaving Nvidia and AMD to define the architecture of the AI era. Efforts to diversify into memory, modems and contract manufacturing consumed billions yet added little strategic value, as none achieved scale or competitive differentiation. The result is a company that once dictated the direction of computing, now fighting to rebuild credibility and profitability across technology, execution and strategy, the very pillars that once made it untouchable.

Intel has struggled significantly but ultimately has remained the largest chip manufacturer in the US, which, of course, makes Trump happy. So much so that the government bought a 10% stake at ~$20, a great investment so far!

Q3 was a pretty strong one, which left investors pleased with Intel’s progress over a year ago. Let’s dive into why.

Financials

Intel reports in three main segments: Client Computing Group (CCG), Data Center and AI (DCAI), and Intel Foundry, with a smaller “All Other” bucket housing Mobileye, IMS, and legacy operations like Altera.

Revenue came in at $13.7bn, up 3% YoY and ahead of guidance, marking Intel’s fourth consecutive quarter of sequential growth and signalling tangible improvement in execution. CCG revenue rose 5% YoY to $8.5bn, supported by a seasonal PC refresh, Windows 11 upgrades, and stronger demand for high-end SKUs (stock-keeping unit, a.k.a processors) like Lunar Lake and Arrow Lake, which lifted average selling prices. The rebound in enterprise PCs also helped, as corporates resumed delayed refresh cycles. AI PCs remain early in adoption but are beginning to contribute to mix improvement and ASP support. DCAI revenue slipped 1% YoY to $4.1bn as supply constraints on Intel 10 and Intel 7 capped shipments, though higher enterprise demand and the launch of Granite Rapids Xeon 6 helped offset weakness. The new chips deliver up to 68% total cost of ownership savings versus legacy servers, meaning better energy efficiency and lower operating costs, an increasingly key factor for cloud customers managing data-centre power budgets.

Intel Foundry posted $4.1bn in revenue, down 2% YoY, as internal wafer volumes partly offset softer external demand. The unit remains unprofitable, but the $2.3bn operating loss narrowed sequentially as impairment charges rolled off and yields on Intel 18A improved. Progress on 18A and early customer interest in 14A nodes mark slow but steady validation of Intel’s process roadmap, though scaling remains capital-intensive. All Other contributed roughly $1.0bn, driven by Mobileye’s resilience before Altera’s deconsolidation mid-quarter.

Intel’s 18A process is its current leading-edge manufacturing node, built around new RibbonFET transistors and PowerVia power delivery that boost efficiency and performance. It will underpin upcoming products like Panther Lake and marks Intel’s re-entry into advanced foundry manufacturing. The next node, 14A, remains in development and will push density and performance further, serving as Intel’s answer to TSMC’s most advanced processes later in the decade. In short, 18A is what’s ramping now, while 14A represents the next stage of Intel’s effort to reclaim process leadership.

Gross margin improved to 40.0% non-GAAP (from 18.0% YoY) as restructuring savings, mix improvement, and lower inventory charges offset ramp costs for Intel 18A. Non-GAAP operating income was $1.5bn (vs –$2.4bn YoY), while EPS reached $0.23, up sharply from –$0.46. Adjusted free cash flow turned positive at $0.9bn versus –$2.7bn last year, reflecting disciplined spending and early returns from cost resets. Overall, Intel’s print showed clear progress in stabilising its core businesses and tightening execution, even as the path to sustainable margin recovery in Foundry remains long. Intel also returned to profitability for the first time since 2023, posting net income of $4.1bn compared with a $16.6bn loss a year earlier. The sharp swing back to profit was driven by higher gross margins, lower operating expenses, and reduced restructuring charges, signalling early signs that the turnaround efforts are starting to translate into bottom-line improvement.

Intel’s financial turnaround, if you can call it that, has been underpinned by a sharp reduction in operating expenses, including a 30% headcount cut over the past year as CEO Lip-Bu Tan flattened management layers and redirected resources toward engineering. The balance sheet has also strengthened from the U.S. government’s converted CHIPS Act grants to its 10% equity stake, joined by $2bn from SoftBank and a $5bn investment from Nvidia, which also expanded its AI partnership with Intel.

There had been growing expectations earlier this year that Intel would look to spin off or sell its loss-making Foundry arm to streamline the business. However, that now appears unlikely. The U.S. government, which has become Intel’s largest outside shareholder, has made it clear that retaining advanced chip manufacturing onshore is a strategic priority, and Intel remains central to that goal. While some non-core fabs are expected to close as part of the restructuring to reduce fixed costs and concentrate production at leading-edge nodes like 18A and 14A, Foundry will stay part of the group, both as a political symbol of U.S. semiconductor independence and as a long-term bet that Intel can re-establish itself as a viable alternative to TSMC.

Price Action

shares were up ~8% but closed the day unchanged. To be honest, this chart stings me as I was contemplating entering a position in the $20s, but ultimately decided not to do anything and now shares are nearly up 100% since that point. I think that was a great trade, so if you did enter this, pat yourself on the back.

Given the positive news and recovery path so far, I can see Intel trading higher, perhaps even getting to the 50-handle in the next few months, but this is heavily reliant on Intel’s execution and also the wider macro/tariff environment, which has high levels of uncertainty. Even beyond that, Intel is still far from competitive, and its dominance in x86 is increasingly at risk with AMD closing the gap. Intel wouldn’t be a top-five holding for me, but it deserves a place in a diversified tech portfolio.

Outlook

Intel enters the final quarter of the year with cautious momentum. Management guided Q4 revenue to $12.8–13.8bn, roughly in line with expectations, signalling a stabilising demand backdrop across PCs and servers. The near-term focus remains on executing the 18A ramp and securing external foundry customers to offset the heavy capital intensity of the business. Encouragingly, management noted that foundry customer engagement picked up materially during the quarter, with multiple design discussions now underway, a positive sign that interest in Intel’s advanced manufacturing capabilities is starting to translate into early commitments. AI PC adoption is expected to build gradually through 2025, supporting mix improvement in CCG, while data-centre recovery should remain uneven as cloud customers optimise existing capacity. Cost discipline and leaner headcount provide a stronger baseline for profitability, though gross margins are likely to stay constrained by high setup costs and lower utilisation in Foundry. Overall, the tone of the call was notably more confident, with CEO Lip-Bu Tan saying he has “a lot more confidence” in the company’s trajectory as execution improves and external interest strengthens, signalling that Intel’s turnaround is beginning to gain credibility beyond its own guidance.

I personally do not see a world in which Intel takes a meaningful amount of customers away from TSMC at this stage, as the gap in manufacturing scale, ecosystem integration, and customer trust remains too wide. TSMC’s consistency, efficiency, and deep alignment with major chip designers like Apple, AMD, and Nvidia make it an entrenched partner, while Intel is still proving that it can deliver on schedule and at competitive yields. Even with government backing and early design interest, meaningful share gains in foundry will likely take years and depend not just on technology parity, but on rebuilding confidence that Intel can execute reliably over multiple product cycles. Having said that, this doesn’t mean shares won’t continue to rise…

Final Thoughts

ISRG continues to perform well as a business as the da Vinci 5 rollout gains traction, supported by strong procedure growth and expanding adoption across new geographies. The combination of scale, recurring revenue, and continued innovation keeps its competitive moat firmly intact. Intel, on the other hand, remains a turnaround story. Progress is clear, but so are the challenges. Execution will determine whether it can truly rebuild credibility in advanced manufacturing or remain a politically backed underperformer, which I’m not entirely optimistic about. Nevertheless, it remains an interesting story to follow.

Share

Thank you for taking the time to read this article. I hope you gained value from it. Please make sure to subscribe and share it with others, as it helps us grow! Stay tuned for next week as the heavy hitters report.

Subscribe now

Disclaimer: This article is for informational purposes only and should not be considered financial or investment advice. The views expressed are my own and based on publicly available information, market trends, and personal analysis. I own Intuitive Surgical shares which may change at any time. Readers should conduct their own research and consult a financial professional before making any investment decisions.

This week, risk started on the back foot with equities moving sharply lower, credit spreads widening and gold continuing its rally, as President Trump threatened 100% tariffs on China. Then the TACO (Trump Always Chickens Out) trade became evident as he then said he was having a bad moment and everything would be fine. The causation of this was the Dutch seizure of Nexperia, a Dutch semi-manufacturing company, which was sold to China in 2017. Nexperia produces low-margin semiconductors such as power management chips, diodes, and transistors used widely in consumer electronics and Europe’s auto industry. The seizure followed U.S. pressure to remove Nexperia’s Chinese CEO over security concerns. In response, China imposed new export restrictions on key minerals, including gallium, germanium and graphite used in chip and battery production. But for now, all seems fine on that front.

ASML

If you haven’t read the Q2 article, I highly recommend you do, as it’ll make things easier for you to follow. Since that article, ASML shares have been flying and are up ~42%. I added to my position just as it broke above €700. I think it’s evident that the pessimism at the start of the year was overdone, although it’s easy to say that looking back, ASML is simply not doing as badly as most people expected it to. And the Q3 results reiterated that.

Financials

Total net sales came in at €7.5bn, down slightly from Q2 but up 16% YoY, and broadly in line with expectations. System sales made up €5.4bn of the total, while the Installed Base Management (IBM) segment contributed €2.1bn. Gross margin for the quarter came in at 51.6%, down from 53.7% in the prior quarter due to continued dilution from early High-NA system deliveries, which still carry higher production costs. Operating margin was 32.8% broadly unchanged YoY. Net income was €2.1bn, reflecting solid operational performance despite slightly softer sequential sales.

One of the more encouraging signs this quarter was the resilience in orders, showing that demand for advanced lithography tools remains intact despite broader macro uncertainty. System bookings reached €5.4bn, with €3.6bn coming from EUV and a sharp pickup in memory, which accounted for 47% of the total. Memory sales rose sharply this quarter, mainly due to a rebound in DRAM investment, particularly in high-bandwidth memory (HBM) used for AI accelerators. As AI workloads expand, chipmakers like Samsung and SK Hynix are increasing investment in HBM to support GPUs and data-centre demand, which requires additional and more advanced lithography steps, with SK Hynix also taking delivery of its first High-NA system to support future DRAM development. This drove a surge in orders for ASML’s EUV systems from memory customers, marking the first clear sign of recovery in that segment after several weak quarters.

As a quick sidenote, DRAM is essentially the base technology behind high-bandwidth memory (HBM), which is used in GPUs and AI accelerators to enable faster data transfer and training efficiency. Demand for this has surged as AI infrastructure spending accelerates, with Micron, SK Hynix and Samsung all reportedly operating near full capacity and fully booked for HBM orders well into next year.

A new product was also shipped in Q3, ASML’s first packaging-focused lithography tool, the XT:260. Packaging refers to how chips are connected and assembled once fabrication is complete, a process that increasingly involves stacking or combining multiple chips to improve performance and reduce power use. These steps still require lithography to print the copper interconnects and through-silicon vias that link the chips together.

The XT:260 is an i-line system, using a 365nm wavelength suited for the larger features in packaging layers rather than the ultra-fine patterns in chip fabrication. Despite being based on older technology, i-line remains ideal for speed and cost efficiency in this area. ASML’s new design boosts productivity by up to, supposedly, four times compared with existing solutions, giving the company its first foothold in the fast-growing 3D integration market.

Another interesting thing that happened in Q3 was that ASML also announced a new partnership with Mistral AI, a French company developing large language models for enterprise applications. While ASML is best known for its hardware, software now plays an increasingly important role in optimising chip production. The collaboration will allow ASML to integrate AI models across its product suite to enhance system performance, improve yield for customers, and accelerate development. The €1.3bn investment gives ASML an 11% stake in Mistral and a seat on its strategic committee, reflecting its intent to embed AI deeper into its operations and strengthen its position at the intersection of hardware and intelligent automation.

On the topic of China, management reiterated that while demand in China remains solid in 2025, it’s expected to decline in 2026 as export controls tighten and the market normalises after two years of unusually strong activity. They described recent demand as “very high and in no way normal,” noting that 2026 should mark a return to more reasonable levels following significant front-loading of orders. ASML expects this to be a temporary correction rather than a structural shift, with EUV and service growth helping to offset the decline as TSMC and Samsung continue investing in advanced logic and high-bandwidth memory capacity.

The financial results weren’t groundbreaking, but they are steady and in line with expectations, which ultimately is a positive relative to the negative sentiment earlier in the year.

Valuation

At around €870 a share, ASML trades on roughly 36× current earnings and around 28× forward earnings based on 2025 estimates. That’s a slight premium to peers like Lam Research and Applied Materials, but justified given ASML’s monopoly in EUV lithography. The valuation no longer looks stretched after last year’s correction, and if execution remains steady, the multiple could hold as earnings continue to compound through the decade.

Price Action

The results didn’t make much noise and left shares broadly unchanged. The next level in sight is the almighty €1000. There remain several catalysts that could support ASML moving higher over time, including continued AI infrastructure spending, which drives demand for GPUs and high-bandwidth memory, both reliant on ASML’s EUV systems. TSMC’s upcoming 2nm process is another key driver, as each new node transition brings a fresh wave of tool orders. Positive signs are also emerging from Intel, and if its foundry turnaround gains traction, it could once again become a meaningful buyer.

While I’m not looking to add at current levels, the long-term setup still looks favourable. ASML’s performance is closely tied to its customers’ capex cycles, and with structural demand for compute only growing, its position remains strong. Macro factors may dictate the pace, but fundamentally, the business continues to trend upward. For long-term investors, it may make more sense to build a position gradually and add on weakness rather than sit on the sidelines waiting for the perfect entry

Outlook

For Q4, management guided total net sales between €9.2bn and €9.8bn, with the Installed Base Management segment expected to contribute around €2.1bn. Gross margin is forecast to come in between 51% and 53%. Looking ahead, 2026 is expected to serve as a transition year with softer sales as China demand normalises, before growth reaccelerates in 2027 and beyond. ASML continues to see healthy momentum in both EUV and services, each projected to grow roughly 10% annually as customers expand advanced logic and memory capacity. The company also reaffirmed its long-term outlook from Capital Markets Day, targeting €44–60bn in revenue by 2030 and gross margins of 56–60%. Management highlighted that the structural drivers behind demand remain intact, supported by increasing chip complexity, accelerating AI adoption, high-bandwidth memory growth, and continued node transitions at leading foundries.

Q3 results haven’t changed my view that ASML should be a core holding in a tech portfolio, as it’s the only company capable of doing what no one else can replicate. But beyond that, the catalysts mentioned in the price action segment give me the confidence to continue holding and adding on any weakness (depending on the cause).

TSMC

Since the Q2 article, TSMC is up ~24%, performing strongly in line with the rest of the sector. Like ASML, TSMC is, in my humble opinion, another core position within a tech portfolio that reported its highest quarterly profit in Q3.

Financials

Total revenue came in at $33.1bn, slightly above guidance and up 10% QoQ and 41% YoY, supported by resilient demand for advanced process technologies and strong seasonal strength in smartphones. Gross margin expanded to 59.5%, improving 90bps sequentially. Gross margin benefited from efficiency gains and improved utilisation at N3 and N5 but remained partially offset by higher depreciation and ramp costs tied to overseas fabs in Arizona and Japan, which is expected. Operating margin rose to 50.6%, while net income reached $14.1bn, up 14% QoQ and 39% YoY, translating to EPS of $0.55. The solid profitability reflects improved utilisation at leading nodes and operating leverage from higher volumes, with performance well ahead of initial expectations earlier in the year. Free cash flow came in at $4.3bn, after $9.7bn in capex and $3.9bn in dividend payments.

Advanced nodes continued to dominate, with 3nm contributing 23% of wafer revenue, 5nm 37%, and 7nm 14%, meaning technologies 7nm and below accounted for 74% of total wafer sales. The mix underscores TSMC’s strong leadership at the cutting edge of logic manufacturing, with growing exposure to AI workloads, HPC chips, and next-generation smartphone processors. Volume production for 2nm is set to begin in Q4, marking an important milestone for TSMC’s next technology cycle.

By platform, HPC (high performance computing) remained the largest contributor at 57% of revenue, showing its growing importance as AI and data centre demand offset a pause in traditional compute orders. Smartphones accounted for 30% of revenue, rising 19% QoQ as Apple’s seasonal ramp for its 3nm A18 chip drove volume growth. IoT and automotive each contributed 5%, up around 20% and 18% QoQ, respectively, as chip content in connected and electric vehicles continues to increase. The only weak spots were DCE (-20%) and Others (-8%), both reflecting inventory adjustments in mature nodes.

CapEx came in at $9.7bn this quarter, largely in line with expectations, as TSMC continues to expand 3nm capacity and lay the groundwork for its 2nm node. Spending remains elevated due to ongoing fab construction in Arizona, Japan, and Germany, though management noted that each project is backed by customer demand and long-term commitments. Around three-quarters of total CapEx continues to flow into advanced nodes, underscoring TSMC’s focus on staying ahead of competitors as the industry shifts toward 2nm production in 2026.

A significant portion of this investment is going into advanced packaging and backend expansion. Management said demand for CoWoS (Chip-on-Wafer-on-Substrate) packaging remains well above supply and is set to grow further as AI chip complexity increases. TSMC plans to boost CoWoS capacity by over 60% next year and continue scaling its SoIC 3D integration and InFO technologies. Mask complexity and backend testing also remain areas of heavy investment, as each new node requires more layers and tighter inspection. These technologies are quickly becoming just as critical as lithography itself in sustaining TSMC’s performance edge.

Share

A notable milestone this quarter was that the first Nvidia Blackwell GPU was manufactured at TSMC’s Arizona plant, marking a key validation of the company’s overseas expansion strategy. It’s an early but important proof point that the heavy investment in U.S. manufacturing is starting to pay off, both in terms of operational readiness and customer trust. While production in Arizona is still limited compared to Taiwan, it signals that TSMC’s diversification efforts are moving in the right direction and could help strengthen its strategic positioning with major U.S. clients over time.

Price Action

Shares in initially jumped to a high of 3% of the previous day’s close but were rather unchanged at close. I think a lot of the upside is priced in at this point, so it makes it harder for strong earnings to propel shares higher. For this reason, I have no intention of adding to my existing position at this point. TSMC trades on roughly 30× current earnings, which sits above its long-term average but remains reasonable given record margins and its dominant position in advanced foundry manufacturing.

Outlook

The tone this quarter was notably more upbeat than Q2, with management raising guidance and expressing confidence that AI-driven demand remains stronger than anticipated. Full-year revenue growth guidance was set at the mid-30% range in USD terms, up from around 30% previously. The upgrade reflects stronger-than-expected AI demand and sustained momentum in advanced nodes, particularly 3nm. For Q4, revenue is guided between $33.5bn and $34.3bn, with gross margin expected to hold around 59–60%. While 2026 is expected to be a year of slower growth as customers rebalance inventories, management sees a strong reacceleration into 2027 as 2nm enters volume production and new AI chips ramp. They reaffirmed their long-term outlook for mid to high teens annual growth, supported by secular demand from AI, HPC, and automotive.

I think as a business, the future is bright for TSMC, but ultimately, I feel that the next most likely outcome is a leg lower than a leg higher. If I’m wrong, bluntly speaking, I don’t care, as I still hold a position, and if I’m right, then it will give me a chance to add to my position, which is always good (as long as it doesn’t keep going down).

Final Thoughts

As I’ve said for both companies, they’re, in my view, core holdings in any tech-focused portfolio, given their critical roles in the semiconductor supply chain. Without them, the modern world simply wouldn’t function as it does. I see more upside for ASML at this stage, but over the next five years, I expect both to be trading higher than today, with inevitable bouts of volatility creating opportunities to add or start a position. Overall, two solid sets of results that reinforce just how central both companies remain to the future of technology.

Thank you for reading this article. I hope you found it insightful. Feedback is always welcome, so please feel free to fill out the survey below. Make sure to be subscribed to receive the next exciting article straight into your inbox!

Feedback Survey

Disclaimer: This article is for informational purposes only and should not be considered financial or investment advice. The views expressed are my own and based on publicly available information, market trends, and personal analysis. I own both ASML and TSMC shares, which may change at any time. Readers should conduct their own research and consult a financial professional before making any investment decisions.