Focus: The ASML way covers ASML’s rise to its current dominant position with a de facto monopoly over EUV. I read the book because I wanted to get a better idea of ASML’s technology and culture. The book is interesting, well-paced, and contains colorful stories about key figures and the industry, like how one Intel fab suffered a drop in chip yields because cow farts (i.e. methane gas) from a nearby dairy were drifting into their clean rooms. There are also factoids like how it takes an entire year to fabricate just a single mirror in ASML’s next-gen High NA lithography system and the entire High NA system will cost a fab $400M. Here are a few things that stood out:
One feature of ASML’s culture is its culture of debate and directness, which is both a product of Dutch directness and its longtime technical director, Martin van den Brink:
Technicians at ASML are expected to continually challenge one another, to never blindly accept what someone has said simply on the basis that they hold a higher position in the company. Egos need to be checked at the door, and for good reason… this flatness gives a degree of security and protection essential to the smooth running of such an intense operation. ‘Even as the boss, if you’re completely in the wrong with a decision, you can at least be sure someone will say something about it.’
The directness and flatness means that problems do not stay hidden for long, which is important because EUV is the most complicated machine ever created by people, with hundreds of thousands of parts coming from hundreds of suppliers. In fact, no single person understands how an EUV works end-to-end, according to Frits van Hout, ASML’s former Chief Strategy Officer:
“The people who know how it should be done don’t know why it should be done, and the people who know why it should be done don’t know how it should be done.”
I also did not realize just how long ASML spent developing EUV. It took 16 years from the research beginning in earnest until the first prototype was delivered. Martin van den Brink also takes exception to the idea that US politicians have floated that the ‘US did all the hard work on EUV but the Dutch capitalized on it.’
The development of EUV has been going on for 25 years. [The US] did the first five years, but we’ve done the last twenty.
Given just how many difficult problems ASML had to solve to get EUV to work, it’s not all that surprising that the culture is light on praise, with technical milestones celebrated “often with only a t-shirt or a photo… ‘Easy problems’ are not interesting enough to distinguish yourself as the smartest person in the room.”
Another interesting aspect to ASML is how collaborative it is with its suppliers. ASML has over 5K tier one suppliers and while it will crack the whip in an effort to get suppliers to increase production, it doesn’t pit suppliers against each other:
Because the most complex components of the machine are single-sourced – there’s only one supplier for these crucial parts – a high level of trust is essential. This approach differs from conventional sourcing strategies, which typically pit suppliers against each other to drive down costs and mitigate risk. But ASML’s model has an advantage. The suppliers effectively become co-developers of the machine, and their deep involvement in the material means they are able to solve problems they are not the cause of themselves.
I’ll end with an observation about co-development that I’ve begun to notice as I learn more about the AI infra space. Unlike the supply chain space, which is highly fragmented and more commoditized, the AI infra space is concentrated and differentiated. This means relationships are longer-standing, which creates more opportunities to build trust and much higher returns from co-development.