The most important chip company you've never heard of is paying its people to stay

Information Technology 4 min 80 sources

ASML raised its forecast twice this year and is handing staff €20,000 to stick around — a window into where the real bottleneck in the chip boom actually sits.

Key takeaways

ASML, the only maker of the machines behind advanced chips, raised its forecast twice this year and is paying staff €20,000 to stay — a sign the real constraint is keeping skilled people, not selling machines.
The chip boom is straining the whole chain at once: TSMC posted record 77% profit growth, while Google, Amazon-backed startups, and China race to own more of the stack.
Two more breaches landed — the AI hub Hugging Face and a vendor to US hospitals — a reminder that attacks usually arrive through a supplier you depend on.

The loudest story in tech is still artificial intelligence. But the quiet one that decides how far it can go runs through a single factory town in the Netherlands — and this week that company did something that tells you more than any model launch.

The company that makes the machines that make the chips

ASML raised its full-year sales forecast for the second time this year, now expecting €43–45 billion, up from €36–40 billion [14][21]. Order intake, its CEO said, is “extremely strong,” and it plans to add 30% more manufacturing capacity two years running [14][22]. ASML is the only company on earth that builds EUV machines — extreme ultraviolet lithography tools, the room-sized systems that use focused light to etch circuit lines finer than a virus onto silicon [14]. No advanced chip today is made without one.

Then came the stranger move. On Monday ASML confirmed it will hand every eligible employee a €20,000 retention bonus — a conditional stock grant — to stay from 2027 through 2030 [17]. The company employs 44,500 people, more than half in the Netherlands [17]. Read that together: a firm with more orders than it can fill, paying its own staff not to leave. That is the tell. The scarce thing in this boom is not demand, and it is not the machines. It is the people who know how to build and run them.

The demand is real, and it runs through Taiwan too

You can see the pull on the other side of the world. TSMC — the world’s largest contract chipmaker, meaning it manufactures the chips that companies like Nvidia and Apple design — posted a 77% jump in second-quarter profit, a record, far past what analysts expected [3][6]. It is accelerating its Arizona factory buildout and expects “strong, multi-year” demand for AI processors [8][10]. TSMC’s fabs — fabrication plants, the sites where chips are physically made — run on ASML’s machines. So TSMC’s record quarter is, in a real sense, ASML’s order book. The whole chain is straining upward at once.

The giants want to own more of the stack

The scramble isn’t only about buying chips — it’s about owning the parts of the stack that cost the most. Google is designing a custom chip, known internally as “Frozen,” to run its own AI models more efficiently [80][47]. The point is inference — the ongoing work of running a trained model to answer questions, as opposed to the one-time cost of training it. At Google’s scale, inference is the recurring bill, and a specialized chip it owns cuts both the cost and its dependence on Nvidia.

The reach goes further down. Jeff Bezos is backing CuspAI, a British startup that has raised close to half a billion dollars to use AI to discover new materials for chipmaking [19]. On Monday it launched an “AI Materials Foundry,” a coalition of more than 48 firms including Nvidia, Meta, and Hyundai, pooling computing power to find better materials faster [19]. If you build on AI infrastructure, the cost of inference is the number to watch — the giants are all racing to control it.

China’s other front: cheaper, open, and raising fast

The race also has a supply side, and China is pressing it. Within days of each other, Moonshot (with its Kimi model) and Alibaba (with Qwen) unveiled models they claim rival OpenAI and Anthropic at a fraction of the cost — and released them openly, not locked behind a paid interface [11][56]. DeepSeek, another Chinese lab, is raising fresh capital at a valuation reported near $74 billion ahead of a domestic listing [18]. And memory-chip maker CXMT is preparing an $8.6 billion IPO — an initial public offering, its first sale of shares to the public [16]. The contest isn’t only at the frontier of what models can do; it’s over who controls the hardware and capital beneath them.

Elsewhere: the breaches keep landing

Two more this week worth an audit. Hugging Face — the widely used hub where developers share AI models and datasets, effectively the GitHub of AI — confirmed a breach affecting internal datasets and credentials, and urged users to rotate their keys [38]. Separately, a technology vendor relied on by thousands of US hospitals said hackers stole a “significant” amount of data [32]. The pattern in both: the attack came through a supplier. If you ship software, the dependency worth checking is rarely your own code — it’s the third parties your systems quietly lean on.

02 · Lesson · why it matters

The machine you can copy. The people you can't.

The hardest advantage to steal isn't a design or a machine — it's the know-how living in the people who build it, and money can't buy the years that grow it.

A booming company, paying people not to leave

ASML has more orders than it can fill. It just raised its forecast for the second time this year. So why turn around and hand every worker €20,000 to stay put through 2030?

Because its real asset walks out the door every evening. A machine sits in a factory; you can point at it, insure it, put it on a balance sheet. The knowledge that builds and runs that machine can’t be pinned down like that. It lives in people. And people can leave.

What ASML sells — and what it can’t

The EUV machine is one of the most complex objects humans make. Hundreds of thousands of parts. Light bounced off mirrors so smooth that, blown up to the size of a country, the largest bump would be a fraction of a millimetre. It etches lines onto silicon finer than a virus.

But the machine is only the visible tip. The real product is decades of accumulated know-how, spread across tens of thousands of people and a web of suppliers who each perfected one piece. There is no shop that sells that. You cannot order it, and you cannot download it.

The knowledge that can’t be written down

There are two kinds of knowledge, and the difference is the whole story.

One kind is explicit — blueprints, manuals, patents, specifications. It can be written, copied, stolen, or bought. The other kind is tacit — the feel of when a part is seated right, the judgment built from ten thousand small failures, the fix nobody ever wrote down because it lives in a person’s hands. This kind transfers only one way: doing the work, for years, beside someone who already can.

This is why a rival can hold every blueprint and still not build the machine. The secret isn’t locked in a safe. It’s spread through thousands of trained people, and there’s no shortcut to growing more of them.

Why that forces ASML’s hand

Follow the logic and the €20,000 stops looking strange. If the company’s value is its people, then losing them is losing the company. A hot market makes those engineers precious. Rivals circle. Customers who want to build their own capacity would love to hire a few away.

So ASML does the one thing that actually protects the asset: it keeps the humans. The bonus is a quiet confession. It says out loud what no balance sheet can print — our worth is not the machines we ship, it’s the people who know how to make them.

The same rule reaches your desk

Step back and this stops being about chips. Any advantage built from accumulated human know-how is sticky and hard to buy — a surgeon’s hands, a kitchen that has cooked together for years, a team whose shorthand took a decade to grow. It’s why some skills outlast the machines that were supposed to replace them, and why some work can’t be moved offshore by writing a cheque.

Look at your own work the same way. The part of it that could be written into a manual is the part most easily copied, automated, or handed to someone cheaper. The part that only lives in you — the judgment you can’t fully explain, the feel you earned by doing it wrong a hundred times — is the part that’s genuinely yours. You are inside this pattern, not watching it from above.

What no single seat can see

From the outside, ASML looks like a machine company, and its power looks like a patent or a clever trick. It’s neither. It’s thousands of people who each hold one piece of a knowledge no one person carries whole. Not the CEO, not a rival with billions, not a government — no one can see or inventory the entire thing, because it doesn’t sit anywhere you can look.

The most durable advantages are exactly the ones that can’t be counted. That makes them easy to underestimate and hard to buy. It’s worth holding our judgments — about who is ahead, and why — a little more loosely, knowing how much of the real work is invisible to every seat, including our own.

03 · Lab · your turn

Close the Gap

Rehearse why a chipmaking rival can buy its way to a plateau but only reaches the top by patiently growing its own people.

04 · Hope · carry this

The one thing no fortune can rush is a person who has quietly learned to do a hard thing well. In a year of machines that spread everywhere overnight, that makes the patient work of getting genuinely good at something worth more, not less.