[1123581321]

My understanding is that production and installation do not scale since the machines are essentially replications of a successful physics experiment environment. Manufacture is a long, multi-step process requiring technical expertise (which is limited). Then, machines are re-assembled at the customer site--again, carefully and with a large technical staff.

To commit engineering to achieving 10x manufacture rate would probably require giving up the lead on next generation processes.

Personally, I think there must be an end to this. There are steps we are going to learn to more reliably replicate, and reliably reassemble. Machines won't become smaller, but other heavy industries have learned to do faster, de-skilled installations so semiconductors should eventually as well. However, I don't think we'll get the 'good for the world' benefit because by the time every country can make leading edge semis, some other critical, supply-restricted part will become what's contested.

[konschubert]

You might be right.

On the absolute forefront, it has to be artisanal.

But we can learn to scale making the machines that were cutting edge two years ago.

Maybe ASML just doesn’t do this because they leave it to other companies? Either way, somebody should be the one that’s always two years behind but makes 2000 machines a year.

And of course, when you remove a bottleneck, you find another. But that other bottleneck will be wider.

[brnt]

Asml does this, learning how to scale the cutting edge of two years ago. This is what scaling cutting edge physics looks like. Academic groups sometimes take the better part of a decade (or more) to replicate a setup by another group. There are few components you can buy off the shelf in this space: you'll have to understand, design and make them all yourself. Market's way too small to suppory competing suppliers. Takes longer than two years to farm all of that out and create a market of sufficient size. Plus, buyers tend to not want (or pay) for one or two decade old stuff, so that market (for part suppliers) may never quite materialise.

And then, these machine are indeed more like physics experiments than a device. If you compare it with that, ASML is actually quite fast.

And the market knows all of this, hence their valuation.

[konschubert]

I think you might be misunderstanding me. I meant “somebody should scale making those machines that ASML could make two years ago”.

And maybe not two years, but 5, but you catch my vibe here.

[brnt]

But they are! It just takes way longer than two years, both to scale production in house (creating the first takes way longer than than, creating the second in 2 is already fantastic), and to create any supply chain market which others could use to compete. Canon surely has been trying, and failing, because it's hard.