Scottish Mortgage is Baillie Gifford's flagship investment trust (founded 1909, LSE ticker SMT), known for its maximalist growth style — long-term stakes in Tesla, Amazon and ASML plus bold allocations to private companies like SpaceX and ByteDance. It is the UK retail investor's flagship vehicle for global disruptive growth.
This article focuses on ASML's central role in AI chip manufacturing. CEO Christophe Fouquet noted that without EUV (extreme ultraviolet lithography), there would be no 3nm or 2nm nodes, and therefore no AI. ASML's EUV machines contain more than 700,000 components, making them the most complex equip
Christophe Fouquet, an 18-year company veteran who took over as ASML CEO two years ago, came up through the applications engineering ranks. His conversation partner is Lawrence Burns, an investment manager at Scottish Mortgage. The podcast introduction notes that under Fouquet, ASML has become the most valuable company in European history; the new-generation EUV system — as heavy as a blue whale and a decade in the making — has begun shipping. The throughline: how EUV has become the physical infrastructure of AI compute, and how deep the commercial moat is at this position. The most consequential judgment in the entire episode comes from Burns: ASML is, in essence, a monopolist levying a "neutral royalty" on AI demand — whether OpenAI, Anthropic, or Google wins at the model layer, the chips needed for training and inference cannot bypass its lithography machines.
Fouquet's technical judgment leaves no room for doubt: without EUV, there is no 3nm node and no 2nm node, and therefore no AI. The logic chain: AI compute power comes from the number of transistors on a chip, and transistor patterns are "printed" onto wafers by lithography machines with nanometer-level precision. The podcast introduction provides a sense of the complexity: inside ASML's newest machine, three successive laser pulses bombard pollen-sized tin droplets 60,000 times per second, producing a kind of light that normally exists only in outer space.
Burns places this judgment within an investment framework. In his view, the AI value chain has three layers: the application layer (autonomous driving, AI healthcare, and other vertical scenarios), the model layer (Anthropic, OpenAI, Google Gemini), and the supply chain layer (ASML's lithography machines). The risk distribution is completely different — the application layer's winners are the hardest to foresee, and the market's labels for "who benefits from AI" shift extremely quickly; the model layer's frontier players have converged to three, but whether it will be single-winner or multi-winner is unknown; the supply chain layer, meanwhile, presents a "bottleneck with an almost unbelievably strong competitive position," corresponding to very high gross margins and strong free cash flow. He said: "Whatever happens at the application or model layer... whether it's OpenAI, Anthropic or Google, ASML wins... effectively an agnostic royalty on AI demand, and more broadly, compute demand." (In other words: no matter what happens at the application or model layer — whether OpenAI, Anthropic, or Google wins — ASML wins; it is, in substance, levying a neutral royalty on AI demand and, more broadly, on compute demand.)
Extrapolation: Fouquet notes that the industry is currently in a period of intensive investment in AI infrastructure, and the investment must continue — NVIDIA is already developing products two or three generations ahead; compute facilities built today will need upgrades in two to three years, and again in five to six years. This means lithography equipment demand is not a single wave but an unbroken stream. At the same time, he offers an honest "don't know": how much additional chip demand end-use will create once models truly enter consumer and industrial products — "the jury is still out" — and clarity will emerge over the coming months. This is the guest's own acknowledged uncertainty, not a conclusion.
Fouquet breaks down the claim that "Moore's Law is dead": it depends on which version — the cost version died long ago; the density version is not only alive but accelerating in the AI era. The cost version says chip costs halve every two years; that one has been dead for years, because technology costs have kept rising. The density version says transistor density doubles every two years; it has not stopped in 40 years, and certainly has not stopped today. He cites the order-of-magnitude shift: the performance gain of NVIDIA's most advanced chips is not 2x every two years, but 16x every two years — "you could say NVIDIA has put it on steroids" (in other words: you could say NVIDIA has put Moore's Law on steroids).
He also cites Moore's own original definition: Moore's Law was, from the start, a two-legged framework of "miniaturization + integration" — miniaturization uses lithography to pack more transistors into a unit area; integration stacks more lithography layers. Burns adds the technical structure: an advanced chip is built from dozens to hundreds of stacked lithography layers, and only the most critical few layers require EUV-grade nanometer precision; the large remainder use DUV. EUV is a minority by units shipped; DUV is the "workhorse of lithography," and many of the DUV machines ASML shipped forty years ago are still in service. Fouquet reveals that ASML's internal roadmap is already planned out to 2035–2040.
The flip side of accelerating demand is the supply barrier. Fouquet's moat logic is a war of movement: to maintain their lead, customers require ASML to keep investing heavily in EUV — today ASML's R&D spending on EUV is higher than it was a decade ago, when the technology was being taken from 0 to 1. As long as the leader keeps moving at the same speed or faster, the chaser faces a moving target. Only when you stop does someone begin to dream of catching you. Historical data supports "why they can't catch up": the extreme ultraviolet light source took 20 years of R&D to become commercially viable; Burns relays the industry consensus that hand an EUV machine to a competitor or even a country, and reverse engineering would take roughly 10 to 15 years — during which time ASML would still be advancing the next generation. Order-of-magnitude details: an EUV system has over 700,000 components; the latest-generation optical system is about 1,000 times more precise than the Hubble Space Telescope; the wafer stage carrying the chip pattern accelerates at 32g, while fighter pilots withstand only 4–5g — anyone sitting on it would be killed by the acceleration. The optical mirrors come from a supplier in which ASML holds a stake, keeping a critical supply chain link in its own hands.
Customer lock-in is the softer side of this moat: Fouquet says the two sides begin envisioning the 5-10-15-year future together before designing the next-generation tool, forming a "chain of trust, chain of knowledge, chain of support." As an applications engineer, he once slept in TSMC's fab repairing machines, and during the pandemic he insisted on flying to Taiwan to solve problems face to face.
China is the only disruptive variable discussed openly. Last year China contributed 33% of ASML's revenue, even though the Dutch government, at the request of the United States, has banned EUV sales to China; in April 2025, U.S. lawmakers from both parties proposed further restricting immersion DUV exports to China. Fouquet's assessment of the current policy situation is "still messy" — the discussion has not yet reached a balance — and he believes overly aggressive restrictions would invite competition: "if you are too aggressive in restricting, you are extremely inviting in creating competition somewhere else" (in other words: if restrictions are too aggressive, that is tantamount to inviting competition elsewhere). Burns, as an investor, offers a hedged assessment of the same risk: China's EUV R&D has been compared within the industry to "Manhattan Project" level, and he is unwilling to bet indefinitely against Chinese innovation; but even if China produces an EUV machine, pressure from the U.S. government would most likely lead TSMC, Samsung, and Intel all to refuse to adopt it, confining the threat to China-region revenue — and this revenue "was already declining; the market has already seen the risk." A caveat is warranted: in the export-control passage, Fouquet is also playing the role of policy lobbyist, using the argument that "overly aggressive restrictions breed competition" to contend that controls should not be too tight; readers should factor in the industry side's interest position.
Burns explicitly opposes "timing ASML as a cyclical stock" — this is precisely why Scottish Mortgage was slow to build a position in its early years. They first invested in ASML 13–14 years ago, but for several years before that they repeatedly tried to "time the cycle," always wanting to wait for a better entry point. He draws two lessons: first, fixating on the cycle makes you "miss the forest for the trees" — working out the long-term structural growth of demand is the real trend; second, timing is itself a new source of risk — the cost of mistiming the cycle can exceed the cycle's own volatility. The realization that ultimately "liberated" them was: acknowledge that this is a cyclical industry, but do not let the cycle dictate decisions — do not greedily exit at the top, do not fearfully add at the bottom; focus instead on long-term structural demand. The current moment happens to be inside "the largest structural demand driver in history," namely AI.
Fouquet's judgment on cyclicality is not inconsistent with this: the industry's nature is to pile in collectively when it sees a big opportunity, then pause collectively — so the cycle is most likely still there, and geopolitics could become a new cycle-maker. But he also acknowledges that over the past few years the industry has been "unusually insensitive" to geopolitical shocks. On the demand side, he stays candid about how many additional chips AI terminal applications will consume: the jury is still out — this is the strongest confirmation he can offer, rather than overpromising.
In Burns' framework, uncertainty is not an obstacle but a precondition. When they first invested in ASML, they could not determine from academic journals or industry interviews whether EUV could be mass-produced — the strongest signal at that point came from 2012: Intel, TSMC, and Samsung, the three largest customers, injected €4bn into ASML, pledging to fund EUV R&D. Burns says this was the industry's biggest players voting with real money: this is the only path that can unlock the bottleneck, and they would put up the money to make it happen. Combined with the asymmetry of "significant long-term upside if it succeeds, controllable risk if it fails," the bet was worth taking. He distills it to one sentence: truly great investments carry uncertainty — sometimes they succeed, sometimes they don't — but once these companies truly break out, the returns can be remarkable.
| Entity | Guest Stance | Key Data |
|---|---|---|
| ASML | Bullish (core holding) | EUV contains over 700,000 components; 2025 sales forecast raised to €36–40bn; China contributed 33% of last year's revenue |
| NVIDIA | Neutral (no position call; repeatedly cited as the demand engine) | Most advanced chip performance improves 16x every two years; already developing products two or three generations ahead |
| TSMC / Intel / Samsung | Neutral (customer and signal roles, not position calls) | Jointly injected €4bn into ASML in 2012 to fund EUV R&D |
| Anthropic / OpenAI / Google Gemini | Neutral (model-layer outcome undecided; Anthropic is a portfolio holding) | Frontier model-layer players have converged to about 3 |
| Canon / Nikon | Neutral (historically exited EUV R&D) | — |
| xAI / Meta | Neutral (Scottish Mortgage AI portfolio holdings, cited as demand-side evidence) | — |
| China EUV self-sufficiency (unnamed) | Risk alert | Described as "Manhattan Project"-level; corresponds to China's 33% share of ASML revenue |
1. Burns' "neutral royalty" thesis: whether the model layer or the application layer wins or loses does not affect ASML's ability to collect — the EUV monopoly makes ASML a structural rent collector on AI compute demand.
2. Fouquet's Moore's Law dichotomy: what died is "chip costs halving every two years"; what lives is "transistor density doubling every two years." NVIDIA has pushed the latter from 2x every two years to 16x every two years — equivalent to putting Moore's Law "on steroids."
3. Burns' asymmetric bet view: whether EUV would succeed could not have been known in advance, but the customers' €4bn injection was the strongest signal the industry could give; truly great investments carry uncertainty, and the basis for the bet is that "the upside if it succeeds is sufficient to compensate for the risk."