One series worth tracking next to the ~90 immersion tools a year is ASML's own China mix. On its Oct 14 call, ASML put China at roughly 20% of 2026 total revenue, down from 29% in 2024 (with a 49% peak in Q2 2024) and about 33% in 2025. That is a revenue share, not an immersion unit count, so it can't confirm the import figure directly, but it is the most regular public read on whether the flow is already slowing before any MATCH Act passes. If China's share climbs back toward 30% in ASML's 2027 guidance or quarterly reports, the hole is still widening; if it holds near 20%, part of the gap may be narrowing through post-stockpiling normalization rather than legislation. One caveat cuts the other way: the share also falls when EUV sales elsewhere grow faster, so a lower percentage alone doesn't mean fewer tools reached China.
In 2006 as part of an US NSF delegation, I visited IMEC in Belgium https://www.imec-int.com/en - this is a consortium of semi-conductor companies doing advanced research with companies from all over the world. ASML was a big part, as were Intel and IBM. I wonder if the Chinese companies are there now?
one thing from here the chip-count math may miss: a lot of money is going into squeezing mixed fleets, not just buying more chips. infinigence raised ¥2.3bn this month for a layer that runs zhipu, moonshot, deepseek and others across 16 different chip types, claiming inference cost fell 10x in a year. does the model treat that kind of software efficiency as fixed, or could it narrow the gap on its own?
One series worth tracking next to the ~90 immersion tools a year is ASML's own China mix. On its Oct 14 call, ASML put China at roughly 20% of 2026 total revenue, down from 29% in 2024 (with a 49% peak in Q2 2024) and about 33% in 2025. That is a revenue share, not an immersion unit count, so it can't confirm the import figure directly, but it is the most regular public read on whether the flow is already slowing before any MATCH Act passes. If China's share climbs back toward 30% in ASML's 2027 guidance or quarterly reports, the hole is still widening; if it holds near 20%, part of the gap may be narrowing through post-stockpiling normalization rather than legislation. One caveat cuts the other way: the share also falls when EUV sales elsewhere grow faster, so a lower percentage alone doesn't mean fewer tools reached China.
In 2006 as part of an US NSF delegation, I visited IMEC in Belgium https://www.imec-int.com/en - this is a consortium of semi-conductor companies doing advanced research with companies from all over the world. ASML was a big part, as were Intel and IBM. I wonder if the Chinese companies are there now?
one thing from here the chip-count math may miss: a lot of money is going into squeezing mixed fleets, not just buying more chips. infinigence raised ¥2.3bn this month for a layer that runs zhipu, moonshot, deepseek and others across 16 different chip types, claiming inference cost fell 10x in a year. does the model treat that kind of software efficiency as fixed, or could it narrow the gap on its own?