Given the sparse data set (ACML, N=1), the upper bound is set too low. It’s entirely possible that it is “never” - there’s no guarantee they will be able to create 7/5 nm lithography tools at all.
1. Not having to explore all the dead ends that ASML explored is surely a huge speedup. Maybe we can even quantify it based on public historical data from ASML?
2. The increasing isolationism of the US might decrease its value for the EU and NL, so the US might have less say in NL export controls in a few years? Then Chinese firms might be able to buy ASML EUV machines again.
3. Knowing that EUV is possible should make it easier for Chinese firms to win huge investments.
But:
1. If the NL export controls might weaken, investors might predict that, and the development of DUV/EUV in China might slow. And do the export control already extent to the next generation High-NA EUV machines? Or other even more speculative technologies?
2. Especially when the first GPU and TPU chips come out built using High-NA EUV, the US could loosen the export controls on the previous generation that all the US firms want to get rid of, thereby crashing the prices in China and the demand for Chinese EUV machines. Sort of how Saudi Arabia has tried to destroy the US fracking market by subsidizing oil. (It might work if Chinese innovators are not prepared for it.)
Nice! I do worry that "size of engineering force" is a significant driver. How many engineers does ASML have? i think we should provide some `sqrt(N)` scaling relative to the size of the Chinese engineering force.
It seems that the Chinese AIs are just distilled versions of US AIs, so export controls might cause them to fall behind. This would slow down their AI assisted R&D. It would also slow down their demand for chips (why make more chips if you're AI is worse?).
Thanks for sharing. I do think you missed one important point, which is ASML doesn't refer to its machines as machines. It refers to them as tools. A machine works out of the box. A tool needs constant refinement. It's not semantic. It's a complete statement of philosophy. I spent months researching ASML, and it's that customer obsession and tool focus that makes all the difference. If you stand back and just compare with machines versus machines, you're missing a philosophical and important perspective.
Given the sparse data set (ACML, N=1), the upper bound is set too low. It’s entirely possible that it is “never” - there’s no guarantee they will be able to create 7/5 nm lithography tools at all.
Such important research! Thank you!
Some notes on speedup:
1. Not having to explore all the dead ends that ASML explored is surely a huge speedup. Maybe we can even quantify it based on public historical data from ASML?
2. The increasing isolationism of the US might decrease its value for the EU and NL, so the US might have less say in NL export controls in a few years? Then Chinese firms might be able to buy ASML EUV machines again.
3. Knowing that EUV is possible should make it easier for Chinese firms to win huge investments.
But:
1. If the NL export controls might weaken, investors might predict that, and the development of DUV/EUV in China might slow. And do the export control already extent to the next generation High-NA EUV machines? Or other even more speculative technologies?
2. Especially when the first GPU and TPU chips come out built using High-NA EUV, the US could loosen the export controls on the previous generation that all the US firms want to get rid of, thereby crashing the prices in China and the demand for Chinese EUV machines. Sort of how Saudi Arabia has tried to destroy the US fracking market by subsidizing oil. (It might work if Chinese innovators are not prepared for it.)
Nice! I do worry that "size of engineering force" is a significant driver. How many engineers does ASML have? i think we should provide some `sqrt(N)` scaling relative to the size of the Chinese engineering force.
Don't forget to take into AI export controls!
It seems that the Chinese AIs are just distilled versions of US AIs, so export controls might cause them to fall behind. This would slow down their AI assisted R&D. It would also slow down their demand for chips (why make more chips if you're AI is worse?).
Thanks for sharing. I do think you missed one important point, which is ASML doesn't refer to its machines as machines. It refers to them as tools. A machine works out of the box. A tool needs constant refinement. It's not semantic. It's a complete statement of philosophy. I spent months researching ASML, and it's that customer obsession and tool focus that makes all the difference. If you stand back and just compare with machines versus machines, you're missing a philosophical and important perspective.
https://sparkmind.capital/podcasts/leaders/asml-monopoly-by-consent/
Yeah
amazing, thanks