ASML's €5B Profit Quarter Reveals AI Chip Bottleneck

ASML Is Printing Money. The Problem Is Everyone Else Wants More.

ASML made €5 billion in profit in a single quarter. That’s more than the entire annual net income of most publicly traded companies on Earth. Revenue was €11.7 billion, up from €9.3 billion a quarter earlier, and the company raised its full-year forecast again — this time to €44 billion or more in sales for 2026.

The story most people are walking away with after reading these numbers is: ASML is incredible. That’s not wrong. But the more important story — the one that should worry governments, investors, and anyone betting on artificial intelligence happening at scale — is far less celebratory.

ASML just told its customers that even with a supply expansion of 30 per cent planned for 2028, they won’t get enough machines. Not enough. The company is now investigating whether it can push that capacity increase beyond 30 per cent. Customers are placing orders for 2029 and 2030. The bottleneck isn’t closing. It’s accelerating.

The Numbers Behind the Bottleneck

Let’s look at what actually happened in Q3, because the raw figures tell a story that a press release smooths over.

New lithography systems sold jumped from 86 in Q2 to 113 in Q3. That’s a 31 per cent quarterly increase — not the kind of trajectory you see in normal industrial companies. Used systems also climbed, from five to eight, which sounds trivial until you remember that ASML’s entire installed base of roughly 7,000 machines needs constant maintenance, optimization, and upgrade work. Installed Base Management revenue alone came in at €3.3 billion, above guidance, driven by customers upgrading existing systems to squeeze more productivity out of silicon they already own.

The gross margin of 57.5 per cent is extraordinary for a hardware company. For context, that’s closer to what Apple extracts on its services business than what any traditional manufacturer achieves. ASML is effectively charging a premium for something the world cannot build without it.

Cash on hand nearly doubled quarter over quarter, from €7.6 billion to €12.7 billion. ASML is sitting on a war chest and it knows it.

Why 2028 Is Not Close Enough

Christophe Fouquet, ASML’s CEO, did something unusual in the earnings call. He didn’t celebrate. He looked straight at the camera and said the company’s current supply plan for 2027 and 2028 will not meet demand. That’s a remarkably direct statement from someone who, by professional convention, would prefer to say things are “challenging but manageable.”

The reason is structural. ASML’s extreme ultraviolet lithography machines — the ones required to print the circuits that power AI training clusters — are not widgets. Each EUV system takes roughly a year to build, involves over 100,000 components sourced from across a fragmented supply chain, and requires calibration work that only a handful of engineers worldwide can perform. There is no factory in China, no competitor in Taiwan, no overnight solution.

ASML’s customer order book for 2029 and 2030 is growing faster than the company’s ability to deliver in 2027 and 2028. That gap is what Fouquet is describing when he says they are “investigating ways to increase capacity beyond the 30 per cent increase previously communicated.”

Thirty per cent sounds like a lot. In semiconductor terms, where capacity builds take two to three years from announcement to first silicon, it’s barely a dent.

The AI Infrastructure Paradox

Here’s what the ASML numbers reveal about the broader AI economy: the demand side is moving faster than the supply side ever could, and that mismatch is going to define investment outcomes for the next half-decade.

Every major cloud provider and AI lab is building training clusters that require exponentially more compute than the previous generation. Google, Meta, Amazon, Microsoft — they’re all spending tens of billions on H100 and H200 equivalents, next-generation blackwell chips, and whatever comes after. None of that silicon gets printed without an ASML EUV machine. None of it happens fast.

The result is a paradox. AI promises are growing louder and more expensive, but the physical constraint — the number of EUV tools in the ground and being built — is growing slower than anyone’s projections assumed. ASML isn’t the bottleneck because it’s inefficient. It’s the bottleneck because it exists at the narrowest point in the entire global semiconductor supply chain.

Fouquet noted that customers are increasing their short- and medium-term equipment demand forecasts, and that this is “driven primarily by their views on the size and duration of AI-related investments.” Translation: the demand side doesn’t trust the supply side to keep up, so they’re over-ordering to hedge. That’s rational behaviour. It also means the shortfall is going to feel worse than the numbers suggest, because every customer is already planning for scarcity.

Who Wins, Who Loses

The winners are clear. ASML’s shareholders have seen returns that dwarf the Nasdaq over the past three years. Its suppliers — particularly those providing optics, lasers, and precision mechanics — are benefiting from a sustained investment boom. Intel, TSMC, Samsung, and the major cloud providers all benefit from higher wafer output enabled by ASML’s productivity upgrades.

The losers are harder to identify in a quarterly report but unmistakable in hindsight. Startups that need access to cutting-edge compute and can’t secure allocation will fall behind incumbents. Smaller countries hoping to build domestic AI capabilities without navigating the ASML queue will watch their ambitions stall. Investors who priced in near-term supply solutions that don’t exist are already facing earnings disappointments.

There is also a geopolitical dimension that deserves attention. ASML’s EUV technology is subject to US export controls that restrict sales to China. That policy has created a secondary market where Chinese buyers are paying premiums for older DUV systems, while ASML itself loses revenue it would otherwise capture. The restrictions aren’t likely to loosen, which means the company is operating with one hand tied behind its back even as demand explodes on every other front.

What Comes Next

ASML expects Q4 sales between €14.2 billion and €15.2 billion, with gross margin around 57 per cent. Full-year 2026 revenue of €44 billion to €45 billion is now the base case, not the optimistic scenario. The company is investing heavily in capacity expansion — the €1.2 billion R&D figure represents continued development of next-generation high-numerical-aperture EUV tools that will be required for the node generations powering third- and fourth-generation AI models.

But the real takeaway is simpler than any of these numbers. The world wants AI at a pace that the physical infrastructure cannot match, and ASML is the closest thing there is to a single point of failure in that equation. The company is doing everything it reasonably can. It still won’t be enough.

That’s not a problem for ASML to solve. It’s a problem for everyone else to plan around.