On all metrics, token consumption is exploding. According to OpenAI’s recent enterprise report, API token consumption for reasoning increased 320-fold year-over-year. In the meanwhile, the current shift to AI agents—which communicate with each other over multiple steps—is again compounding reasoning volumes. Goldman notes that China enterprise token usage grew 162% from June to December, while China is even more short on compute than the West:

We can see on OpenRouter’s data that token consumption growth has strongly accelerated again as from January of this year:

While optimization techniques and hardware improvements have been driving sharp declines in the cost per token, demand is ramping at such a fast pace that capacity remains the binding constraint. This is UBS on OpenAI’s presentation at GTC:

“The cost of intelligence continues to fall rapidly, but demand growth still appears likely to keep the world compute constrained. The speaker highlighted sharp declines in token costs (99.7% decline in 3yrs) driven by both hardware and software improvements, while also suggesting demand is ramping fast enough that long-term compute availability remains a binding constraint. The broader implication is that “gigawatts” alone are too coarse a metric, with the shape and composition of compute becoming increasingly important as workloads diversify.”

A key focus in the coming years is on reducing latency. You don’t want a software engineer that costs $300-500k per year to sit idle for too long as he waits while Claude is generating code. This is again UBS on this trend after watching the presentation from Nvidia’s Bill Dally and DeepMind’s Jeff Dean:

“Models have improved materially on coding, math, and longer-running agentic workflows, with inference speed now emerging as the next key bottleneck. The panel suggested that agents can already sustain materially longer autonomous workflows than a year ago, with the next step less about raw capability and more about scaling throughput and lowering latency so agents can operate at practical speed (that far exceeds human speed). In that new paradigm, certain software tools developed for use at a human speed may need to be reimagined for much faster agentic workloads.”

This is the reason why Nvidia is now bringing to market a unique and low latency inference solution. Whereas Nvidia’s Rubin GPU will handle the memory-intensive prefill phase with its massive capacity of HBM, the new Groq LPX chip will accelerate the token generation in the decoding phase. Essentially, Groq LPX is a chip that heavily makes use of SRAM, resulting in an extreme amount of bandwidth and thus low token latencies:

This is Jensen on this new solution for high-performance AI workloads:

“So we acquired the team that worked on the Groq chips and licensed the technology, and we’ve been working together now to integrate the system. At the most valuable tier, we’re now going to increase performance by 35x. I would add Groq to maybe 25% of my total data center, and the rest of my data center is all Vera Rubin.

The reason why Groq was so attractive to me is because their computing system is a deterministic data flow processor. It is compiler scheduled, meaning the compiler figures out when to do the compute as the data arrives at the same time. All of that is done statically in advance and scheduled completely in the software.

The architecture is designed with massive amounts of SRAM, it is designed just for inference. And as the world continues to increase the amount of high-speed tokens it wants to generate, the value of this integration is going to get even higher. We’re going to fuse Groq and Vera Rubin, and use Groq to process the very last stage of auto regressive language models. That last stage is extremely bandwidth-intensive.

It would take a lot of Groq chips to be able to hold the parameter size of Rubin as well as all of the KV Cache that has to go along with it. So that limited Groq’s ability to really reach the mainstream and take off, until we had a great idea. What if we disaggregated inference altogether with a piece of software called Dynamo so that we could put the work that makes sense on Vera Rubin and then offload the low latency, decode generation part of the workload on Groq. And so we unified processors of extreme differences, one for high throughput and one for low latency.

It still doesn’t change the fact that we need a lot of memory. And so we’re just going to add a whole bunch of Groq chips, which expands the amount of memory. And so for a 1 trillion parameter model, we have to store all that in Groq chips. However, it sits next to Vera Rubin, where we hold the massive amounts of KV Cache that’s necessary in processing all of these agentic AI systems. So, the attention part of decode is done on Vera Rubin and the token generation part is done on the Groq chip. We run Dynamo, this incredible operating system for AI factories on top of it, and you get a 35x increase for token generation the world has never seen.

We’re in production with the Groq chip, and we’ll ship it in the second half, probably about Q3 time frame. I want to thank Samsung, who manufactures the Groq LP30 chip for us, and they’re cranking as hard as they can. The sampling of Vera Rubin is going incredibly well. And in fact, Satya texted out already that the first Vera Rubin rack is already up and running at Microsoft Azure. We have now set up a supply chain that can manufacture thousands a week of these systems, essentially multi-gigawatts of AI factories per month inside our supply chain.”

Groq LPX in combination with Rubin especially boosts the bandwidth and thus the interactivity per user:

UBS came away thinking that the risk-reward is towards the upside of their revenue estimates:

“In a very well attended session, management highlighted the potential sources of upside to the updated backlog slide presented yesterday and we walked away a bit more optimistic as to how much LPX (the new fast inference solution from Groq) could add in C2027. We took from the discussion that ~$50B could be a reasonable estimate of how much LPX would add on top of the ~$460-470B implied baseline for data center revenue in C2027. On top of this, of course, would be additional business that comes in within lead times – which could be as high as ~$50-75B or even more. This is because we estimate data center will end up at least $40-50B higher in C2026 than what was implied when the company first provided this backlog last year – ergo, with 21mos still left until the end of C2027, the company should be able to add at least this much to C2027. So net of all this, we could see a data center revenue number potentially pushing $600B in C2027 (UBSe ~$520B).”

The golden question is how AI demand will come in over that fiscal ‘29-’31 period for Nvidia, as we already know that this year (FY27) and also next year (FY28) will come in strong. However, the market is assuming we’re now heading close to a peak in terms of AI capex spend which we can also see in UBS’ numbers:

As a thought exercise, we tried to estimate Nvidia’s future annual revenues based on where AI demand will go five-six years from now in 2031. We start from current AI capacity, which is basically the sum of Nvidia’s revenues over the last three years. So, this should be roughly the current installed base excluding Google’s TPU-based data centers. This results in $407 billion of current AI capacity.

We know that this is by far not enough—the market is still undersupplied while token consumption will continue to explode. If we assume that current AI capacity satisfies 10% of future AI demand in 2031, we need an installed base of just over $4 trillion of Nvidia equipment by 2031. If we assume a lifespan of 6 years for this equipment, that demand will be fulfilled by the equipment Nvidia ships over the 2026-2031 period (6 years in total). This would mean Nvidia would sell on average $678 billion of equipment per annum in this time period:

AI demand going up 10x by 2031 is our personal best guess. Of course, it could easily be 15x, or it could only be 5x. There are a lot of moving parts—on the one hand, models are growing larger but also optimization techniques for inference are driving the cost per token down. What we feel very confident about however is that we’re very early in the adoption curve. This can easily be witnessed in the steep growth in token consumption, as well as the continuously expanding use cases of AI—from coding and AI agents in SaaS to realistic video generation, robotaxis, humanoid robotics etc. This is UBS on how AI agents will become an important driver for SaaS revenues for example:

“Also particularly intriguing to us was the CEO’s perspective that the future for software industry with AI, might shift toward monetization focused more on AI token reselling instead of historical software tool licenses/SAAS. In this new paradigm, future software companies’ COGS increase and come in a form of token consumption. At the same time, NVDA expects cloud business model to evolve from instance rentals to monetizing tokens per second (with various tiers offered at different prices). NVDA sees hyperscalers aiming to optimize tokens-per- second-per-watt.”

However, a potential problem for Nvidia is that all hyperscalers are developing their own ASIC accelerators and will gradually shift more demand towards this in-house silicon, similarly to what happened at Google already. Or what has been happening in the data center CPU market with the shift towards Graviton ARM-based CPUs at AWS. This is William Blair on Meta’s recent progress with MTIA as an example:

“In a blog post yesterday, Meta provided greater details on its custom ASIC roadmap. Meta noted that the current generation of MTIA 300 chips is already in full production, with three additional generations targeted for release in 2026 (MTIA 400) and 2027 (MTIA 450 and 500). All four generations leverage Broadcom as a key design partner, with the company contributing significant IP for I/O, networking, and chiplet connectivity. Recall that on its most recent earnings call, Broadcom noted it expects Meta to deploy several gigawatts of its next-generation XPUs in 2027 and beyond. The announced roadmap provides greater details on the design priorities for these new chips (with a refocus on LLM inference) and helps instill greater confidence that Meta will be able to scale its ASIC deployments over the next few years.”

We know that hyperscalers will be aggressively ramping ASIC capacity from Broadcom’s calls. This is from JP Morgan:

“We expect Broadcom to generate close to $65B+ in AI (ASIC + networking) revenues in FY26, led by its primary TPU customer, Anthropic ramp, the first- generation ramp of SoftBank/ARM XPU ASIC programs, and incremental revenue contributions from other AI ASIC customers such as Meta and Bytedance, along with continued strong AI networking revenue. For FY27, Broadcom anticipates close to 10 GW of total compute capacity being deployed across its six XPU customers. Specifically, OpenAI has now become a qualified customer and is expected to deploy over one gigawatt of compute capacity. Anthropic’s order deployment is projected to increase 3x Y/Y in FY27 to 3 GW. Meta remains on track with its internal ASIC program, with multiple gigawatts expected to be deployed in FY27. Customers 4 and 5 (which we believe to be SoftBank/ARM and Bytedance, based on a process of elimination given Broadcom’s disclosures) are expected to have strong shipments this year and more than double in 2027. All of this is in addition to the continued strength of Broadcom’s lead customer and the ramp of the TPU v8 (Sunfish) program. Overall, we estimate Broadcom can drive $12–15B of revenue per gigawatt and conservatively see AI revenues at $120B+ in FY27. Reflecting increasing AI customer diversification, we estimate Google TPU business will decline to less than 45% of overall AI revenues in FY27, down from 50–52% in FY26 and 60% in FY25—even as total AI revenues grow to over $120B in FY27.”

Based on these numbers we can do some modeling. We know that Nvidia makes around 50-60% of their revenues from hyperscalers, so if we then compare these revenues to Broadcom’s AI revenues, we know that Broadcom will grow its market share at the hyperscalers to around 27% by next year:

Will ASICs disrupt the Nvidia investment case? Next, we’ll dive further into developments in AI and three trades we’re making.

Additionally, humanoid robotics should be one of the big growth stories in tech in the coming decades. While the robotics field is competitive, we recently came across a well-positioned components supplier with good engineering prowess that started signaling high growth from both humanoid and quadruped robotics in the coming years. Typical investment criteria we look for are—high market shares and margins, combined with a long runway of growth so that we can simply buy and hold for the next 5-10 years with the aim of seeing shares compound into attractive multibaggers.

Read more

HBM (high-bandwidth memory) is a workhorse in the current AI boom. In HBM, DRAM dies are stacked on top of each other and connected through TSVs. This design allows for loads of fast working memory to be located close to the GPU. Bonding these DRAM dies together is a key step in back-end manufacturing (advanced packaging) as it determines the interconnect density, and thus speed and bandwidth.

This bonding process was dominated by Hanmi Semi over the last few years as the key supplier to SK Hynix (and Micron). However, the SK Hynix relationship soured when the latter started evaluating equipment from other suppliers. Currently, HBM die bonding is dominated by a technique called thermo compression bonding (TCB), and Hanmi Semi’s erratic behavior towards their key customer has now allowed ASMPT to get a strong foothold in the TCB market. ASMPT’s growth has exploded in the TCB market, and is now guiding for a 35-40% market share in this high-growth TAM. This is the CEO on the recent call:

“TCB momentum strengthened further in 2025 with significant new orders across logic and memory, solidifying our TCB technology leadership. We established deep engagement with both logic and memory customers and saw encouraging traction in areas such as HBM and C2W (Chip-to-Wafer) ultrafine pitch applications. This continues to reinforce our position as a leading provider of advanced packaging solutions as customers move to more complex chiplet-based and high-density architectures.

This time last year, we expected the TCB TAM to reach around USD 1 billion by 2027. Since then, the landscape has evolved meaningfully. The acceleration of AI-driven investment—especially in advanced logic and high-bandwidth memory—has expanded the market significantly more than our earlier assumptions. Based on our latest projections, we now estimate the TCB TAM to grow from roughly USD 759 million in 2025 to USD 1.6 billion by 2028, representing a CAGR of 30%. This reflects sustained adoption of 2.5D architectures, higher HBM stacks, and the industry’s move towards final pitch interconnect. These are all areas where TCB is increasingly the preferred solution.

Our target market share remains at 35% to 40%. With our breakthrough into the competitive HBM market, we grew TCB market share significantly, achieving record TCB revenue growth of 146% year-on-year. In 2025, our advanced packaging (AP) revenue growth of 30% year-on-year was driven by TCB.

Now, let’s look at TCB more closely. In logic, our C2S (Chip-to-Substrate) solution maintains its dominant position as a process of record with a steady flow of orders from key OSAT customers in 2025. Extending into early 2026, we are pleased to share that we have secured additional orders for 9 more TCB tools from the same customer. We are well positioned for further order wins as the market shift towards larger compound lines. At the same time, our C2W ultra-fine pitch platform secured orders for 2 tools in February 2026 from a leading customer. Since the announcement, we have secured 2 more such TCB tools from the same customer. As the industry transitions from mass reflow technology to TCB, ASMPT stands to benefit significantly as the preferred C2W solution provider offering plasma enabled capabilities.

In memory, we deepened our engagement with several customers and continue to expand our share with shipments in Q4 2025. Our tools have demonstrated superior performance with industry-leading production yields and interconnect quality. We were also the first to secure HBM4 12-high orders from multiple players, and we are now leading HBM4 16-high development with our flux-based TCB tool deployed for sampling, and our fluxless AOR-TCB process under qualification. These are important milestones for our technology leadership as HBM architectures scale further.”

During the Q&A we got more details on ASMPT’s positioning for the coming years:

“Currently, the HBM portion in the TCB TAM is definitely much larger than logic. But as we go a few years out, this dynamic will actually shift where logic, especially CoW, should take a larger share. But we cannot share the specific split for confidentiality reasons or competitive reasons. ASMPT is very strong in CoS and we are also making wins on CoW. Recently, we won two tools for C2W application and then we won two more. So I think it is a signal that we are also being recognized as a solid solution provider for the C2W space.

Now on the memory side, I think the competition landscape is different. We have a strong incumbent in the memory space, but a year ago, we broke into the HBM market and so we have gained market share there as well. In 2024, we had practically zero share in HBM. In 2025, we have a strong foothold in the memory market. For 2026, orders will come as our customer rolls out 16-high HBM in volume. As the industry moves from 12-high to 16-high, I think all equipment suppliers, including myself for TCB, are waiting anxiously for that particular customer to allocate TCB demand. So at this moment, we feel that 2026 will be a year whereby there will be new true demand for TCB for 16-high, but exact timing, I cannot give you any visibility at this point in time, but it cannot be too long before we know in our opinion.”

For years, Hanmi Semi (a South Korean equipment maker) was the dominant supplier of mass reflow bonders to SK Hynix, the leading player in HBM. However, diversifying the supplier base is fairly common practice for all leading fabs. For example, Intel for a long time kept sourcing a smaller share of its less crucial lithography tools from Nikon, even though those had far inferior performance compared to ASML. The reason is that they wanted to keep Nikon alive in the lithography race and not become fully reliant on ASML. So, it’s fairly common to give 70-80% share to the best equipment maker, and then 20-30% to a secondary player so that the fab can keep multiple options open. This also provides funds to the second player to stay alive in the innovation race.

So, in our view, it’s fairly normal that SK hynix started placing small orders with Hanwha Semi and ASMPT to test out their TC bonders. However, according to media reports, Hanmi subsequently recalled 60 engineers from SK Hynix’s fabs which were servicing the bonding equipment. Additionally, Hanmi notified SK Hynix of a 25% price increase. This led SK Hynix to accelerate its shift towards another strong supplier.

Because ASMPT’s equipment had apparently been successful in testing, this allowed the company to capture substantial TCB order flow in 2025. The Elec mentions that ASMPT has captured half of HBM4 TC bonders so far:

“SK Hynix is expected to place a large order of a hundred set of TC bonders for the production of HBM4 in March. This is when it plans to expand 1b DRAM production at its M15X fab at Cheongju. It remains to be seen how the chipmaker places orders for these bonders to ASMPT, Hanmi Semi and Hanwha Semi. SK Hynix uses around fifty sets of TC bonders for the production of HBM4 at the current stage, and half of these have been supplied by ASMPT. Hanwha Semitech’s bonders were not in use, sources said.”

According to Electrical Engineering World, ASMPT’s bonders are outperforming those of Hanmi Semi:

“A person familiar with SK Hynix said, ‘During the process simulation of HBM3E 16-layer, we found that the performance of ASMPT equipment exceeded that of Hanmi equipment as the number of stacked layers increased.”

So, it’s not really surprising that Hanmi Semi has been reporting disappointing revenues. This is JP Morgan on the company’s recent results:

“Hanmi’s 4Q earnings were a miss, with the OP coming in 40-60% below JPMe/ Bloomberg consensus with TCB sales recording the lowest since 4Q23. After two quarters of earnings misses from HBM4 qualification delays, we expect TCB earnings to recover in 1H26 from memory makers resuming TCB orders. Despite memory capex entering an upcycle, however, we see unfavorable risk/reward with Hanmi trading at a 30%-35% premium to peers with Samsung TCB optimism priced in. We remain UW with a PT of W150K based on 29x FY27E P/E and recommend investors trim at strength.

Hanmi shares have outperformed peers YTD which we attribute to: 1) memory capex entering an upcycle; 2) Samsung TCB optimism; and 3) the market’s perception of Hanmi being a higher beta to memory stocks. We agree that memory capex is entering an upcycle, but we emphasize that the upside risk is on the front-end capacity and conventional DRAM more than HBM where we do not sense any upward revision to HBM capacity in the next 12M. Samsung is testing both HCB and TCB for 16-Hi HBM4E, and while we expect Samsung to converge to TCB, we see limited reasons for Samsung to diversify TCB vendors.”

Samsung is heavily relying on its internal equipment maker, SEMES, to supply its TCB tools. This means that it is unlikely Samsung will adopt Hanmi’s equipment, unless Hanmi’s equipment strongly outperforms SEMES tools. To make matters worse for Hanmi, Macquarie points out that Hanmi could also lose share at its other key customer, Micron, which has been testing out ASMPT as well as Besi:

“Hanmi continues to face Hanwha Semi and ASMPT in SK Hynix’s value chain. At Micron, Hanmi remains dominant, but Micron has tested ASMPT and Besi, increasing competitive risk. While Hanmi entering Samsung Electronics’ HBM bonding is a catalyst that the market awaits, we see Samsung prioritising its subsidiary SEMES and considering alternatives rather than adopting Hanmi tools.”

Asian broker CGSI models in revenue growth for Hanmi with as demand source Chinese packaging fabs:

However, this is a somewhat unreliable source of demand as increasing sanctions pose a looming risk, especially as Huawei is the end customer which has been a constant target of sanctions. From CGSI:

“As China pushes for advanced chip localization, centered on Huawei, we expect Chinese outsourced semiconductor assembly and test players (OSAT) to place large-volume orders for critical packaging process tools to Hanmi and key Japanese players, driven by Huawei’s production ramp-up for artificial intelligence (AI) chips amid the depletion of pre-secured HBMs. We expect Chinese HBM players to start mass volume shipment in 2H26F and project their monthly HBM through silicon via (TSV) capacity at 60k by end-2026F and 127k by end-2027F. Assuming 50% thermo-compression bonder (TCB) market share of Hanmi in China, we expect its China-bound bonder exports to jump to W164.6bn in FY26F and W183.9bn in FY27F from W58.9bn in FY25F, contributing 38% and 40% in total TCB sales in FY26F and FY27F, respectively, up from 17% in FY25F.”

Overall, the TCB semicap market is an interesting space for long term investors. Not only does it play a crucial role in the HBM market, but it’s also seeing increasing adoption in the logic market for 2.5D chiplet integration which is crucial to pack more compute into a single system. As Moore’s Law has slowed, the obvious way to pack more transistors into a single computing system is to simply cobble more dies together. AMD has been at the forefront of this trend:

ASMPT is forecasting a 30% CAGR in the TCB TAM, while ASMPT is the player with the momentum in this market. Obviously, the company is well-positioned to capture a strong market share of this future TAM.

Is it time to invest in ASMPT? Or are other names even better positioned to capture the value from the attractive outlook in advanced die bonding for logic and HBM end-markets? We’ll dive further into this space next.

HBM/Die Bonding Semicap Outlook

Read more

MRAM (Magnetoresistive Random Access Memory) combines the best of both worlds in memory—the permanent storage capabilities of NAND and the high speed read and write capabilities of DRAM. Its permanent storage capabilities are even better than NAND, as NAND flash physically degrades every time you erase and write new data to it. So, SSDs have a limited lifespan. On the other hand, MRAM has no physical wear and tear which results in a near-infinite endurance. MRAM also has been called ‘universal memory’, as it theoretically could tackle every use case.

However, the MRAM market remains small today, with as key reason that the cost per bit remains high. Nevertheless, the MRAM market should see attractive growth rates in the coming decade(s) as it plays a crucial role in edge intelligence such as satellites, autonomous vehicles, industrial robotics etc. FMI forecasts the MRAM market to grow at an 18% CAGR over the next 10 years:

“The global MRAM market is expected to grow from an estimated USD 912 million in 2025 to USD 4,773 million by 2035, reflecting a robust CAGR of 18.0% over the forecast period. Growth is primarily driven by the increasing demand for high-performance memory solutions in automotive, aerospace, industrial, and consumer electronics applications.

MRAM offers distinct advantages over conventional memory technologies, including non-volatility, low power consumption, high endurance, and faster read/write speeds, making it an attractive option for next-generation computing systems and IoT devices. Adoption is particularly strong in embedded applications where energy efficiency and reliability are critical, such as in automotive ECUs and industrial controllers.”

Aerospace (satellites) should become a key end-market as MRAM is radiation immune. Traditional memory, like SRAM and DRAM, stores data using electrical charges. In the harsh environment of space, cosmic rays and high-energy particles can alter charges in traditional memory causing bit-flips, i.e. accidentally changing a 1 to a 0. Because MRAM stores data magnetically rather than electrically, it doesn’t have this problem. So, you can use MRAM on satellites without the bulky shielding required for standard memory. Additionally, LEO satellites have a short lifespan of around 5 years or less, making this a business with regularly recurring revenues as satellites have to get replaced.

Everspin (ticker MRAM, enterprise value $159 million) is a leading company in the MRAM space. The CEO guided for their revenues to nearly double in the coming years:

“Before I close, I would like to discuss our long-term strategy, which entails reaching $100 million in annual revenue over the next 3 to 5 years. We believe this growth will be driven by the ramp of new products, most notably our new XY parts in our PERSYST product portfolio, such as the 64-megabit part I described earlier and continued solid growth in our Toggle MRAM and licensing business. Our new PERSYST X5 parts are getting solid traction. They are offered in densities from 4 megabit to 256-megabit and include the power-optimized SC families and the high reliability or HR families with quad and optical SPI interfaces.”

As we know, memory is currently supply constrained with manufacturing capacity heavily being reoriented towards HBM and DRAM. Can this have downstream effects where excess demand starts flowing over into the MRAM market next? Is Everspin next to benefit from supply shortages in the overall memory industry? We’ll take a deep look at all relevant topics next.

Everspin & the MRAM market

Read more

We’re interested in the power semi space for a number of reasons. First, these types of semiconductors will start playing an increasing role in both the AI data center and their connected power sources. Second, these are not expensive stocks, as these companies are still crawling out of the semi downturn post the covid electronics boom. Thirdly, these companies are exposed to a variety of attractive end-markets, ranging from robotics to EVs, drones and green energy. Robotics and increasing levels of automation in the wider economy can provide a big tailwind over the coming ten years, making mature semis in general interesting plays on the long term growth in edge AI.

William Blair notes that ON Semi can make $100k per MW of AI compute:

“Power is the key bottleneck to AI infrastructure, making power density and efficiency critical. Onsemi now estimates its power semi content per next-gen 1 MW AI rack has doubled from $50,000 to $100,000, supported by its broad SiC and GaN portfolio. The company generated $250 million from these applications in 2025 with high- teens sequential growth, with much larger upside expected as higher power architectures scale toward 2030.”

Infineon is another key player in power semis and guides for $12k–$15k of content for a current 130kW AI rack. Taking the midpoint, $13.5k of power semi content per 130kW rack means $104 per kW, or $104,000 per MW.

If 30GW of AI compute will be added per annum, this means a power semi TAM of around $3.1 billion from AI compute alone:

Based on Infineon’s recent comments on the conference call, it looks like the German leader in power semis will grab a large share of this market:

“Our unmatched portfolio breadth, system understanding, speed of innovation, quality and delivery capability make us the partner of choice for all relevant GPU makers and cloud providers. Our revenues from AI power solutions keep growing steeply. We re-confirm our target of around EUR 1.5 billion for this fiscal year. This number is still supply limited, capped by how fast we and our manufacturing partners can bring up capacity. And let me be clear, in contrast to some peers, which in their disclosure do not differentiate between AI and other enterprise data center business, the EUR 1.5 billion of Infineon are purely related to AI. Separately, we do another EUR 500 million, give or take, of classical data center revenues.

Meanwhile, projections of build-out of AI data centers and related infrastructure keep going up. More and more, AI is showing tangible benefits in real-life use cases. Several of these we witnessed in our own company, for example, improving speeding up chip design, software development or customer engagement. The demand from our customers for the coming years is still going up quarter by quarter. Therefore, we are doubling down on AI. We have decided to pull in EUR 500 million this fiscal year to accelerate AI-related capacity increases in order to fuel growth beyond the running fiscal year.

These investments, including the conversion of existing IGBT capacities are capital efficient and will underpin our market leadership. Large part of them will be used for a faster than originally planned ramp-up of our new power and analog/mixed-signal module fab in Dresden. Fortunately, we are able to accelerate the opening of the facility to the summer. With the additional capacity, we project to achieve AI-related revenues of around EUR 2.5 billion in our 2027 fiscal year. Said differently, we will add another EUR 1 billion of margin accretive business next year, thus expanding our AI revenues by a factor of 10 within just 3 years.

We are not only engaged with 1 or 2, but with all hyperscalers and relevant GPU makers across the globe. And for us, of course, the second stage part is giving us particular growth momentum as vertical power delivery modules have a much higher value. It’s the biggest opportunity I’ve seen for the company in the last 30 years, but also technically very challenging with very demanding customers.

From a utilization point of view, you can assume that the 300-millimeter factories will soon be fully loaded as we are converting IGBT to AI MOSFETs. Then, we still have some idle capacity in non AI-related technologies, but also here, we expect a certain spillover or halo effect from this AI boom, meaning that MOSFETs going into telecommunication or automotive will also see demand.”

Based on what the various competitors are saying on their conference calls, we believe that Infineon currently has a fairly dominant share in this market.

Wolfspeed illustrates that power semis are also used in cooling, as well as to convert power from the grid so that a single GPU can work at a lower voltage level. This will mainly be done by SiC (silicon carbide) whereas GaN (gallium nitride) is more heavily used at the rack level.

The reason is that SiC is better at handling high power, whereas GaN is better at high frequencies. Yole estimates the SiC market to grow at a 21% CAGR from 2026 to 2030, but GaN should grow faster as it will take the larger market share in the AI data center.

Let’s go through some of the players in this market. If dominant player Infineon can’t keep up with demand, it’s likely that the other players will see a demand boost as well.

Wolfspeed, the SiC Specialist

Wolfspeed is a pioneer in SiC and has historically held a strong market share in this market. Infineon actually tried to acquire Wolfspeed in 2016, which is how we first gained knowledge of Wolfspeed’s strong SiC capabilities. However, this acquisition got blocked by the US government on national security grounds, as Wolfspeed makes SiC modules for the US military.

Wolfspeed has now completed a state of the art SiC fab in New York State for $6 billion, however, as the EV market went into a downturn over the previous years, the company blew up due to excessive leverage. Basically, Wolfspeed loaded up on debt to build this fab but when SiC demand came in lower than expected, shareholders got wiped out.

This debt load has been restructured now—basically, the old debt holders have now become the new equity holders, and trading resumed a few months ago on the New York Stock Exchange. The good news is that the financial situation is now looking much better—the fab is ready, allowing the company to limit capex in the coming years, while the balance sheet has now $1.3 billion in cash and with a reduced debt load of $2 billion, giving an overall net debt figure of $700 million.

The bad news is that fab utilization is still low, which resulted in negative gross margins last quarter, and so the company will need to see either strong growth coming from EVs and industrials, or alternatively, start winning deals in the AI data center space. The cash burn was $74 million last quarter, so the $1.3 billion in cash gives shareholders more than 4 years of runway at the current state. However, new management is heavily focused on cutting costs while at the same time getting those new SiC wins. The company had a design win with Toyota last quarter for example for their battery electric vehicles (onboard charger SiC chip). At the same time, also Wolfspeed’s AI data center revenues are growing tremendously, although this will be from a very low base.

This is new CEO Robert Feurle, who is semiconductor veteran with previous VP roles at Infineon and Micron:

“We have continued to build solid momentum across the business this quarter from achieving 50% quarter-over-quarter growth in AI data center revenue to producing a 300-millimeter silicon carbide wafer and securing key customer wins. We’re concentrating in a few key areas: strict financial discipline, advancing our technology leadership, and driving operational excellence. A central theme across these priorities is diversifying our revenue base, particularly in industrial and energy, including applications tied to AI-related power demand and grid modernization.

We remain focused on driving costs out of our footprint as we navigate underutilization headwinds. We have officially completed the shutdown of all 150-millimeter device production in our Durham campus, roughly a month ahead of schedule, transitioning our entire device platform to higher efficiency 200-millimeter manufacturing. We continue to improve production efficiency and speed to optimize the earnings potential of the business. The results of these efforts will be even more apparent when demand accelerates and we begin to increase fab utilization.

We have organized our go-to-market strategy around 4 verticals that we believe will drive growth in our business in the near to midterm—automotive, industrial and energy, aerospace and defense, and materials, and we are already seeing strong traction from these early efforts. Our first vertical, automotive, remains a core market despite muted EV demand due to a mix of macro and structural factors, which include higher interest rates in the U.S. and Europe, the elimination of certain government incentives in the U.S., excess supply across the market and intensifying competition globally, including China. Despite weaker near-term demand, our portfolio is aligned with OEMs that provide us efficiency, range and power density. A great example of this is our recently announced partnership with Toyota, one of the most respected and quality-driven automakers in the world, to power the onboard charging systems for their BEVs.

In industrial & energy, we are leveraging our expertise to expand our reach, concentrating on AI data center power, grid storage, solid-state transformers and broader grid modernization applications. We have the expertise to extend into the AI data center, which operates at significantly higher voltages than legacy data centers. As voltages increase, we believe an increasingly larger portion of this addressable market will be better served with silicon carbide technology from grid to rack, compared to legacy silicon-based solutions. Our devices are already embedded in critical AI data center power systems, and we have doubled our data center revenue in the last 3 quarters with 50% quarter-over-quarter growth from Q1 to Q2. Further, we are actively collaborating with a broad ecosystem of partners to support the industry transition from legacy 400-volt architectures to next-generation 800-volt AI platforms.

Data center build-outs and widespread electrification have driven a surge in global energy demand. We’re already seeing silicon carbide adoption across wind and solar applications as evidenced by our recently announced collaboration with Hopewind to advance the next generation of wind power solutions.

Turning to aerospace and defense. We believe there is a growing opportunity due to the tailwinds from defense modernization and electrification, including direct energy platforms. U.S. government has already recognized silicon carbide as strategically significant to national security with both the Department of War and the Department of Energy designating it as a critical material. Additionally, the U.S. government has emphasized the strategic importance of secure domestic semiconductor supply chains for national security applications.”

During the Q&A, Feurle gave more details on what they are working on in the AI data center:

“Today, you’re around 100 kilowatt per rack. That’s moving in 2 years to like 600 kilowatt per rack and into like a megawatt rack in the 2029-2030 time frame. This means you have to figure out how do you power these racks and this is where Wolfspeed can play, stepping that voltage down. And then, as more and more renewables come into the mix, you also need a lot of energy storage systems in between. So, that’s the next portion where we are focused on. Then, you need to get this energy into the data center with transformers, right? And there is a transition happening from traditional transformers to solid-state transformers where silicon carbide is the perfect solution.”

Wolfspeed is currently making around $475 million in annual revenue from power semis and $200 million from SiC wafers. From previous management, we know that the SiC power semi fab has a revenue capacity of around $2 billion, which could result into $1.2 billion of annual FCF. We’re not sure yet whether new management will confirm these numbers, but they provide some guidance on what a bull scenario could look like.

The current market cap is $722 million, and with the net debt of $700 million, this results in an enterprise value of $1.4 billion. So, currently, the shares are valued at an EV-to-Sales of around 2x.

Let’s work out a realistic scenario to see if there is substantial upside. If a cyclical recovery in EVs and industrial end markets doubles revenues, and then the company wins another $500 million in AI data center sockets, it’s well possible that over time we can add another $1 billion in revenues. Assuming a 50% contribution margin, this should result in $500 million of additional FCF. With further cost optimizations, this could well result into $200-300 million of overall positive FCF for the company on an annual basis. Valuing this at 15x FCF would give a market cap of around $3.8 billion.

So, Wolfspeed could be interesting here. There should be a coming boom in power semis driven by AI demand and we’re also at the start of a cyclical recovery in industrial end-markets, going through conference calls from TI, ADI and Microchip. And then if Warsh comes in to cut rates, this should boost automotive demand such as EVs.

We also think that Wolfspeed should see more design wins due to its US-based fab capacity as hyperscalers and GPU makers increasingly secure their supply chains, such as by shifting more advanced logic manufacturing to TSMC Arizona. At the same time, Wolfspeed’s current cash position and cash burn give new management ample runway to turn things around.

The main risk would be a macro downturn in the coming years. Should there be a credit, geopolitical, or other event impacting the global macro, the risk would be that Wolfspeed’s cash burn deteriorates further which will no doubt send the shares quite far south. However, due to the strength of the current AI investment cycle and the demand this is creating for power semis, we do think the risk-reward is towards the upside here for Wolfspeed. Needless to say, this remains a speculative story as the company will need a substantial amount of design wins related to AI data centers to turn things around.

Next, we will dive into opportunities in power semis and the overall semis industry, including a few likely layups for investors. Finally, we’ll discuss our findings on AMD.

Read more

One of the big debates in AI is whether LLMs will be a commodity, or whether these will be attractive businesses where a handful of big companies will dominate like we have in big tech today. The commoditization camp will point to charts like the one from Nvidia below, pointing out that open source models are only six months behind and with the gap in capabilities continuing to narrow:

A few key points though to note here. Whenever we’ve tried open source models, the output we get is of lower quality than the leading models. What is likely going on in the chart above is that open source models are heavily trained to perform strongly on benchmarks, whereas with leading models there is also a strong focus to perform in the real world. This is the classic statistical phenomenon of in-sample versus out of sample performance. Just because a statistical or quant model performs very strongly on in-sample data doesn’t mean it will perform strongly once you test it on real data it hasn’t seen before. For example, in quant investing, a backtested and data-mined performance shouldn’t get confused as a reliable indicator for real world results.

Despite the chart showing a narrowing performance, our view is that real world performance between leading models and the rest of the followers will continue to widen. Before, training was heavily dominated by pre-training, which is the phase where you train the model on the available body of human knowledge. However, over the last twelve months, both reinforcement learning and chain-of-thought reasoning have become crucial to get state-of-the-art problem solving capabilities. With reinforcement learning, models can now be trained on unlimited data in particular fields such as mathematics and coding as the model’s proposed solution to synthetic problems can objectively be verified.

This is similar to DeepMind’s training process of AlphaGo nearly a decade ago, where the model that beat the world champion in the boardgame was not trained on human data, but purely on synthetic data by having the model continuously play Go against itself. So, we believe that scaling laws in current AI training are alive and well, and that the increasing capital intensity will make it hard for new players to enter the field. You either need large revenues or access to large amounts of funding in order to compete in this game.

Going forward, another differentiator is that leading firms will have the capabilities to build high-value data sets themselves. This includes hiring investment bankers to teach models how to do financial modeling, lawyers to set up research workflows for legal cases, scientists for scientific research, etc. As the model arrives at its final answer via chain-of-thought reasoning, competition aiming to distil a leading model over API can only memorize this final result. However, these distilled models will struggle to resolve new real world problems themselves.

A positive for the big tech firms is that they also have access to unique data sets. Good examples here are Meta with Facebook and Instagram, xAI with X, and Microsoft with LinkedIn. While other providers of high quality data sets such as Reddit will likely continue to steeply increase pricing in return for training access. Amazon is also already blocking access to their shopping site for AI agents, but presumably for a hefty fee this could get unlocked as well.

The key point from the above data points is that capital requirements in order to compete in this race will only continue to grow. Historically, these types of markets have not resulted in highly competitive landscapes with a plethora of competition, but rather a handful of competitors at best. While there is a widespread narrative that LLMs are commodities, we don’t believe this is actually the case today and that it will be even less true in the future. OpenAI and Anthropic scaled to $10-20 billion ARRs within a few years—the fastest ever— providing objective evidence that these businesses have clear differentiated value.

The only counterargument to the above we can see is that big Chinese tech firms can legally purchase high-quality training sets via Chinese data exchanges. For example, while US medical data is fragmented across thousands of private hospitals on different systems, Chinese AI firms have access to large scale anonymized medical records giving an obvious competitive advantage in fields such as medicine and biology.

Another advantage is that China has a ton of engineers which could more cheaply assist models to build workflows. A decade ago, China was famous for data labeling farms, where low-skilled workers tagged traffic photos to train autonomous driving models on. Today, companies like DeepSeek and Baichuan use expert-in-the-loop systems where junior engineers and field specialists write chain-of-thought reasoning paths for models to follow. A big drawback for China is that the country remains heavily short on compute. However, in the long term, it’s likely that China will have the advantage in terms of power generation.

Nvidia vs AMD – Apple vs Android?

At CES, Jensen explained how Rubin will give 5x the performance of Blackwell, despite the number of transistors only increasing by 60%:

“This is the Rubin GPU. It’s 5x Blackwell in floating performance. But the important thing is to go to the bottom line, it’s only 1.6 times the number of transistors of Blackwell. That tells you something about the levels of semiconductor physics today. If we don’t do extreme co-design at the level of every single chip across the entire system, 1.6x puts a ceiling on how far performance can go each year. One of the things we did is called NV FP4 tensor core. The transformer engine inside our chip is not just a 4-bit floating-point number, it is an entire processing unit that understands how to dynamically adjust its precision to deal with different levels of the transformer.

So, you can achieve higher throughput wherever it’s possible by losing precision, and go back to the highest possible precision wherever you need to. You can’t do this in software because obviously it’s just running too fast, and so you have to be able to do it adaptively inside the processor. That’s what an NV FP4 is. When somebody says FP4 or FP8, it almost means nothing to us. Because it’s the tensor core structure in all the algorithms that makes it work. I would not be surprised that the industry would like us to make this format and this structure an industry standard in the future. This is completely revolutionary.”

Next, we will dive into Nvidia’s approach and compare this to AMD’s coming solution. Subsequently, we will dive into more topics in semis and software.

Read more

A key reason for the AI supply chain outperforming in ‘25 was token consumption coming in much stronger than expected, which resulted in cloud capex upgrades throughout the year. The chart below highlights how the bulk of capex is being driven by the hyperscalers and especially the major clouds, which is where enterprises run key infrastructure (alongside on-premise data centers).

Despite narratives of a bubble and circular financing, the bulk of investments is carried by hyperscalers who have some of the best balance sheets and cash flow generation in the world. While neoclouds are highly leveraged, these remain small in overall capex spend. However, most of their capex spend is backed by long term contracts and with high quality players such as hyperscalers. In addition, national governments are participating in building out dedicated AI infrastructure, while financing these by tax revenues and government bonds issuance. Again, these are not lower quality participants as they actually have the cash to participate.

If we retrieve quarterly accounts, we can see that the four hyperscalers are actually close to peak FCFs again at $60+ billion per quarter (black line below, axis on the right). Despite elevated levels of capex, the returns on those investments are obviously already healthy. And all these installed AI servers will generate revenues for 6 years, so the real rewards in FCF still have to come as we’re still in the investment phase.

Hyperscaler cash flows of operations have been growing at attractive CAGRs since the start of this investment cycle:

If we take into account rising capex via depreciation, EBIT has been growing at a 26% CAGR over the last five quarters:

And the outlook continues to be extremely good. Cloud revenue growth has already been massively accelerating throughout 2025, while the key leading indicator of cloud bookings accelerated again to 54% YoY growth last quarter:

Concerns around large capex spend have been weighing on hyperscaler share prices, especially Meta. However, the financial returns are clearly there already, while we continue to be confident in the demand picture. We expect AI demand growth to remain strong as inference applications get increasingly inserted in the real economy and model reasoning capabilities continue to increase. The latter is becoming increasingly token intensive, with the ratio of the number of tokens generated in reasoning relative to the number of tokens in the final output being on an upwards curve. We’re moving away from an old world, where there was a 1:1 relationship, to a reasoning world where models consume huge amounts of internal tokens to solve complex problems such as in coding, science, etc. At the same time, advancements in AI agents mean that AI models can now work independently for longer periods on their own.

Overall, we see the hyperscalers as building out extremely valuable infrastructure, where enterprises are effectively locked in at the major clouds due to high switching costs in software services and data gravity effects.

Another fallacy we regularly encounter is the so called ‘circular financing’ narrative, i.e. that this capex cycle is being driven by circular deals. While there is some amount of circular financing, most notably in some Nvidia-neocloud deals or in the OpenAI-AMD deal, these remain a very low proportion of overall capex spend. Circular financing doesn’t even make 5% of Nvidia revenues on our estimates.

The main risk for the AI hardware names is that forecasts at some stage will start showing moderate growth and with the risk of a cyclical pullback. Going over the financials above, we think that the economics are very healthy for the AI bull cycle to continue. Even when capex growth rates start to moderate later this decade, for the major clouds, share prices can actually do extremely well as free cash flow growth accelerates (as we’re post the investment cycle). At the same time, more moderate capex growth still translates into revenue growth for the hyperscalers, albeit at a slower pace. Another factor is that pricing power for cloud software services remains strong, which will further boost margins at the major clouds.

AI hardware names on the contrary need ever more data center capacity being added to show attractive growth rates. A tailwind is that as AI data center equipment is becoming ever more advanced, ASPs will automatically increase. So, AI hardware names can still get growing revenues even by selling the same volumes of equipment, but at higher ASPs. Key cyclical indicators to watch out for in the coming years will be token consumption growth in the cloud, GPU rental pricing and availability, and capex guidance at the hyperscalers. We’ve seen quite a number of semi booms and busts over the last few decades, and so overcapacity at some stage later this decade is a key risk factor to watch out for. Once you’re in the overcapacity scenario, it will likely mean a multi-year downturn in AI equipment sales.

Another risk is that while AI demand should continue to come in strongly, we simply run into bottlenecks in the AI value chain somewhere, which reduces growth rates for the overall industry. To some extent this has already been happening since the start of this boom. First, in CoWoS, then in HBM, and now in Optics.

Overall, we see the environment as very healthy for AI. We expect token consumption to keep growing at high rates while the hyperscalers are reaping the rewards to pay for increasing investments. This brings us to where these investments will be allocated..

Next, we will dive into the entire semis space as well as key topics in AI. As a bonus, we’ll discuss our outlook for the quantum computing space, and we’ll also review a US-based, state-of-the-art fab trading on a sizeable discount to intrinsic value. Finally, we’ll discuss our top stock picks for 2026 across tech.

The Battle for the Accelerator

Read more

“We revisit why high NA is likely to be adopted and update the latest evidence/data supporting its adoption: Why? 1) new data suggest high NA could lead to 20-40% cost saving vs double or even triple patterning critical layers; 2) We identify 3 specific critical layers for logic likely to adopt high NA and 2 for memory; 3) Advanced logic roadmap confirms that the below 20nm metal pitch at A10 node (tools in 2029-30, mass production 2031) will likely require triple patterning low NA if high NA is not adopted; When? 4) latest data suggests tools availability likely to reach >90% in ‘26 which we believe is likely to trigger high NA orders in H1/H2 26 for insertion for A14 2nd gen (tools in 2027-28, mass production 29).

Our analysis suggests that High NA could account for 15-20% of system sales by the end of the decade, driving a 3-5 percentage point increase in lithography intensity share due to cost reductions achieved in critical layers of other manufacturing processes, such as deposition and etching. As for catalysts, we view the SPIE industry event scheduled for February 22-26 as a key date to monitor progress toward achieving >90% readiness for High NA, which may pave the way for its adoption in 2027 for the A14 2nd/3rd generation node. Furthermore, we believe the introduction of High NA is significantly more impactful for ASML compared to low NA multi-patterning, as High NA results in a greater lithography share, whereas multi-patterning leads to a lower lithography share. This dynamic could meaningfully influence ASML’s valuation premium relative to its peers.

ASML is trading at a c20% premium relative to U.S. peers, compared to its 10y historical average of 86% with positioning near the all-time low range of 5-15%. Given High NA could add 3 to 5 percentage point WFE share relative to US peers we believe high NA is not only important for EPS but could trigger further upside to relative premium multiple to a more normalised level vs history of around 40-60%.”

The other reason we like ASML is that TSMC capex is still fairly depressed. TSMC built up a lot of N7 capacity during the covid semi boom and has been able to shift those tools to strengthen its N5 and N3 lines during the AI boom. However, this tailwind in terms of lower required capex should now be dying down, and with all capacity TSMC is planning, it looks likely that we’re looking at a bull case for semicap. Even modeling in moderate capex-to-revenues levels for the coming years, capex comes out much higher than what the market is expecting:

On Micron’s earnings call this week, it became clear that AI demand continues to strengthen:

“Server unit demand has strengthened significantly, and we now expect calendar 2025, server unit growth in the high teens percentage range, higher than our last earnings call outlook of 10%. We expect server demand strength to continue in 2026.

In the medium term, we are only able to meet about 50% to 2/3 of our demand from several key customers. So we remain extremely focused on trying to increase the supply here and making the necessary investments.”

Obviously this bodes well when it comes to litho demand from the memory market. Especially as the key driver of memory demand, HBM, is extremely capex intensive. This is again Micron’s CEO:

“The dramatic increase in HBM demand is further challenging the supply environment due to the 3 to 1 trade ratio with DDR5 and this trade ratio only increases with future generations of HBM. Additional cleanroom space is necessary to address this increased demand and lead times for cleanroom build-out are lengthening across geographies.”

We wouldn’t worry about China’s Manhattan project to produce EUV tools at this stage. Even if China produces the tools by the end of this decade, it seems unlikely that the US will allow any of the semi manufacturers amongst its allies to purchase that equipment; so that includes TSMC, Samsung, SK Hynix, and obviously Micron and Intel. Similarly, Western enterprises are focused on reducing Chinese components in their supply chains and this trend will only continue to strengthen with the recent Nexperia debacle.

So, even if Chinese EUV tools will see adoption by 2030 or so, which is a big if, we see the risk as fairly limited to ASML at this stage. ASML can’t sell their EUV tools into China anymore anyways, and so Chinese EUV tools will mainly be sold to SMIC and the Chinese memory players. Where it would impact ASML is on their Chinese immersion business, as the likes of SMIC have been purchasing plenty of those to do multiple patterning and still produce fairly advanced silicon of around N7 levels.

ASML’s valuation still remains fairly reasonable, while we should be looking at a bull case in demand in the coming years:

Next, we will dive into market share movements in XPUs vs GPUs, and Broadcom’s sell off.

XPUs vs GPUs

Read more

The beauty of coding for CPUs is that it is extremely simple, you can leverage languages such as C, Fortran, Rust, etc., and even make use of multi-threading to parallelize tasks over multiple CPU cores where possible. However, to move these workloads onto GPUs, this requires a complete algorithmic and codebase overhaul. GPUs demand massive thread-level parallelism with tens of thousands of threads, and a codebase that has to be rewritten for CUDA. This code has been built up over decades, and so rewriting it is an enormous task. This is precisely the reason why it has not been done before despite GPUs being much faster.

What will no doubt help here is that LLMs have gotten extremely good at coding, and so they can now take over large parts of this workload. Basically, we will have GPUs creating code to run on GPUs, creating a virtuous circle of GPU demand.

This is Synopsys’ CEO on how verification is currently a major compute bottleneck in chip design:

“We started redesigning some of our products about 7 years ago on NVIDIA GPU using the CUDA layer and in a number of cases, we’ve seen a significant speed up. When you talk about 10x, 15x, 20x speed up on a work that may take 2, 3 weeks and bring it down to hours; customers will adopt it. The bottleneck of designing these complex chips or complex systems is limited by the computation requirement to verify these systems. The worst thing you can do is assume that you’re ready to go, launch a chip and it doesn’t work as intended. That costs hundreds of millions of dollars. So, you spend a lot of energy in the design and simulation phase. We have a number of products already in use at customers, still very early stages in terms of a broad adoption, but that’s why we refer to an expanded opportunity for both companies.”

Jensen highlighted how EDA and CAD software will play a major role in digital copies of the physical world, i.e. digital twins:

“The expansion of the market opportunity for Synopsys goes from the EDA industry – addressing a several hundred billion dollar chip industry – to now addressing a multitrillion dollar, every product industry. In the future, every product will be designed in digital twins. This partnership essentially enables this industry to address the entire R&D budget of the whole world’s GDP. Everything that’s going to get designed and built will be done first in a digital twin.

In 2016, 90% of the world’s scientific supercomputers running physical and biological simulations were general purpose CPUs. Today, it has flipped completely, CPU-only supercomputers are 10%. NVIDIA-accelerated computing is now 90% of the world’s physical science simulation computers. This is going to happen also in the EDA industry. For industries that are fundamentally based on physics, this capability is really important.

The challenge, of course, is to reformulate the algorithms so that it could be accelerated on CUDA. And that is a multiyear journey. It took some 7 years probably to do cuLitho. Reformulating Synopsys’ applications to take advantage of this acceleration is what this is all about. All of it will be CUDA accelerated over time. And so we’ve got a lot of work to do, but that’s what this partnership is really about, focusing the two engineering teams across the entire companies so that we could take this capability to market as soon as possible.

The partnership is nonexclusive. There’s no obligations whatsoever for Synopsys to only buy NVIDIA. And they’re welcome to continue to work with their rich ecosystem of chip partners, and we’re going to continue to work with our ecosystem of really important EDA and CAD partners like Cadence, Siemens and Dassault. We also partner with Intel, all of NVIDIA’s EDA – the way we do chip design and system engineering today – is still largely based on x86 CPUs. This is really the beginning of that platform shift so that in the future, that’s going to be accelerated by NVIDIA GPUs. And I’d be delighted for all of the chip industry to buy NVIDIA GPUs for designing their chips.”

This is obviously a bit of a sideshow for Nvidia as GPU-based chip design will be a small end-market compared to AI. However, it is of key strategic importance to accelerate their own chip design cycle. Nvidia has recently moved to a one year cadence in releasing a new major chip, and no doubt removing the verification bottleneck will have a major positive impact once this work is complete.

Clearly this is an evolution that can have major impacts on the chip design industry overall. Most notably, it will bring down costs as weeks of chip simulation can be reduced to a single day or less. This opens the door for more chip experimentation and also more custom designs to be done, even by less experienced players who have the capital (e.g. large enterprises), as you can iterate on your chip designs quickly and remove bugs via verification.

The winners from this should be the EDA companies, Synopsys and Cadence, as lower chip design costs should boost demand for chip design software from more enterprises. Additionally, Synopsys and Cadence have solid pricing power. When they can provide a tool that speeds up work by 20x, they will have the ability to raise prices for GPU-accelerated modules. While Nvidia’s investment could give Synopsys an edge here, also Cadence is pursuing GPU accelerated workloads and has historically been able to build out a solid simulation portfolio.

Marvell’s New Data Center Tech

Marvell made a new acquisition to increase the company’s focus towards the data center and AI in particular. This is a strategy that has been working well, with the company’s revenues growing 41% year-over-year this quarter (excluding the recently divested automotive ethernet business to Infineon), and the data center end-market now comprising 74% of revenues. This is Marvell’s CEO on the tech and team they’ll be acquiring, which sounds pretty interesting:

“I’m excited to share details on the strategic acquisition we announced today of Celestial AI, which brings an entirely new disruptive technology, a photonic fabric platform purpose-built for next-generation scale-up interconnect. This acquisition is the latest in a series of decisive moves to further strengthen our data center portfolio. Since 2019, we have continued to increase our focus on the data center, divesting our WiFi business and acquiring Avera, Aquantia, Inphi and Innovium. These transactions have driven significant revenue growth and scale, and have each proven to be an absolute home run. The acquisition is expected to close in the first quarter of next year, subject to regulatory reviews in the United States.

AI is reshaping data center architecture at an unprecedented speed. Next-generation accelerated systems are no longer confined to a single rack, they are evolving into multi-rack scale-up fabrics that connect hundreds of XPUs in a high-bandwidth and ultra-low latency any-to-any fashion. These advanced fabrics demand purpose-built switches and interconnects, creating a new TAM for companies like Marvell. Industry analysts are forecasting the merchant portion of the scale-up switch market to approach $6 billion in revenue in 2030. On the interconnect side, as it attaches to both the XPU and the switch, the opportunity actually doubles to over $10 billion. These are both very large and incremental opportunities for Marvell.

As we first evaluated Celestial AI, it reminded us of our early look at Inphi and the products we saw in their PAM technology to transform the scale-out interconnect market. We see even greater potential for Celestial AI’s photonic fabric to transform scale up interconnect. Interconnect technology is critical to couple hundreds of XPUs together. Copper-based interconnects used in today’s scale-up systems are approaching their fundamental limits in reach and bandwidth, creating a compelling need for optical solutions. Celestial AI’s photonic fabric technology was purpose-built for this inflection. It enables large AI clusters that scale both within and across racks using a high bandwidth, low latency, low power and cost-effective optical fabric. This breakthrough enables a true optical solution with greater than 2x the power efficiency of copper interconnects, but with far longer reach and significantly higher bandwidth.

Celestial AI’s solution provides nanosecond class latency and the thermal stability is a significant competitive differentiator. It enables reliable operation in the extreme thermal environments created by large multi-kilowatt XPUs. This allows the photonics technology to be co-packaged vertically with high-power XPUs and switches in a 3D package, enabling the photonic connection to be made directly into the XPU rather than from the edge of the die. This stands in sharp contrast to many other CPO implementations where the photonics engine must connect at the die edge. Celestial AI’s approach results in a more compact and integrated solution, freeing up highly valuable die edge beachfront that can be repurposed to significantly increase the amount of HBM within the XPU package.

Celestial AI’s first-generation product is a photonic fabric chiplet (PF chiplet), which integrates all the required electrical and optical components. This delivers an unprecedented 16 terabits per second of bandwidth in a single chiplet, 10x the capacity of today’s state-of-the-art 1.6T ports used in scale-out applications. Celestial AI is deeply engaged with multiple hyperscalers and ecosystem partners and has already secured a major design win with one of the world’s largest hyperscalers who plans to use PF chiplets in its next-generation scale-up architecture. These PF chiplets will be co-packaged into both the hyperscaler’s custom XPUs and scale-up switches to provide connectivity.

This is just the beginning of a broad set of new applications which can be enabled from this technology. Our base case forecast shows Celestial AI’s revenue reaching a $500 million annualized run rate in the fourth quarter of fiscal 2028, doubling to a $1 billion run rate by the fourth quarter of fiscal 2029.”

Marvell’s fiscal ‘28 ends in January of ‘28, so significant revenues will start coming in two years from now (Q4 of calendar ‘27). The main goal of this new tech is to scale up (providing low latencies) across multiple racks. Traditionally, when you add racks to your compute, this is called scaling out (or horizontal scaling). However, Marvell-Celestial AI are calling this scale up here – a term traditionally used to refer to adding more compute into a single server or rack – as due to low latencies, they could make a cluster of 5-10 racks behave like one rack.

Current scale-up fabrics (like NVIDIA’s NVLink or Credo’s AECs) rely on copper, which confines a computing node to a single rack (e.g. 72 GPUs). To grow larger, you currently have to step down to slower scale-out networks such as Ethernet or InfiniBand, which forms a bottleneck especially in AI training today. During the Q&A, it became clear that Amazon AWS will be the first customer, and so this tech will first be implemented for Trainium-based AI training clusters. As shipments start in Q4 2027, this will be either for Trainium v3 or v4 based training clusters.

With this acquisition, Marvell is building out their data center connectivity portfolio with what looks like a very promising next-gen technology – high bandwidth, low latency, long reach, and thermal stability. Marvell has a current stronghold in DSP-based connectivity (800G and 1.6T pluggables), which are the go-to for hyperscalers to wire up AI clusters. The company acquired this expertise with the Inphi deal in ‘21, which was obviously a great investment and it is currently the key driver of growth.

So, the company has a fairly dominant market share in DSPs, but the main concern would be increasing competition from Broadcom which is launching competitive products. However, end-market strength is expected to offset headwinds from competitive pressures. This is HSBC on these dynamics:

“We believe Marvell is maintaining its market leadership in this segment through its 800G PAM4 DSP, which is seeing strong demand, and the launch of the subsequent 200 gig per lane 1.6T PAM4 solution, for which the company anticipates accelerating adoption over the next few quarters. However, we believe its market leadership may be threatened by Broadcom, which has caught up with the launch of its own 800G and 1.6T DSP solutions. As a result, while we do believe that the growth of its AI optical business, which is primarily driven by PAM4 DSP solutions, will decelerate — we believe the growth would still be higher than Street estimates due to the overall TAM expansion.

Amidst the massive AI infra demand expanding beyond traditional hyperscalers and into enterprise, sovereign AI, etc., the demand for DSP solutions continues to rise. Hence, while we expect AI optical business y-o-y growth to decelerate from 54% in FY26e to 38% in FY27e and 20% in FY28e, our FY27e and FY28e estimates remain above consensus estimates of 20% y-o-y growth in FY27e and 3% y-o-y contraction in FY28e. Our bullish estimates are based on our belief that while 800G will continue seeing strong momentum, the 1.6T solution will also see faster adoption driven by the TAM expansion.”

Next, we’ll dive further into Marvell. Will Marvell get disrupted? Is it time to take profits after the strong share price rally since September? We’ll provide much more analysis below..

Read more

Secondly, when it comes to running workloads in the cloud, Nadella described Azure as the “customer’s portal to the entire AI stack”. Microsoft’s strategy is to provide high margin web services in their cloud—like storage, databases, networking, security, and further apps—while they don’t need to own the raw GPU for every job. Nadella explains:

“Azure is the portal... For pure GPU workloads—like massive training runs or inference at scale—we can burst to neoclouds. It’s fine for us because now we have line of sight to demand. If they bring their capacity into our marketplace, that customer uses the neocloud for compute, but pulls storage, databases, all the rest from Azure. That’s a great win-win. They get specialized capacity fast, and we capture the full workflow.”

We see this as a sound strategy. Azure has a wide portfolio of web services in the cloud with its only competitors being Amazon AWS and Google Cloud Platform. At this stage, as future AI growth rates remain hard to predict, Microsoft can leverage capacity from a variety of neoclouds, control their capex expansion and remain free cash flow positive. The more attractive and sustainable AI workloads can selectively be powered by the Azure platform, with excess capacity being outsourced to neoclouds. In the future, more of those types of workloads can always be brought in-house again, with bursty AI demand being offloaded to neoclouds.

Thirdly, while OpenAI’s apps can be powered by other clouds, the APIs will be exclusively on Azure. This is an important point as the latter part of the business will likely be very lucrative – loads of other software and apps connect to OpenAI’s APIs to send their ChatGPT queries over JSON. If OpenAI’s consumer-facing app at some stage can disrupt Google Search, obviously it could be even more lucrative, but Nadella is happy for those workloads to also be powered by other clouds. The reason is likely that Microsoft doesn’t want to send their capex levels to stratospheric heights at this stage as future AI growth rates remain hard to predict. So, Nadella is striking a balance here, Azure can still accelerate growth and achieve highly attractive 30-40% growth rates, while the company focuses on selectively expanding capacity where there should be long term sustainable demand.

Fourthly, Nadella explicitly confirmed that Microsoft has full access to all of OpenAI’s IP, with one notable exception in consumer hardware i.e. any future OpenAI-branded devices. While no doubt Sam Altman will try to shield some of OpenAI’s innovations in algorithms to train and infer their models, Nadella implied here that Microsoft would have access to everything. This means it’s fairly likely that Microsoft will be able as well to build their own LLM capabilities in the long term, which Nadella referred to as MAI (Microsoft AI).

Finally, Nadella envisions their software products such as Office 365 and their Azure platform being utilized by AI agents. So, the business model should evolve from per-user licenses to per-agent provisioning which could balloon the TAM:

“Our business, which today is an end-user tools business, will become essentially an infrastructure business in support of agents doing work... It’s not per-user anymore, it’s per-agent.”

The only counter argument we’d make here is that a lot of these software products’ key functionality is also available via open-source competitors. While these open-source apps are typically lower quality, thanks to AI their code bases could be improved to increase functionality and reliability. Thus, while we can see a bull case for software as also agents might need to buy licenses or usage credits, there is also increased risk as writing code has become more commoditized. Doing coding on a regular basis, we estimate that one programmer’s productivity currently increases by 5 to 10x with the use of LLMs. And this will only continue to increase as now LLM agents are directly available in popular coding environments, so it’s not like you have to copy-paste code anymore from Claude’s web portal.

Subscribe now

For subscribers, we will dive further into developments in AI. We will discuss CoreWeave, as well as some accounting tricks the company is playing. Next, we will give the outlook for AMD and, finally, we will highlight a smaller, hardcore engineering firm in robotic AI that we came across. We’ll also give an overview of our current buy list in the universe of stocks that we track.

Read more

As the industry moved into this oligopoly structure, pricing discipline improved which resulted in better margins and EPS:

However, as the industry remained extremely cyclical, these stocks weren’t Buffett-style buy-and-hold investments, but rather tactical opportunities to play during an upcycle in the semis market. End-markets also weren’t that exciting. Revenue exposure to the mobile and client markets was high, two end markets that had gone ex-growth:

However, with the rise of AI, end market exposure for the memory players has dramatically shifted towards the data center. Last quarter, the two data center related segments comprised 54% of revenues for Micron. Besides being one of the most attractive growth markets in semis in the coming five-ten years, the data center market also has a number of other attractions. Firstly, customers are taking out multi-year contracts, which enhances revenue visibility and stability for the memory players. This is SK Hynix’s CFO on this week’s call:

“It is true that in the memory market, some businesses have shifted to order first, produce later approach with the emergence of HBM which have massive investment and long lead time. In addition, with strong HBM demand coming from our customers, the company was able to secure visibility into the customers’ demand from the contracting stage with long-term agreements. For both the memory industry and the company, this has led to greater market predictability and business stability than in the past when it was much more volatile.”

Subscribe now

One of the historical problems for the memory players was that memory was basically a commodity. You could swap out one vendor’s DRAM and replace it with that from another. This is now changing in the crucial HBM market where growth is exploding. SK Hynix’s CFO explains:

“Custom HBM will gradually increase from HBM4E. So products will be developed in close collaboration with customers from the early stage of design of GPU or ASIC products, unlike in the existing standardized HBM. This will lead to much longer term and strategic transactions between customers and a small number of suppliers, contributing even more to business stability and profitability improvement on the part of the memory suppliers.”

All of the above has led to better dynamics in traditional memory markets as well. Industry consolidation led to better pricing and investment discipline, and the growth in HBM is resulting in supply constrains and long term agreements in conventional memory markets:

“Now if I may add, memory companies are allocating capacity to ramp up HBM supply. And this has led to supply constraints in conventional memory, resulting in supply shortage of conventional memory, for which demand is actually growing. And as a result, we are seeing an increase in customers who want to sign long-term agreements for conventional memory products as well. Some customers are very much actively responding to the current supply shortage by issuing prepurchase POs for 2026. Now, given the customers’ demand and the company’s capacity for next year, not only HBM, but DRAM and NAND capacity has essentially been sold out.”

For premium subscribers, we will dive further into current developments in the memory market, we’ll do some interesting modelling on Nvidia, Broadcom and AMD, discuss the outlook from here, and finally, we’ll highlight a few trades which we will be making (selling some winners).

Read more

JP Morgan initiated on Credo, a niche silicon player in the AI data center:

“Credo is well positioned to capitalize on rising AEC demand, with industry unit shipments forecasted to expand at a 60%+ CAGR from <1 mn in 2022 to 10 mn+ in 2028, supported by increasing deployments from major companies such as Amazon, Microsoft, and xAI as well as broadening adoption by others, including Meta and potentially Oracle, Google, and more, which are currently engaging with the technology. Underlying the increasing and broadening adoption are expanding use cases beyond the traditional NIC-to-ToR connection inside a server rack, as AECs are now being used for short-reach rack-to-rack connections, intra-switch rack connections, and eventually scale-up use cases. These dynamics are creating an ASP tailwind, as early adoption of short-reach 100G “straight” cables by a single company ( ASPs in the low hundreds ) is transitioning to a variety of ranges, more complex form factors, and data rates extending to 800G for multiple companies ( ASPs in the mid-hundreds ), which, when combined with the shipment forecast, translates into a $4 bn+ market opportunity by 2028

Credo became the first supplier of AECs in 2018, creating both the product and the market, and remained the primary supplier for many years. Despite the more recent entry of well-known players such as Marvell, Broadcom, and particularly Astera – which has gained notable traction in the market recently – Credo still commands an overwhelming majority of the share, estimated to be ~80%. This leadership is driven by Credo’s ability to deliver an entire, qualified, and tested solution, where the silicon to finished cable is based on its in-house IP (vs. peers providing only the silicon or module), which gives Credo advantages in time-to-market, ease of deployment, reliability, and cost. While we expect other peers to gain market share over time as large customers multi-source, we anticipate that Credo will maintain its overall leadership, with other suppliers primarily serving as second or third suppliers.

Beyond Ethernet AECs, Credo’s product portfolio also encompasses other interconnect technologies, such as optical DSPs and PCIe retimers. While Credo currently has limited share and will be competing against established peers, these markets represent significant opportunities – estimated to track to $6 bn and $2 bn in 2028, respectively – and even minor share gains could translate into a solid growth opportunity for the company. Importantly, the aforementioned implies an aggregate interconnect TAM expanding at a +45% CAGR and reaching more than $12 bn in 2028 .

We are forecasting a revenue CAGR of 50%+ from FY25 to FY28E, including growth of 100%+ in FY26E. This forecast is underpinned by increased deployments of AECs by Credo’s two large, more longstanding customers, new deployments across three additional customers, as well as strong growth for Optical DSPs and solid growth currently assumed for other solutions ( both from a low base ). Importantly, our expectations are ahead of industry forecasts for AEC unit shipments, which are projected to grow at a 40%+ CAGR over the same period. We believe this outperformance reflects new use cases beyond NIC-to-ToR (e.g., rack-to-rack, switch rack), the ramp-up of new customers, and pricing tailwinds.

We are projecting earnings growth to outpace revenue, with an earnings CAGR of 70%+ from FY25 to FY28E, driven by gross margins tracking toward the high end of the long- term range (63%-65%), supported by increasing scale and an improving product mix as well as operating expense growth at a pace below revenue. These factors should also drive a solid improvement in free cash flow generation, which, against the backdrop of a very healthy balance sheet, could lead to a greater appetite for capital allocation that has been limited to date and would represent upside to our model. We introduce a December 2026 price target of $165 based on valuing shares of Credo at ~48x our CY27 EPS estimate.”

The below image shows what Credo does – the purple cables are Active Electrical Cables (AECs). These are low-power, copper-based interconnects that integrate advanced silicon to extend signal reach. These are heavily used in AI data center networking for shorter distances with the most common use cases being inter-rack and top-of-rack communication.

The attraction with Credo is an 80%+ market share in the AEC market where volume growth is even higher than in the accelerator market. While competitors such as Marvell, Broadcom and Astera are moving in, Credo has further growth opportunities in other interconnect technologies.

Subscribe now

There are not many companies which can effectively compete at the leading edge as semi design costs continue to explode. So, typically, there are a handful of competitors in these markets, or less. This makes these industries and companies attractive for investors to play the exploding growth in leading edge silicon, as these are not commoditized businesses. This can easily be witnessed in attractive gross margins. You can see that both Astera and Credo are generating similar gross margins to Nvidia:

The attraction with both TSMC and ASML is that these upstream companies in the value chain currently face no or only very limited competition, which gives them clear upside to keep moving gross margins up over time. We expect that TSMC will be moving its gross margins up to 65-70% long term, and similarly, it’s likely that ASML will move towards a 60% gross margin as well. Obviously, Astera and Credo have a handful of competitors, which could give some gross margin pressure over time, especially when the pace of growth slows down at some stage. That said, we think these latter businesses are definitely investment grade as well in the context of a diversified portfolio.

For investors in the name, we’d simply stay long as we continue to be bullish on the AI cycle, while Credo is obviously fairly well positioned.

It’s a punchy multiple, but then again EPS are exploding:

For premium subscribers, we’ll dive further into developments in AI with a number of stock picks. Finally, we’ll take a detailed look at a very small but highly innovative company that should be looking at a long runway of attractive growth.

Read more

“The AI demand actually continues to be very strong, it’s more stronger than we thought 3 months ago. So, we have talked to customers and then we talk to customers’ customer. The CAGR we previously announced is about mid-40s. It’s still a little bit better than that. We will update you probably in beginning of next year, so we have a more clear picture. Today, the numbers are insane.

Talking about the CoWoS capacity, we are working very hard to narrow the gap between the demand and supply. We are still working to increase the capacity in 2026. The real number, we probably update you next year. Today, all I want to say about the everything AI-related, like front-end and back-end capacity, is very tight. We are working very hard to make sure that the gap will be narrow. Right now, we are working with one OSAT, a big company and our good partner, and they are going to build their fab in Arizona. We are working with them because they’re already breaking ground, and the schedule is earlier than TSMC’s two advanced packaging fabs. Our main purpose is to support our customer in the U.S.

If the China market is not available, I still think the AI’s growth will be very dramatic, very positive. I have confidence in my customers, both in GPU or in ASIC, they are all performing well. And so, they will continue to grow, and we will support them. I believe we are just in the early stage of the AI application. So, very hard to make the right forecast at this moment. We pay a lot of attention to our customers’ customer. We discuss with them and look at their application, be it in the search engine or in social media, and then we make a judgment about how AI is going to grow. This is quite different as compared with before, where we only talked to our customers and have an internal study.

The number of tokens increase is exponential and I believe that almost every 3 months, it will exponentially increase. And that’s why we are still very comfortable that the demand on leading edge semiconductor is real. The tokens increase is much, much higher than TSMC’s CAGR we forecasted. Our technology continues to improve and so our customers moving from one node to the next node, they can handle much more tokens. And that’s why we say that we have about a 40-45% CAGR.”

Obviously, a big factor why token growth is much higher than TSMC’s forecasted CAGR is all the algorithmic improvements LLM providers are realizing:

Along with their bullish outlook on AI, TSMC spent a lot of time talking about investment discipline. Below is CC Wei emphasizing once again the company’s discipline, although at the same time he mentions that they are securing a second large piece of land in Arizona to build out a giga fab cluster:

“As the world’s most reliable and effective capacity provider, we will continue to work closely with our customers to invest in leading edge specialty and advanced packaging technologies to support their growth. We will also remain disciplined and thorough in our capacity planning approach to ensure we deliver profitable growth for our shareholders. With a strong collaboration and support from our leading U.S. customers and the U.S. federal, state and city governments, we continue to speed up our capacity expansion in Arizona. We are making tangible progress and executing well to our plan. In addition, we are preparing to upgrade our technologies faster to a more advanced process technologies in Arizona, given the strong AI-related demand from our customers.

Furthermore, we are close to securing a second large piece of land nearby to support our current expansion plans and provide more flexibility in response to the very strong multiyear AI-related demand. Our plan will enable TSMC to scale up through an independent giga fab cluster in Arizona to support the needs of our leading-edge customers in smartphone, AI and HPC applications. In Taiwan, with support from the Taiwan government, we are preparing for multiple phases of 2-nanometer fab in both Hsinchu and Kaohsiung Science Parks. We will continue to invest in leading edge and advanced packaging facilities in Taiwan over the next several years.”

It’s clear that demand for N2 is going to be massive. Historically, leading edge semi demand was driven by Apple’s latest iPhone and all other advanced semi designers – such as Nvidia – were one or two nodes behind Apple. However, due to the explosion in AI demand, it now makes sense for the various players competing in AI acceleration to make the additional investments and move to the leading edge node more swiftly. This will obviously be a big positive for TSMC, as leading edge wafers come at much higher ASPs while also a faster ramp of their latest node boosts margins. This is TSMC’s CFO on how N2 will be very profitable and also margin dilution due to a new node ramp is becoming much less of an issue:

“Sunny, let me share with you, N2’s structural profitability is better than the N3. Secondly, it’s less meaningful nowadays to talk about how long it will take for a new node to reach to a corporate average in terms of profitability. And that’s because the corporate profitability, the corporate gross margin moves and generally, it has been moving upwards. So less meaningful to talk about that, okay?”

CC Wei commented how TSMC will be ramping N2 fast due to high-performance computing (AI) demand:

“Finally, let me talk about our end-to-end A16 status. Our 2-nanometer and A16 technologies lead the industry in addressing sizable demand for energy-efficient computing, and almost all innovators are working with TSMC. N2 is well on track for volume production later this quarter. With full year, we expect a faster ramp in 2026, fueled by both smartphone and HPC AI applications. With our strategy of continuous enhancement, we also introduced N2P as an extension of our N2 family. N2P feature further performance and power benefits on top of N2 and volume production is scheduled for second half ‘26. We also introduced A16 featuring our best-in-class Super Power Rail or SPR. A16 is best suited for specific HPC product with compressed signal routes and dense power delivery networks. Volume production is on track for second half ‘26. We believe N2, N2P, A16 and its derivatives will propel our N2 family to be another large and long-lasting node for TSMC.”

The below chart shows how the various nodes have long life times for TSMC, while more advanced nodes really bring in the revenue growth:

As advanced capacity comes at higher ASPs, TSMC benefits both from wafer volume growth and increased pricing over time:

Looking at recent quarters, we can see that in the June and September quarters, wafer shipments are now picking up strongly:

For premium subscribers, we will dive further into current developments in leading edge semis, TSMC, and ASML; and which name of the two should be the best performer in the coming years.

Read more

While TSMC didn’t introduce scaling at the N2 node as they moved to the new GAA transistor architecture (nanosheets in the image below), we should see scaling again at the next two nodes. This is a positive for ASML as transistor scaling requires more litho processing steps – which either need to be carried out with multiple patterning or with more advanced litho equipment (high-NA). In multiple patterning, as the name suggests, the patterning steps are carried out multiple times to transfer the needed chip patterns onto the wafer. This requires multiple processing steps with ASML’s litho tools, currently EUV or immersion DUV. The below chart is from the leading research house on transistor technology, IMEC, which basically all the key players in advanced semi manufacturing are collaborating with including ASML and TSMC:

Note the metal pitch scaling on the chart above as the key driver for litho intensity. TSMC tends to pause when they move to a new transistor architecture, such as Nanosheet (GAA) this year or CFET in the future (2031E). The reason is that they want to make gradual steps to refine their processes and not implement too many changes at the same time, which can make it hard to optimize yields.

On the second bear thesis, the positive is that high-NA is much more economic than EUV at its early introduction. EUV’s uptime and productivity were extremely low. The technology had all sorts of problems and there were continuously bear theses flying around in the years leading up that the technology will never make it. The positive for ASML though is that even if high-NA gets delayed, more multiple patterning with EUV tools will be required at the most advanced processing steps. This actually increases both revenues and margins for ASML, as high-NA is lower margin than EUV due to current low manufacturing capacity levels.

Subscribe now

This is what we concluded back in July as ASML’s shares sold off following results and weak order numbers:

“So, the long term bull case for ASML is that: 1) TSMC at its next one or two nodes will start scaling again, leading to increased EUV demand as more EUV steps will be needed again. 2) TSMC is building out three fab clusters—one in the US and two in Taiwan, where capacity requirements are increasing due to the boom in AI demand. 3) DRAM is becoming more EUV intensive, due to the scaling roadmap. AI is very DRAM intensive (HBM), boosting needed fab capacity. 4) ASML has zero competition. Thus, for long term investors, we think that the risk-reward is starting to look attractive here for ASML. The valuation is basically the same as where it was ten years ago, while a lot of tech and quality stocks in the S&P 500 have strongly re-rated (i.e. trading currently on much higher PE ratios).”

The recent environment reminded us a lot of the 2015-2016 years for ASML with bear arguments flying around such as “EUV isn’t going to work”, “Everyone already owns this stock – how can you still make money?” and “Tech is in a bubble, sell the space!”. Sounds familiar?

JP Morgan was the latest to come out with a bullish note recently:

“TSMC has started to ramp up its 2nm process and this process ramp will continue in ’26 when their lead customer, Apple, is likely to start using this process. There have also been press reports yesterday that Nvidia is expected to be the first to adopt the upcoming A16 process at TMSC which will implement backside power. If these reports are true, then TSMC would need to ramp up significant capacity for A16 in ’27 (when it is likely to be in production). A16 will need 1 or 2 more EUV exposures than 2nm and thus will be particularly positive for ASML because of the increase in lithography intensity but more importantly because of the volume of wafer capacity that would need to be added if Nvidia were to start using the leading edge process (Nvidia has been using a process that is 1 or 2 generations behind the leading edge on its leading edge processors till now). TSMC would need to start preparing for Nvidia ramp in terms of orders and building capacity in 2H26 which would result in stronger orders/sales for ASML from TSMC in both ’26 but definitely in ’27.

HBM-related demand remains robust and memory pricing has not shown weakness. Indications have been that Samsung is possibly closer to qualification for HBM4 at Nvidia as well, which would be very positive for equipment shipments starting end of ’26 into ’27. Thus, risks that ASML guides to a weak ’26 are becoming much less likely in our opinion and ’27 trends will be strong (though not guided so soon by ASML). Importantly for the stock, once ASML guides to ’26 revenue range on October 15th, only ’27 will matter and though ASML will not guide to it, the market will speculate on positive trends and that should help the stock. ASML also continues to trade below prior average multiple of 30-35x and thus appears attractive. We reiterate that ASML remains our top pick in ‘25.”

So, the peak lithography bear thesis – which has been popular in the markets over the last year – only really holds water at N2, and then around 2031 again for one node as TSMC moves to the CFET architecture. If you look at the industry’s roadmap, plenty more scaling is ahead in the coming decades.

This brings us now to the biggest risk for ASML..

For premium subscribers, we’ll dive further into ASML, semicap’s outlook in general, and finally, we’ll review two more out-of-consensus silicon names – like ASML was earlier this year.

Read more

The seeds of this disaster were already sown almost 10 years ago, when Intel failed to transition to 10nm and lost its manufacturing lead to TSMC. Intel seemed to be on 14nm forever, with its 10nm node being delayed with more than 3 years. To make matters worse, Intel only managed to insert EUV into high-volume manufacturing in ‘21 with its ‘Intel 7’ node. During these delays, TSMC moved swiftly in a regular two year cadence to N10, N7, N5, and then N3. This allowed Intel’s competitors, most notably AMD which had been heavily betting on TSMC, to take market share due to superior products. The above revenue chart doesn’t capture just how bad things are going for Intel. Gross profit margins are a better reflection of how pricing has been evolving for Intel’s products:

Collapsing profit margins led to collapsing operating cash-flow, and as the company had to step up capex to become competitive in manufacturing again, this led to a heavy cash-burn. Over the last twelve months alone, Intel burned another $11 billion of cash. It’s beyond comprehension that this company was still paying out a dividend up until Q3 of last year, and obviously all dividend payments should have been halted already in ‘22. By this time it had become also clear financially just how difficult the company’s current position is:

Intel isn’t going bust anytime soon, the company has still $21 billion in cash with another $3.5 billion coming in from the sale of Altera, $1 billion from reducing its stake in Mobileye, another $2 billion from Softbank, and $5.7 billion from the US government. So, in total we have another cash-inflow for the company of $12.2 billion in the coming months. In addition, the company still has a stake in Mobileye currently valued at $8.8 billion, and there is the possibility to sell various other assets. So Intel still has a lot of runway, however, it is also clear that the company’s financial performance has to start turning around in the coming years. This is no easy feat in leading edge semis.

Due to the heavy cash burn, Intel is making it clear that they want to add external customers first to make the investments in their coming 14A node. This is the company’s CFO at the recent Deutsche conference:

“Lip-Bu wants to execute on 18A and he wants to execute on 14A, but he wants to maintain financial discipline while he's doing that. And so what we're trying to communicate is the philosophy around why we're getting into the Foundry business. There's opportunity there. But the volumes required at these new nodes in terms of the spending level, it's hard to get an ROI unless you've got more volume than we have. We were able to do it with 18A. I think it will turn out to be a good node for us from an ROI perspective without Foundry customers, although I do think that we have an opportunity in the second wave to get Foundry customers into 18A. But 14A is different. So we do need outside customers to drive volume and make sure that we get the right ROI. We're not going to put capital in place until we have firm commitments from customers that they're going to use 14A. Now Lip-Bu, every day, he kicks the tires on 14A. He becomes more and more confident around our ability to be successful there. The fact that 18A is now making steady improvement on the yield, it's also giving us a high degree of confidence that 14A will be successful.”

We do think that starting a foundry business is the right move for Intel. The company has good manufacturing tech, while there is a desire from customers to diversify their supply chains. For example, Nvidia allocated their gaming GPUs to Samsung’s foundry not too long ago while Tesla will be sourcing its next-gen AI6 chip from Samsung. At the same time, on-shore semi manufacturing capacity is seen as a national priority in Washington. The latter could help Intel secure certain volumes for its 18A and 14A nodes.

President Trump is known to regularly get involved at the more micro level to help out US businesses. For example, just a few months ago, Trump got involved in the acquisition of US Steel by Nippon Steel, securing all sorts of guarantees to increase US steel manufacturing capacity and a golden share with veto power in about a dozen key areas. Recently, Trump met with Intel’s new CEO, Lip-Bu Tan, and mentioned afterwards that he “really liked him”. This meeting came after the attacks launched by Senator Tom Cotton, which portrayed Lip-Bu as being tied to the Chinese communist party. However, after the meeting, Trump mentioned that he thought that Lip-Bu was unfairly a victim here.

Subscribe now

It’s not hard to put all the above observations together. Firstly, Intel needs volumes in order to invest in its next leading edge node. Secondly, Washington wants more semiconductor manufacturing capacity, which it sees as crucial for national security. And, finally, Trump is known to get involved into business deals to help out US companies, especially when it comes to US-based manufacturing capacity, and Intel’s new CEO has successfully built a new key relationship here. So, we do think it is well possible that at some stage, there will be some horse trading going on behind the scenes, where the leading US semi designers will be asked whether they can allocate some of their (smaller) products to help out Intel and the country in general. Deutsche Bank made a similar conclusion after their meeting with Intel’s CFO:

“Despite the initial dilution, the company views this as a strategic win to bolster their near-term cash balance while eliminating the uncertainty (and time duration) of receiving cash grants from the US CHIPS Act. Further, this investment aligns their incentives with those of the US government (i.e. a strong US-based foundry and we believe decreased odds of any foundry based spinoff). While largely indirect, we also believe the involvement of the US government may help attract US fabless companies to utilizing the Intel foundry business.

Regarding the necessity of garnering external customers for 14A to generate an acceptable ROI, Intel’s CFO indicated that this should not be read as a sign of uncertainty nor lack of commitment to the foundry business, but rather a statement of Intel’s financial discipline. We believe the statement was also meant to encourage greater collaboration with potential foundry customers that may desire to increase their US-based wafer sourcing and avoid future foundry sole-sourcing. Overall, the company remains confident in its ability to generate a solid ROI in 18A over the next several years and expects customer designs for 14A to fall within the CY26/27 range before ramping in earnest in the CY28/29 time frame.”

In the meanwhile, Intel is also taking other measures to improve profitability such as increasing its levels of tool reuse and increased pricing at the next nodes. This is the CFO at the Deutsche conference:

“Naga and Lip-Bu have been pushing on reuse too, and I don't think we did a particularly good job on reuse. So they're really forcing the organization to not dispose of equipment and sell it at pennies on the dollar, but reuse it in the process. I think on the Foundry side, we're in pretty good shape. 18A and 14A have very good cost structures relative to older nodes. The pricing is better. So they should be on a steady path to improving both gross and operating margins as we go into '26 and beyond.”

A foundry business would be lucrative for Intel shareholders. If the company would be able to capture a 12.5% market share over time in a future foundry TAM of $250 billion, this new venture alone could generate over $2 of EPS:

Valuing the above on a 25-30x PE would already result in a future share price of around $60 for the Foundry business. This is really the only credible bull case we can see in Intel at this stage, and needless to say, this remains a speculative story. Yields on its coming 18A and 14A nodes will have to be sufficiently high in order to attract volumes. We do believe that Washington will be pushing to help, but if Jensen explains that the yields are too low, Intel will obviously be in a very tough spot. Under that scenario, it would make much more sense for the US government to give further incentives to TSMC to build out their massive fab hub in Arizona. Overall, Intel remains a speculative and difficult turnaround story. At this stage, we’re more inclined to stay on the sidelines for now. Greater clarity in a roadmap to successfully build out a foundry business is a key factor that could bring us onboard as investors in this name.

Fortunately, for premium subscribers, we will analyze 7 investment ideas next. These will range from hardcore semi engineering firms to AI names and fintech. We’ll cover US, international, and emerging markets stocks.

Read more

This is Nvidia’s CFO on what’s driving increased spend in the AI data center:

“As we continue to deliver both Hopper and Blackwell GPUs, we are focusing on meeting the soaring global demand. This growth is fueled by capital expenditures from the cloud to enterprises, which are on track to invest $600 billion in data center infrastructure and compute this calendar year alone, nearly doubling in 2 years. We expect annual AI infrastructure investments to continue growing, driven by several factors: reasoning agentic AI requiring orders of magnitude more training and inference compute, global build-outs for sovereign AI, enterprise AI adoption, and the arrival of physical AI and robotics.

Blackwell has set the benchmark as it is the new standard for AI inference performance. The market for AI inference is expanding rapidly with reasoning and agentic AI gaining traction across industries. Blackwell's rack scale NVLink and CUDA full stack architecture addresses this by redefining the economics of inference. New NVFP4 4-bit precision and NVLink 72 on the GB300 platform delivers a 50x increase in energy efficiency per token compared to Hopper, enabling companies to monetize their compute at unprecedented scale. For instance, a $3 million investment in GB200 infrastructure can generate $30 million in token revenue, a 10x return.

Adoption of NVIDIA's robotics full stack platform is growing at a rapid rate with over 2 million developers, and 1,000-plus hardware and software applications taking our platform to market. Leading enterprises across industries have adopted Thor, including Agility Robotics, Amazon Robotics, Boston Dynamics, Caterpillar, Figure, Hexagon, Medtronic and Meta.”

Nvidia will continue to see automatic ASP increases as next-gen products will be manufactured on more advanced nodes and with higher amounts of HBM:

There was some analysis circulating online that Nvidia missed with its data center revenues, however, this was caused by the halt in H20 sales to China where revenues declined $4 billion quarter on quarter. As we know, Jensen has been regularly talking to the Trump administration while understanding in Washington continues to grow of the importance of having the dominant AI stack (hardware + software). President Trump himself mentioned at a recent press conference that he would be happy to allow sales of a modified version of Blackwell into China. So, at this stage, it seems quite likely that billions of China revenues will be coming back in the near future. This is Nvidia’s CFO on the China situation:

“We have not included H20 in our Q3 outlook as we continue to work through geopolitical issues. If geopolitical issues subside, we should ship $2 billion to $5 billion in H20 revenue in Q3. And if we have more orders, we can bill more. We continue to advocate for the U.S. government to approve Blackwell for China. Our products are designed and sold for beneficial commercial use, and every license sale we make will benefit the U.S. economy and U.S. leadership. In highly competitive markets, we want to win the support of every developer. America's AI technology stack can be the world's standard if we race and compete globally.”

So while current quarterly growth rates took a dip due to H20 sales winding down, next quarter we will see already very brisk growth rates with revenues growing 17% sequentially, which don’t include China sales:

As Nvidia’s CFO expects they could ship $2 to $5 billion of H20s into China, there is a lot of upside to next quarter’s number. Consensus for next quarter is basically at $54 billion, the same amount where Nvidia’s CFO guided, but revenues could even come in above $60 billion if China sales are resumed, in combination with a beat on revenues ex-China.

As the Trump administration is currently negotiating a trade deal with China, H20 and future B30 sales could well be included as part of the agreement, with China agreeing to buy more US tech in order to reduce the US trade deficit – a big focus point of Trump. This is Jensen on Nvidia’s opportunity in China:

“The China market, I've estimated to be about $50 billion of opportunity for us this year if we were able to address it with competitive products. And if it's $50 billion this year, you would expect it to grow, say, 50% per year. As the rest of the world's AI market is growing as well. China is the second largest computing market in the world, and it is also the home of AI researchers. About 50% of the world's AI researchers are in China. The vast majority of the leading open source models are created in China. And so it's fairly important for the American technology companies to be able to address that market.

Open source is created in one country, but it's used all over the world. The open source models that have come out of China are really excellent. DeepSeek, of course, gained global notoriety. Qwen is excellent. There's a whole bunch of new models that are coming out. They're multimodal and they're great language models. And it's really fueled the adoption of AI in enterprises around the world because enterprises want to build their own custom proprietary software stacks. And so open source models are really important for enterprise, really important for SaaS who also would like to build proprietary systems, and it has been really incredible for robotics around the world.

We're talking to the administration about the importance of American companies to be able to address the Chinese market. H20 has been approved for companies that are not on the entities list, and many licenses have been approved. And so I think the opportunity for us to bring Blackwell to the China market is a real possibility. We just have to keep advocating the sensibility of and the importance of American tech companies to be able to lead and win the AI race, and help make the American tech stack the global standard.”

A bear thesis on Nvidia is that LLMs aren’t generating revenue. Jensen gave some interesting stats on the call that revenue growth is huge, with revenues up tenfold this year:

Altimeter did a similarly interesting analysis in late ‘24 how OpenAI has been growing revenue much faster than Google and Meta in their first years – obviously the latter two became two of the largest companies in world:

And since then, we know that token usage in the cloud has been really exploding in ‘25, with all hyperscalers such as Microsoft and Google seeing extremely high month-on-month growth rates.

Bears will argue that the cash burn is huge at OpenAI, but it seems likely to us that pricing can be increased much, much further. Currently users seem to be getting a near unlimited amount of analysis for around $20 per month, while obviously OpenAI could limit the amount of questions you can ask, or alternatively price an unlimited bundle at a much higher ASP. Currently, xAI’s Grok has started limiting usage to 50 questions every few hours.

There was some regular clickbait analysis in recent years on X that Nvidia is cooking the numbers, with as “evidence” that Nvidia’s Singapore revenues are unusually high. Nvidia’s CFO gave the following simple explanation:

“Singapore revenue represented 22% of second quarter's billed revenue as customers have centralized their invoicing in Singapore. Over 99% of data center compute revenue billed to Singapore was for U.S.-based customers.”

We’ve never paid much attention to this clickbait analysis as all obvious channel checks are pointing to demand being very strong – when Microsoft tells you that they need to sign multi-year contracts with CoreWeave because they can’t get their hands on sufficient GPU capacity, and with all other hyperscalers such as Google and Meta confirming the same, obviously Nvidia’s demand is real.

Nvidia’s challenge is that they need to sell more and more GPUs every year to be a growth company, which introduces obvious cyclical risks at some stage, something semi investors are hugely familiar with. As the current outlook remains very strong, with GPUs still being in short supply, we continue to happily hold our position, especially as also consensus numbers in the coming quarters look too low. Currently also valuation for Nvidia remains unexcessive at a 31x forward PE, a multiple where the company usually trades when the market thinks that EPS are peaking:

You don’t really run this cyclical risk in the hyperscalers. For example, Nvidia will be doing $210 billion in sales this year, if its growth rate would halt or slightly decline at some stage in the coming years, the public clouds would still be adding similar amounts of GPU capacity like they were doing before. These bring in extra revenues they can make from renting these GPUs out in their clouds. So, under that scenario, Nvidia has gone ex-growth for the time being while the public cloud continues to grow at an interesting pace. Therefore we continue to see Microsoft as the lowest risk play to get AI exposure. The company is extremely well positioned with Azure to keep growing at high rates for the coming 5-10 years and possibly much longer, while still trading on a reasonable multiple. This is one of the best positioned businesses in the world and it is fairly non-cyclical, two factors which typically warrant a premium multiple anyways:

For premium subscribers, we will dive into a niche, hardcore semi engineering company that is seeing even much higher growth rates from the AI data center than Nvidia. For one of its key products, dollar semi content per accelerator will go up tenfold.

Read more

“We have the largest pipeline we have ever had at this point of the year and feel very good about our product portfolio and innovation momentum. Our leading indicators remain positive. As such, we are increasing our full year 2025 revenue guidance to be between $586 million to $590 million. Now let's turn to Q3. We are positive about the long-term outlook for the U.S. federal and the public sector business. Near term, the uncertainty in the spending and restructuring environment in the public sector is likely to impact our Q3 revenue growth as this is the largest quarter for the U.S. federal government business. In addition, we continue to see opportunities to convert on-premise government contracts to SaaS, which we have contemplated in our guidance.”

This is CEO discussing the long term opportunity:

“These engines of our business continue to feed off three main drivers at the heart of our industry. Number 1, finance is in the initial phase of its digital transformation. Having utilized outdated legacy financial systems for decades, CFOs are recognizing the crucial need for a platform to provide a single view into financial and operational data across the enterprise to effectively steer the business. Number 2, the role of the CFO is evolving and expanding. CFOs are transitioning from focusing on reporting to providing more strategic insight and operational planning to help drive business execution. And number 3, the use of advanced tooling is enabling finance teams to drive business performance, not only measure it. We are seeing a shift in customers looking for applied solutions that are purpose-built for finance. Proven AI solutions have the potential to give finance teams greater predictability and visibility into their business with more speed and agility.”

The risk-reward in the shares is starting to look attractive here with this short term government headwind. Long term, the company should be looking at high growth rates while valuation is currently at 7.5x consensus ‘26 revenues. The analyst at William Blair makes a similar conclusion:

“OneStream reported solid second-quarter results that beat consensus across all key metrics. While OneStream expects growth to face some pressure during the third quarter given potential headwinds to its public sector business, the company believes that billings will quickly return to 20%-plus growth in the fourth quarter given the strong pipeline trends it is seeing. Management noted that the company’s pipeline is the largest it has ever been exiting the second quarter. In the near term, OneStream believes that recent DOGE uncertainty could impact demand in the company’s public sector business. In order to account for these headwinds, management took a prudent approach to guidance around the third quarter given it is the largest federal spending quarter. While billings are expected to face some headwinds, management expects a reversion to billings growth of at least 20% in the fourth quarter given strong pipeline trends in its commercial business. Management also highlighted that the public sector represents a significant long-term opportunity for the company, especially since OneStream is the only cloud-based FedRAMP certified CPM provider. OneStream trades at 7.7 times our calendar 2026 revenue estimate versus peers at 11 times. We believe OneStream’s leadership position in a large market that is early in its digitization cycle should help the company deliver durable growth and expanding margins over the next few years.”

So this sell off should provide a nice entry point both for shorter term as well as long term investors:

After a poor start of the year, AMD’s shares have been on a roll since early April where macro fears around the new tariff announcement took the whole market down. However, since then, the US has successfully concluded trade agreements with its major trading partners including the EU, UK, Japan, South-Korea, Indonesia etc., with even a trade deal with China now in the making.

This is Deutsche on AMD’s quarter:

“AMD delivered a solid top-line beat in its 2Q25 report ($7.7b, +4% above DBe) and provided an optimistic 3Q25 guide ($8.7b, +2% above DBe). While aggregate revenues being above DBe is a positive, the drivers were more mixed with Gaming driving upside in 2Q (lower quality semi-custom) and Instinct (DC GPUs) driving the upside in 3Q (higher quality). On the all-important Instinct business, AMD appears to be ramping its MI355 product slightly ahead of plan, yielding DBe +75% q/q in 3Q despite no China-focused MI308 shipments being included. We expect this Instinct goodness to persist through 2H25 and into 2026, incrementally raising our ests accordingly (DBe $6.5b/$11b in 2025/26). Beyond revenue, GMs remain impressively steady despite mix gyrations, with opex continuing to increase as AMD accelerates investments into the AI opportunity. Overall, we remain impressed with AMD’s execution across its array of product lines, expecting solid growth to continue in CPUs and likely accelerate in DC GPUs. However, even with these growth drivers reflected in our model (CY25E EPS ~unchanged, CY26E ~ +10%), we see EPS power in the ~$6-7 range as largely being reflected in a current valuation (~25-30x P/E). Consequently, we maintain our Hold rating, but raise our P/T to $150 (~26x CY26E).”

We think the above conclusion from Deutsche is too negative. The penetration of AI automation in the wider economy is still very low, so we could be looking at high growth rates for AI semis for the foreseeable future. At the start of the LLM revolution, AMD was massively behind Nvidia especially in its software stack. However, since then, its ROCm computing platform has become much more capable and AMD is becoming a viable option, especially in inference. So we think that long term, AMD should be able to take its share of the data center GPU market, while the shares are not particularly aggressively priced at 34x:

Next, for premium subscribers, we will discuss the massive sell off in The Trade Desk’s shares (-40%) and whether this represents a buying opportunity. Subsequently, we will discuss a smaller semi name for which a large opportunity in the AI data center is now emerging.

Read more

Token consumption in the cloud is doubling on an almost monthly basis. This is Sundar from Google:

“The growth in usage has been incredible. At IO in May, we announced that we processed 480 trillion monthly tokens across our surfaces. Since then, we have doubled that number, now processing over 980 trillion monthly tokens, a remarkable increase. The Gemini app now has more than 450 million monthly active users, and we continue to see strong growth in engagement with daily requests growing over 50% from Q1. And Veo 3, our new short video product reached nearly 1 million monthly active users.

Overall queries and commercial queries on Search continue to grow year-over-year. And our new AI experiences significantly contributed to this increase in usage. We are also seeing that our AI features cause users to search more as they learn that search can meet more of their needs. That's especially true for younger users. We know how popular AI Overviews are because they are now driving over 10% more queries globally and this growth continues to increase over time. AI Overviews are now powered by Gemini 2.5, delivering the fastest AI responses in the industry. We also saw strong growth in the use of multimodal search, particularly the combination of lens or Circle to Search together with AI overviews. This growth was most pronounced among younger users.”

Obviously, as queries move more to LLMs and AI agents, disruption in Search is a long term threat to Google. Although at the moment the company’s bread-and-butter search business is still growing well.

Google is obviously not an expensive name for attractive EPS growth, although the key risk is that its key earnings driver—the search business—gets disrupted over time:

One of the key beneficiaries of the explosion in AI is SK Hynix, the South-Korean memory manufacturer who holds the strongest position in the HBM market, and the company described the current business environment well:

“The past quarter began with concerns about a slowdown in demand due to trade tensions and overall economic uncertainty. However, demand for AI memory continued to grow, driven by aggressive AI investments from big tech companies. This was further supported by customers' preemptive purchasing to prepare for external risks, resulting in a more favorable environment than initially expected. Both DRAM and NAND shipments exceeded our original guidance and pricing conditions also improved. As a result, second quarter revenue increased by 26% Q-o-Q and 35% Y-o-Y, reaching an all-time high of KRW 22.2 trillion.

Given demand is expected to grow from new product launches in the second half, we believe the likelihood of sharp demand corrections due to customers' inventory adjustments would be low. Looking at demand by application, we expect healthy demand growth in the server market, driven by increasing investments from big tech companies despite macro uncertainty. In the AI market, big tech companies have begun full-fledged competition by unveiling AI agents with enhanced reasoning models. These models perform iterative thought processes to solve problems autonomously, requiring systems capable of processing significantly more data than conventional AI models. This evolution is expected to further drive demand for high-performance, high-density memory. Additionally, ongoing investments by governments and corporations for sovereign AI are likely to become a new long-term driver of AI memory demand. In the PC and smartphone market, the proliferation of AI applications is expected to drive replacement demand this year. Memory content per device are also expected to grow with wider adoption of high-performance memory required to support the AI functions.

Our HBM3E products are receiving high marks from all customers for their performance and supply reliability, and we plan to maintain industry-leading competitiveness as we transitioned into HBM4. In March, we supplied industry's first samples of HBM4 to customers. We are currently working closely with partners to optimize system-level performance and are preparing to deliver products on time for our customers' needs. Construction of our Yong-in fab Phase 1 is on track for completion in Q2 2027. And through these facilities, we aim to secure the manufacturing infrastructure necessary to meet future demand. Our total investment for this year is expected to increase compared to the previous plan as we now have secured visibility on next year's demand figure based on discussions with customers. We assessed that some proactive investment this year is necessary to ensure timely support of our HBM products.”

Thus, the outlook for AI in the cloud market remains very strong due to the two scaling laws—the first being the expansion of model sizes and reasoning via reinforcement learning in the training phases, and the second being increasing levels of reasoning in the inference phase. With the low penetration rates of AI workloads in the general economy, we’re looking at a strong runway of growth in the coming years, resulting in the current explosive growth rates we’re seeing in the cloud today.

This is UBS on SK Hynix post the results:

“Today, SK Hynix's stock price declined 9%, amid debate on HBM 2026 contract negotiations. We reiterate our Buy rating and PT of Won350k for the following reasons: 1/ Our channel checks continue to indicate that SK Hynix will likely secure a largely stable market share in the HBM market in 2026 around 50% of total bits. 2/ This is on the back of continued lead at Nvidia, but also as we recently noted, strong share gains at ASIC vendors. 3/ We see upside in overall cloud AI demand, as evidenced by TSMC comments in today's 2Q25 conference call, the recent announcement regarding H20 exports to China resuming ( link ), and general pent-up demand from tier 2/3 customers more so into the GPU segment into 2026.”

Goldman downgraded the stock due to possible HBM pricing headwinds in ‘26:

“We downgrade SK Hynix (Hynix) to Neutral from Buy with an updated TP of W310,000 implying 5% upside. While we continue to be constructive on the long-term outlook for the HBM market powered by the rising need of the high bandwidth to improve the efficiency of AI workloads by both the GPU and ASIC customers, we believe HBM pricing could decline for the first time in 2026, with increasing competition and pricing power gradually shifting to the major customer where Hynix has an outsized exposure.”

SK Hynix is currently trading on a peak-of-the-cycle multiple, and so if its earnings continue to move higher in the coming years—which looks likely—the shares will continue to do well:

Tesla is one of the most interesting names in edge AI applications due to its development of humanoid robotics and robotaxis. Tesla could well be the best hardware engineering firm (ex semis) in the world today, and so the company is well positioned to capture its share of these potentially vast new markets. While China is becoming impressive as well at hardware engineering with companies such as BYD and Xiaomi, we don’t think that the US government will allow Chinese humanoid robots to be inserted into US homes and factories. So the competitive field for Tesla will actually be limited. While humanoid robots will only start driving strong revenue growth in 4-5 years or so, the robotaxi market could be more imminent, with Tesla looking to cover 50% of the US population in the near future (they will focus on large cities). Elon is known for frequently setting over-ambitious goals, but he is also known for achieving incredible breakthroughs at all his companies—from SpaceX to Neuralink, Xai and Tesla. This is Elon on their robotaxi product:

“In a couple of weeks or so, we're getting the regulatory permission to launch in the Bay Area, Nevada, Arizona, Florida, and a number of other places. So as we get the approvals and we prove out safety, then we will be launching autonomous ride-hailing in most of the country. And I think we'll probably have autonomous ride-hailing in probably half the population of the U.S. by the end of the year. That's at least our goal, subject to regulatory approvals. I think we'll technically be able to do it. So assuming we get regulatory approvals, it's probably addressing half of the population in the U.S. by the end of the year. But we are being very cautious. We don't want to take any chances. And so we're going to go cautiously. But the service areas and the number of vehicles in operation will increase at a hyper-exponential rate.

The Cybercab, which is really optimized for autonomy, has got probably sub-$0.30 per mile potential over time, maybe $0.25. It's really -- like if you design a car from scratch to be a cost-optimized robotic taxi like Cybercab -- for example, we're not trying to make its cornering like incredibly well like a Model 3 would or Model S would, all of our cars that are driven by people are super fun to drive. They've got incredible acceleration, incredible cornering capability. But we're confident that very few people in a Cybercab want to be hurtling around. So we've produced the top-end speed, which means we can use more efficient tires. We don't need as much acceleration. We want stopping distance, but we're not expecting it to be heavily used. It's a gentle ride. Essentially, if you design it for a gentle ride and then you have a much more optimized design point. So that's why it seems probable we could achieve that. Especially, Optimus is serving, cleaning up the car and doing maintenance and stuff. And doing automatic charging. So I think the actual cost per mile of Cybercab will be very low. The cost per mile of our existing fleet will be higher, but still very competitive. So maybe something over $0.50, I'm just guessing, yes. So really, it tells us robotaxi will go from tiny to gigantic in terms of operations in a pretty short period of time. Like my guess is it has a material impact on our financials around the end of next year.

On the full self-driving front, we're continuing to make significant improvements just with the software. So we're expecting to increase the parameter count. Actually, at this point, we think we can probably 10x the parameter count. So this is actually a very tricky thing to do because, as you increase the parameter count, you get choked on memory bandwidth. So still a lot of improvement on the existing hardware to happen.”

Obviously if Cybercab can launch successfully and charge $0.4-0.5 per mile, we’re looking at an Uber killer. The customer will have the vehicle to him or herself while travelling immensely cheaply, so demand is likely to be huge under that scenario. The key milestone will be proof that the vehicles can drive around much safer than human drivers can. As soon as one of Tesla’s robotaxis will injure a human, or merely be involved in an accident, expect the media to initiate a large scale attack. This is an obvious risk factor to a successful rollout.

The other exciting product is Optimus, which Tesla is successfully training to do simple tasks such as robotaxi maintenance and factory tasks. This is Elon on the humanoid robot:

“We're getting Optimus to the point where it is a phenomenal design. So we're in Optimus version 2 right now, sort of 2.5. Optimus 3 is an exquisite design, in my opinion, and I predict it will be the biggest product ever. It's a very hard problem to solve. You have to design every part of it from physics-first principles. There's nothing that's off the shelf that actually works. So you got to design every motor, gearbox, power electronics, control electronics, sensors, the mechanical elements. We also got to train Optimus to use its limb sensors with a neural net. But we'll be applying the same techniques that we applied for our car, which is essentially a 4-wheel robot.

And Tesla, it is important to note that Tesla is by far the best in the world at real-world AI. Google is good at AI, but they're not good at real-world AI. And by far, Tesla has the best inference efficiency. And people talk about parameters, but the actual constraints in the hardware are how many gigabytes of RAM and how many gigabytes per second can you transfer from RAM. Therefore, it is not a parameter constraint. It is a byte constraint. And Tesla has the highest intelligence density of any AI by far. Intelligence density will be a very big deal in the future.

There will probably be prototypes of Optimus 3 end of this year and then scale production next year. We're going to try to scale Optimus production as fast as it's humanly possible to do, so we'll try to get to 1 million units a year as quickly as possible. We think we can get there in less than 5 years, it's my sort of guess. That's a reasonable aspiration and it seems like an achievable target.”

So, Tesla is a name we like, but you have to believe that they can capture a share of the robotaxi or humanoid robotics markets, otherwise the current valuation at $1 trillion doesn’t make any sense. For example, some modelling on the humanoid robotic market alone shows an attractive IRR, and it’s well possible that the company can sell hundreds of millions of these over time if these bots can be useful in factories and homes..

While 20 million in 2032 already looks unlikely, you can model in a much higher PE multiple if these things start to take off. The annual smartphone market is 1.34 billion units. If Tesla can sell 100 million units of Optimus over time per annum, the revenues would be around $3 trillion. Put that on 10x Sales and you get a $30 trillion market cap.

Turning to robotaxis, capturing 5% of the global number of miles driven in passenger vehicles results in a 37% IRR:

Again, a 5% penetration rate in 2030 is likely too optimistic, but you could model in a much higher PE multiple if robotaxis really take off as growth rates will be explosive.

In the meanwhile, in mature semis, Texas Instruments delivered a disappointing guidance. This is Deutsche:

“TXN delivered a trademark beat for 2Q (revs +2% above mid-pt), but the surprise came in their guide being a “seasonal” +4% q/q, slightly below DBe on both revs and EPS. After guiding (and reporting) well above seasonal revenue growth in 2Q, TXN’s 3Q guide returning to “seasonal” appears to be a reflection of the 2Q strength being driven by tariff-related pull-ins (China especially strong q/q) more so than the desired emergence of the cyclical tailwinds.

Positives: 1) TXN delivered a typical beat off of an already above-seasonal guide in 2Q led by its Industrial business (DBe +14% q/q), with all other segments also growing q/q except for a somewhat surprising lsd% drop in from the co’s Auto segment. 2) The co remains confident that it has turned the cyclical corner and that the long-awaited upturn has begun, as all segments (ex-Auto) are showing sequential improvements. The co’s view is that the Auto upturn is just a matter of time. 3) TXN described in general terms the benefit it expects to see from the tax legislation recently passed in the United States, with the net take that its GAAP tax rate will be slightly higher in 3Q/4Q25 before falling in CY26 and beyond. More importantly, its cash tax rate should also be lower in CY26 and beyond.

Overall, we believe TXN’s shift to a more conservative outlook and the resulting slight reduction in our rev/EPS estimates is not the outcome implied in the co’s premium valuation multiple (~25% above 5yr avg). Consequently, we expect the valuation to somewhat normalize, maintaining our Hold rating and $190 P/T.”

We disagree with the above conclusion from Deutsche as TI is currently trading on bottom-of-the-cycle earnings, so naturally this results in a premium PE multiple:

The downturn can easily be seen in semi shipments—we’re currently in one of the worst downturns in history post the covid semi boom:

And TI can start generating an attractive 5.5% FCF yield in ‘28 on consensus numbers, as either capex levels normalize or revenue growth returns:

All of the above means that semicap should continue to be an attractive industry—AI is driving high growth in leading edge semis, while mature semi markets are still at the bottom of the cycle. Meanwhile, due to the rising complexity of leading edge semis, capital intensity and wafer ASPs will continue to increase.

Bears are out for ASML claiming the story is broken. For premium subscribers, we’ll dive into ASML and a number of other names we like in semicap.

ASML’s Bears & Outlook

Read more

Over the past several months, token consumption has been exploding in the cloud. Anecdotally, the AI workloads we’re running in the cloud have continued to slow down over the past few months, especially when US business hours start, which is a good indicator that GPU utilization has been increasing. In particular, ChatGPT regularly feels slow, while Grok3 slowing down might just be a function of xAI shifting GPU capacity to the mighty SuperGrok. So, it’s no surprise that TSMC raised its outlook for the year. This is TSMC’s CEO on the call:

“We believe the demand for semiconductors is very fundamental and will continue to be robust. Recent developments are also positive to AI's long-term demand outlook. The explosive growth in token volume demonstrate increasing AI model usage and adoption, which means more and more computation is needed, leading to more leading-edge silicon demand. We also see AI demand continuing to be strong, including the rising demand from sovereign AI. Therefore, we now expect our full year 2025 revenue to increase by around 30% in U.S. dollar terms, supported by strong demand for our industry-leading 3-nanometer and 5-nanometer technologies underpinned by growth in our HPC platform. Amidst the uncertainties, we will remain mindful of the potential tariff-related impact and be prudent in our business planning going into second half 2025 and 2026 while continuing to invest for the future megatrends.”

TSMC’s revenue growth is being driven by the high-performance computing segment..

.. which continues to grow at 70-80% year-over-year rates:

TSMC’s CFO clarified that due to strong AI demand, they don’t expect the CoWoS market to balance anytime soon:

“The demand from AI is getting stronger and stronger, and so now we are in a mode trying to narrow the gap for CoWoS. I don't want to use the word balance. The last time you guys misunderstood what I said. So, I will say we try to narrow the gap, all right? So, momentum is still there and very healthy.”

A major new tailwind for AI demand, which was only announced this week, is that Nvidia will again be allowed to export its H20 chip into China. The logic is that the US wants to limit China’s ability to develop a competing AI stack, running on Huawei silicon and software, and manufactured by SMIC. While this silicon is much less powerful than Nvidia’s, if you network a lot of them together into a giant cluster, you still get a huge amount of computing power. Thus, it makes much more sense to allow Nvidia to capture a large share of the Chinese AI market, which will get developed anyway as the genie is out of the bottle, and then have those profits flow back to the US. It makes much more sense to have that capital being reinvested in US R&D as opposed to Huawei-SMIC R&D. The Chinese AI market will be huge and so this is an obvious positive for both Nvidia and TSMC.

Demand for TSMC’s next node, N2, is very high. This is TSMC’s CEO:

“We expect the number of new tape-outs for 2-nanometer technology in the first 2 years to be higher than both 3-nanometer and 5-nanometer in the first 2 years, fueled by both smartphone and HPC applications.”

Historically, Apple made up all of the demand in the initial year(s) when TSMC introduced a new node, while the high-performance silicon designers such as Nvidia, AMD, Broadcom etc. remained one or two generations behind. With the explosion and competitive race in AI, these high-performance silicon players are now looking to transition to the leading edge more swiftly. This is a major positive for TSMC as its leading edge wafers come at much higher ASPs, and it will allow the company to reach target gross margins faster on the new node due to larger volumes.

Due to the huge demand for leading edge silicon, TSMC is building out three massive fab clusters—one in Arizona with 6 advanced fabs (including N4, N3, N2, A16 and A14) and 2 advanced packaging fabs, and two fab clusters in Taiwan including 11 fabs and 4 advanced packaging fabs in total. We also got an update on the call on the progress in A14, TSMC’s next major node after N2:

“Featuring our second-generation nanosheet transistor structure, A14 will deliver another full node stride from N2 with performance and power benefits to address the increasing structural demand for high-performance and energy-efficient computing. Compared with N2, A16 will provide 10 to 15 percent speed improvement at the same power or 20% to 30% power improvement at the same speed, and about 20% chip density gain. Our A14 technology development is on track and progressing well with device performance and yield improvement on or ahead of schedule. Volume production is scheduled for 2028. We will continue our strategy of continuous enhancement with A14, including a Super Power Rail offering planned for 2029. We believe A14 and its derivative will further extend our technology leadership position and enable TSMC to capture the growth opportunities well into the future.”

This innovation roadmap is extremely important for investors as next nodes always come at much higher ASPs, driving strong revenue growth for TSMC. Wafer volume growth hasn’t been a growth story at TSMC over the last years, meaning that the strong revenue growth since the covid peak has come from wafer mix—i.e. selling more N3 and N5 wafers at much higher ASPs versus N7 and N16 wafers.

Although wafer shipments picked up tremendously during the last quarter as TSMC shipped 3.7 million wafers, a number reaching similar levels as shipments during the peak phases of the covid semi boom:

This is helped by also mature semi markets such as automotive and IoT picking up..

.. although automotive is also being driven by the rise of autonomous driving. Nvidia has been winning plenty of OEMs to build autopilots on its Thor and Orin chips, with this business now generating over $500 million per quarter:

All of the above is naturally positive for semicap as well, as TSMC’s capex will make new highs this year:

And the bull case for semicap stocks is that despite peak capex levels, TSMC’s capex-to-revenues ratio remains modest:

The only short term negative is that TSMC is currently seeing gross margin headwinds due to the weak dollar—wafers are priced in $, while costs are in TWD. But it sounds like the company will be working to increase wafer ASPs, as well as make use of other levers to increase gross margins. This is the CFO on this topic:

“NT dollar is the reporting currency of our financial statements. Nearly all of our revenue is in U.S. dollars, while about 75% of cost of goods sold is in NT. Therefore, fluctuations in the exchange rate between U.S. dollar and NT will have a sizable impact to our reported revenue and gross profit margin. The sensitivity of the revenue to dollar NT exchange rate is nearly 100%—that is every 1% appreciation of NT against U.S. dollar will reduce our reported NT revenue by 1%. The sensitivity of our gross margin to the same 1% exchange rate change is about 40 basis points. That is if NT appreciate 1% against the dollar, our gross margin will come down by about 40 basis points.

As a reminder, 6 factors determine TSMC's profitability: Leadership technology development and ramp-up, pricing, capacity utilization, cost reduction, technology mix and foreign exchange rate, which is not in our control. When the foreign exchange rate is unfavorable as it is currently, we will focus on the fundamentals of our business and lean on the other 5 factors to manage through it, as we have successfully done in the past. Thus, even with the unfavorable foreign exchange rate, we believe a long-term gross margin of 53% and higher remains well achievable.”

Coming back to the outlook, JPMorgan reached a similar conclusion about AI demand after meeting with Broadcom’s management, while also seeing a gradual U-shaped recovery in non-AI semi markets:

“We hosted an investor group meeting with Broadcom's President and CEO, Hock Tan, CFO, Kirsten Spears, and VP of IR, Ji Yoo. The discussions centered around strong AI demand trends, which are driving strong growth in Broadcom's custom AI XPU business and driving demand for its high performance networking portfolio. Additionally, there are gradual recovery trends in the non-AI business (seeing positive order inflection) and sustained growth in its VMware business. Key takeaways in our view were:

1) AI demand continues to remain strong, with incremental demand from inferencing workloads that were not accounted for in the prior 2027 SAM forecast, potentially leading to upside (although too early in the adoption curve to make a definitive call).

2) Strong execution in product technology development, with the 2nm 3.5D AI XPU product tape-out on track for this year and the team is on track to tape-out first-generation AI XPU products this year with its two leading prospect engagements (we believe with Arm/Softbank and OpenAI).

3) AI compute workloads are driving strong networking demand and presenting significant dollar content capture opportunities in scale-up networking (5-10x more content versus scale-up networking).

4) The team is seeing a gradual "U" shape recovery in its non-AI semiconductor business, reflected in the bookings/order inflection the team is seeing now, which could contribute to the positive EPS revision cycle next year.

5) The VCF platform conversion is expected to sustain strong growth in its VMware business until late 2026/2027, at which point VMware's revenue base is expected to reach $20B in annualized revenues, with growth returning to more normalized levels of mid to high single-digit% thereafter.

6) Semiconductor operating margins are expected to continue expanding, driven by the operating leverage of the model, despite gross margin headwinds from the AI XPU business.

Lastly, the team’s focus will be on reinvesting in the business over the next 1-2 years on the large AI growth opportunities, rather than pursuing M&A, which could dilute attention and resources.

Overall, we see accelerating AI fundamentals combined with improving fundamentals in the non-AI semiconductor business and further VMware revenue unlock driving strong revenue and earnings growth. Broadcom is the #2 global AI semiconductor supplier and the #1 custom chip ASIC supplier, and we reiterate our Overweight (OW) rating on AVGO.”

Semis are on fire, which is a great environment for stock pickers to find new opportunities. In the second part of this post, we’ll detail a small, niche semi name which is now starting to see (potentially) huge demand from the AI end-market. We’ll do some detailed modeling on the size of this new revenue opportunity, along with giving the insights from industry insiders on the company and this new tech.

A New Niche Semi Winner From AI

Read more

The Token Explosion in the Cloud

The demand for tokens in the cloud is absolutely exploding. This is Nadella on Microsoft’s recent call:

“We processed over 100 trillion tokens this quarter, up 5x year-over-year, including a record 50 trillion tokens last month alone.”

This means that 50% of all tokens over the last year were consumed in March alone. And we’re seeing an even bigger explosion at Google, this is from Barclays:

“We break down Alphabet’s recent disclosure around inference tokens. Key conclusions include: 1) AI Overviews are fueling a huge ramp in inference at GOOGL, 2) GOOGL is processing 5-6x more tokens than MSFT (Azure) as Search is ~6x the size of ChatGPT, which is ~2-4x the size of Gemini, 3) GOOGL is using nearly 10% of its total AI compute capex for inference tokens, the bulk (90%+) is still likely used for training new models and powering AI products like recommender systems, etc., and 4) all these inference tokens only cost GOOGL around $750m in 1Q25 using Gemini 2.5's rate card and a few assumptions, hence the rate of deleverage from infusing AI into Search appears manageable (which may come as a surprise to some investors). AI Overview costs are likely around 1% of Search revenue compared to core costs at around 18% of revenue (i.e., core costs to run organic and ads, excluding TAC costs).”

In total, token consumption at Google is up 50x over the last twelve months. One of the most straightforward ways to play this explosion in AI applications investment-wise are the big clouds—i.e. Microsoft Azure, Amazon Web Services and Google Cloud Platform. We’ve written a number of times before how enterprises architect workloads in these clouds, using cloud-specific code (SDKs), API calls, and command lines, to make use of the breadth of services on these cloud platforms ranging from AI to databases, data analytics systems, cloud functions, load balancers, secure networking, etc. These are businesses with high switching costs and with high barriers to entry, resulting in Microsoft and Amazon controlling a large part of this market.

In our view, neo-clouds have grown tremendously over the last few years due to the large shortages in GPU supply. What happened is that these clouds are fully committed to the Nvidia ecosystem, and thus Nvidia made sure that these clouds had plenty of GPU supply. Thus, the big clouds’ focus on developing ASICs to run their cloud workloads came back to bite them over the last few years as they didn’t get the GPU allocations they were looking for. It’s well known that enterprises have been struggling to get the GPU access they need in their main clouds, which is why Microsoft decided to interconnect their data centers with CoreWeave. This is a director at Microsoft commenting on this deal:

“CoreWeave is probably the best of the new AI hyperscalers, and they make it really easy to spin up and spin down resources. They are 100% dedicated to AI and one of the big reasons we have the partnership with CoreWeave, is because they had just so many NVIDIA chips, as well as the strong infrastructure. They know how to move data around incredibly well and then to do training and inferencing on those chips. CoreWeave is one of those companies that has focused on their NVIDIA relationship, and that’s really benefited them in terms of their procurement strategy. So, what we've done with CoreWeave is to interconnect our network directly to theirs through optical fiber. And then the customer can bring their data and models into Azure, but it is actually being spun out into the CoreWeave network and their GPU clusters.”

Thus, CoreWeave is clearly a legit company. Basically their data centers are running on the state-of-the-art Nvidia ecosystem resulting in tremendous performance. Where we are cautious is their long term competitive positioning. We basically see them as a play on GPUs remaining in short supply. Should the GPU market become more balanced 2-3 years from now, the main clouds will have the GPU capacity they want and then there’s no need to outsource workloads to external clouds providers. Data gravity effects in the cloud are strong—transferring high amounts of data around is both costly (egress fees) and time-consuming, which is the reason you’d typically want to centralize your data.

That the neo-clouds are struggling to attract enterprise workloads can also be witnessed in their ‘24 disclosure—basically Microsoft brought in 62% of their revenues, illustrating how enterprises are simply using Azure’s APIs and are not even aware that in the back Microsoft is offloading these workloads onto CoreWeave. Thus, we take the view that the neo-clouds don’t have a particularly strong competitive positioning in the long term. The amount of services on their cloud platforms is very limited and doesn’t come close to what you need to deploy an app, let alone run an enterprise’s IT infrastructure. Basically, their current main use is to handle AI workloads as the main clouds don’t have sufficient GPU capacity.

We don’t see having better access to Nvidia’s GPUs as a structural competitive advantage. It is more likely that at some stage, supply will catch up with demand. At the same time, competing AI stacks are becoming viable solutions. Google has successfully developed the TPU - XLA - JAX stack, built from the company’s custom accelerator, the XLA computing platform as the answer to CUDA, and JAX as the answer to the high level Pytorch framework. Also, AMD is becoming an option for inference, with Pytorch being able to compile to AMD GPUs via the ROCm computing platform. Nvidia has an incredibly strong ecosystem in AI, but we don’t see having a preferential relationship with Nvidia as a particularly strong investment case for the neo-clouds over the long term. We’d rather invest in companies that have great assets themselves, as opposed to a great relationship, which may or may not last.

Another potential weakness in their business model is that according to this analysis from Deutsche, most of the NPV for CoreWeave only comes in years 5 and 6 of renting out GPUs:

"This past week, CoreWeave hosted an update call on its financing strategy with sell- side analysts. In this note, we review the illustrative contract example laid out by the company this past week and also revisit the approach we took to estimating unit economics in our initiation report. At a high level, our conclusion from both are consistent in that the company at least fully recoups its investment within an initial 4-year contract period (including cost of capital) and that a majority of potential NPV attributable to equity holders based on the current financing mix is captured in years 5-6 of GPUs useful life."

Two remarks here. Firstly, GPUs remain in short supply so price normalization is likely once the market becomes more balanced. This will put pressure on rental pricing over time and worsen the economics. Secondly, GPUs can start failing after three years already and there’s also an incentive to move to new hardware given the massive performance leaps in GPUs. This is a manager at Amazon Web Services via Tegus:

“NVIDIA utilization rates are very high. Even if you have money, small enterprise, you may not be able to get a hold on the NVIDIA instances, either at AWS or any other hyperscaler. It's still very hard to get a hold on. A traditional server's useful life now is about five, six years, accounting and technical useful life. For NVIDIA A100, we see it's probably only three years from deployment because it's too costly to maintain them after three years. They are occupying a very precious spot in the data center. We rather swap them out with Blackwell. A lot of A100s are being swapped out soon, and now, if we have more Blackwells coming in. That's the typical useful life for AI chips we see at this moment. It's about three years.”

It’s really important that CoreWeave’s economics work out as the current financials are jaw-dropping—a $77 billion market cap for a company that generated $2.7 billion in revenues over the last twelve months, managed to burn $7.6 billion of cash over the same time period (largely in capex), while being $12 billion in debt.

Overall, from a cloud engineering standpoint, we don’t see the neo-clouds as having a particularly strong position due to the lack of breadth of services on their platform. Instead, we see GPU shortages at the large clouds as the main reason for why these neo-clouds have been able to pick up strong revenue growth. In addition, from a business model standpoint, we’re not sure whether their economics will work out long term. In our view, neo-clouds will always have to discount GPU pricing heavily in order to attract AI workloads away from the big clouds. For long term investors, we see the two dominant big clouds as much more attractive given their strong market positioning, scale advantages, customer stickiness due to data gravity and high switching costs in the cloud, and healthy economics.

Next, for premium subscribers, we’ll dive into a smaller semi name that is winning orders from the large players in AI accelerators.

A Niche Winner in the AI Data Center

Read more

Now that Q1 reporting season is over and as we’re heading towards season 2 for the year, we’ll go through some of the wider developments in the semi industry. Let’s start off with Navitas, which truly has been a stellar performer with the share price nearly quadrupling over the last month.

Navitas’ Entry into the AI Data Center

This is Deutsche Bank discussing the win:

“NVTS announced a non-exclusive collaboration with NVDA on their next-generation 800V HVDC architecture, signifying a major win for NVTS and validating their commitment to GaN and SiC power solutions. This architecture, designed for high-power AI workloads, addresses the growing power demands of data centers by improving efficiency and reducing infrastructure complexity. NVDA's adoption of the co's GaNFast and GeneSiC technologies, spanning the power delivery chain from grid to GPU, reinforces the efficacy and reliability of NVTS' offerings. By enabling the reduction of copper wiring and eliminating multiple conversion steps, NVTS' technologies contribute to significant improvements in power density, efficiency, and cost-effectiveness.

We view the strategic collaboration between NVDA and NVTS as a much needed "win" for NVTS and consider it a significant validation of NVTS' technology and its potential for long-term growth. We believe this ultimately provides a strong endorsement of NVTS' GaN/SiC solutions, positioning the co as a key enabler for next-generation AI infrastructure. This partnership strengthens NVTS' credibility within the AI data center market, suggesting a significant opportunity for increased revenue and market share as the demand for power-efficient solutions grows with the increasing complexity and energy consumption of AI workloads. While the co is unable to specify the long-term financial impacts of the collaboration at this time, we don't expect to see the benefits to materialize until late CY26 and CY27 (in-line with the expected ramp of NVDA’s Ruben Ultra).”

Navitas’ forward revenue estimates have come a long way down due to the downturn in mature semis. The company has an especially good position in the GaN power semiconductor market, which is closely linked to green technologies such as electric vehicles and solar.

As these end markets went into a downturn, Navitas’ growth outlook suffered as well:

Currently, a large part of the company’s revenues are coming from consumer electronics products such as smartphone fast chargers. However, in the coming 5 years and beyond, GaN adoption in semis will continue to rise, especially in industries where power efficiency is important such as the data center, electric vehicles and solar. Thus, Navitas’ should have an attractive long term growth outlook as the specialist in GaN, with the company having a leading market share. The main risk would be mature semi competitors who also play in the SIC space entering GaN and building up capacity, increasing the risk of periods of oversupply. What gives Navitas some protection during downturns is that the company is an asset-lite semi designer. Basically the company develops and designs their semi tech, and then TSMC and X-Fab take care of the manufacturing.

This is Rosenblatt adding further color:

“Given the exponential growth in data center power needs, there needs to be many changes to today's traditional approach to power management. We are convinced that the market is at an inflection point. Today's data center architectures (54V in-rack power distribution) cannot handle the amount of power needed for increasing AI computation (gigawatts of power). This has led to a redesign of data center architectures, such as the 800V HVDC architecture that NVIDIA is developing. This new architecture eliminates several AC-DC and DC-DC conversion steps. Through Navitas' product portfolio, the GaN and SiC technologies cover power delivery from beginning (the grid) to end (server/GPU). Some products in the portfolio include high-power GaNSafe power ICs, medium voltage GaN devices, GeneSiC technology, and IntelliWeave. In conjunction with this collaboration, Navitas announced its latest 12kW power supply unit (PSU) for high-power rack densities of 120kW.”

The question is what to do with the shares now. We definitely think that it makes sense to take some money out given the strong run the shares had while most of the revenues will only come in 2027. Thus, it’s likely that the shares will trade lower again in the interim at some stage. There is also a substantial risk that the company runs into balance sheet problems should we go into a macro downturn. The current cash burn is around $50 million per annum while the company has close to $70 million in net cash on its balance sheet. Thus, in a prolonged macro downturn, things could get ugly and the company would have to raise debt or equity.

The bull case would be that current sales should be closer towards a cyclical bottom—we’ll discuss the cycle in mature semis later—assuming the global macro environment stays healthy. Secondly, it’s likely that due to the high growth in the GaN market, Navitas will be going towards $300 million in revenues later this decade, which would imply the company having a market share of 10-15% in the wider GaN market. Put these revenues on a 7x sales multiple and you can get to a market cap of $2.1 billion later in the decade vs the current market cap of $1.34 billion. These revenue estimates are based on the company’s investor day and from estimates by industry research house Yole, and these were from before this Nvidia announcement. Lastly, the company is also a takeover target given their strong position in the high-growth GaN market.

Thus, given the steep valuation near term, we’re inclined to take profits here at this stage although it’s a name we could enter again at some stage. It’s definitely an interesting story to keep track off.

For premium subscribers, we’ll go through much more names in semis—including Nvidia, AMD, Marvell, BESI, KLA, Entegris, Arm, Intel, Microchip & Texas Instruments.

Nvidia’s Still Large Opportunity

Read more