ASML CEO Christophe Fouquet: No one is coming for us - BERITAJA

Every clip you usage AI, you are, successful immoderate mini way, depending connected a 42-year-old, 44,000-person Dutch institution that spends €4.5 cardinal each twelvemonth to beforehand its technology.

ASML, headquartered successful the Netherlands, makes the machines that make the chips that make AI possible. More specifically, it makes the only machines successful the world could of printing the microscopic patterns connected silicon wafers that specify the about precocious semiconductors — a process called utmost ultraviolet lithography, aliases EUV. The machines are about the size of a schoolhouse bus, return months to assemble, impact hundreds of suppliers, and costs anyplace from $200 cardinal to upwards of $400 cardinal apiece depending connected the procreation (prices that springiness moreover ASML’s biggest customers pause occasionally).

That monopoly has made ASML the about valuable institution successful Europe, worthy complete $530 billion. And pinch the 4 largest American tech companies — Microsoft, Meta, Amazon and Google — committing much than $600 cardinal successful AI infrastructure spending this twelvemonth alone, request for ASML’s machines has surged to the constituent wherever the institution has openly said the world won’t person capable chips for years.

All that request has besides made ASML a target. Substrate, a San Francisco startup founded by a protégé of Peter Thiel, has raised much than $100 million and been weighted astatine complete $1 cardinal connected the declare that it could build a rival lithography machine. Separately, location person been reports that erstwhile ASML engineers successful China person partially reverse-engineered the technology, a imaginable pinch tremendous geopolitical implications.

Christophe Fouquet, who became ASML’s CEO successful 2024 aft much than a decade astatine the company, sat down pinch this editor connected the rooftop platform of his Beverly Hills edifice Tuesday greeting up of his quality astatine the Milken Institute Global Conference. Dressed successful a bluish suit and achromatic shirt, he was relaxed — moreover erstwhile the speech turned to the rivals.

This question and reply has been lightly edited for magnitude and clarity.

TC: Did you spot the AI detonation coming?

CF: No, not astatine all. We worked very hard, but not pinch the thought that this would come. You went from a conception — thing group thought would yet get — to ChatGPT, which was really the first bully illustration of what AI could do. And now I deliberation we look astatine AI arsenic the adjacent revolution, not only business but societal. Did I spot it coming? No. Sitting successful the mediate of it each day, sometimes we aftermath up successful the greeting and still cheque that what is happening is really happening.

The large mobility everyone has is whether the proviso concatenation could support gait pinch demand. Can it?

The request is specified that the marketplace wide will beryllium supply-limited for rather a bit. Right now, the biggest bottleneck seems to beryllium successful spot manufacturing. We, arsenic an instrumentality supplier, travel our customers, and truthful acold we’ve followed them beautiful good — but we cognize we person to measurement up our full proviso concatenation and capacity. If you talk to the hyperscalers, I deliberation they will show you that for the adjacent two, three, moreover 5 years, they’re not going to get capable chips.

TSMC made news precocious saying your latest machines are excessively expensive. How do you respond?

An EUV system, if you look astatine the price, is going to beryllium much costly than a low-NA system, but the costs of making a wafer pinch this instrumentality connected immoderate precocious layers will beryllium cheaper. We could get 20%, 30% costs reduction.

[Editors note: some machines Fouquet is referring to present are EUV machines — the aforesaid basal technology. NA stands for numerical aperture, a measurement of really finely a instrumentality could attraction ray onto a chip. Low-NA EUV is the existent generation; high-NA EUV is ASML’s newest generation, could of printing moreover finer patterns but carrying a value tag of $350 cardinal aliases much apiece. Fouquet is arguing that moreover though the caller instrumentality costs more, it produces chips much cheaply.]

I get a batch of questions about whether it’s going to beryllium this period aliases adjacent period aliases the period after. And I usually opportunity it doesn’t really matter, because we designed high-NA for the adjacent 10, 20 years. You could spell backmost to the property from 2016, 2017, and you’ll find the aforesaid quotes — low-NA EUV was very pricey. We cognize what happened aft that. The aforesaid will hap pinch high-NA.

There’s a startup called Substrate, backed by Peter Thiel, claiming it could build a rival lithography machine. What do you deliberation of it?

Wanting to person it and having it — that’s still a immense difference. The challenges of lithography are many. Being capable to make an image is simply a starting point, but you request to make that image successful very precocious quantity, astatine very debased cost, astatine precocious speed, and pinch nanometer accuracy. I ever opportunity the only logic ASML could build an EUV instrumentality is because 80% of it already existed, based connected erstwhile knowledge and products built complete time. We had to lick 1 problem — getting EUV ray — and that unsocial took 20 years. When you commencement from scratch, the situation is enormous. I’ve seen a batch of claims. I’ve seen a fewer pictures. But we had our first EUV image 30 years ago, and we still needed 20 much years of difficult activity to move it into a manufacturing system.

What about xLight, a laser startup partly backed by the U.S. government that wants to activity pinch you?

xLight is focusing connected 1 constituent of our EUV instrumentality — the root that creates the light. The root we person could beryllium extended for galore years to come, and we cognize really to standard it. What xLight is doing is simply a caller root that still has to beryllium built and proven. The only mobility is whether it provides a capacity aliases costs advantage complete what we have. I deliberation the assemblage is still out. We are moving pinch them truthful they could show their exertion — we consciousness that’s a work connected our side. But it’s still a very agelong journey.

There are besides reports that erstwhile ASML engineers successful China person reverse-engineered your machines.

To reverse-engineer anything, you first request to person the machine. And location is nary EUV instrumentality successful China — we ne'er shipped immoderate devices there. All the devices we person shipped, we cognize wherever they are. They’re either successful usage pinch customers, and we way those, aliases they’ve been dismantled and came backmost to us. The thought that 1 of our systems is successful China is simply wrong. And because our EUV exertion has ne'er been exported there, we besides person nary group successful China trained connected EUV.

Very early on, erstwhile restrictions came in, we created a complete separation wrong the institution betwixt those who could entree EUV technology, documents and training, and those who cannot. Our squad successful China sits connected the different broadside of that line. The facts constituent to very little, if any, advancement astatine all. It’s difficult for group to judge that because entree to this exertion is truthful important.

On export controls much broadly — Jensen Huang was present past nighttime arguing that companies should waste globally, that much firm gross intends much taxation dollars for a company’s location country. He besides said the important point is to support the champion and latest person to home. Do you agree?

I deliberation he’s wholly right. What he adds — and I deliberation this is what Nvidia has done — is that you could support a technological advantage by maintaining a procreation spread successful what you sell. Nvidia sells a fewer generations back, and that lets them find the equilibrium betwixt still doing business and not handing a beardown competitory advantage to countries wherever you won’t waste the latest. We judge the aforesaid attack should use to our products. Today we vessel devices to China — allowed by export controls — but it’s a instrumentality we first shipped successful 2015. If you use Jensen’s accuracy to our situation, Nvidia is moving pinch about an eight-generation gap. We’re looking astatine 2 aliases three. There’s room for rationalization — uncovering the correct equilibrium betwixt not doing business astatine all, losing a awesome opportunity, and powerfully inviting others to compete pinch you.

How do you measure wherever things guidelines pinch the existent management connected each of this?

There is simply a bully dialogue, which is very important. I deliberation there’s a genuine knowing of what business needs, but there’s still the situation of uncovering the correct equilibrium betwixt each the different voices and interests. The speech is there, and we admit that. I’ve been successful Washington galore times. At slightest the chat is happening. But it’s a very analyzable topic.

You don’t look concerned about anyone short-cutting your technology.

People for illustration to person the top technology, but they thin to hide what it took to build it. It’s been galore years of activity — not only astatine ASML but pinch our suppliers. Many different groups of group solving very difficult problems, and past 1 institution bringing it each together utilizing decades of lithography expertise to move it into a manufacturing system. This is successful nary measurement easy. And I deliberation that’s besides our champion protection. It’s simply what it took to put it together.