What I find interesting is how a very specific technology can become a source of strategic leverage between major powers. We often talk about the US-China AI competition in terms of models and computing power, but the machinery needed to manufacture advanced chips may be just as important. The difficult question is how long export controls can preserve that advantage before they accelerate the effort to develop alternatives.
The installed-base point makes the case for periodic review especially important: the effect of a control changes as tools accumulate, substitutes improve and diversion routes emerge. I would want the policy to state what evidence would justify tightening, narrowing or retiring it, rather than treating today’s chokepoint as a permanent category. That would also give allied exporters a clearer basis for coordination. I discuss that problem of keeping security measures tied to a defined risk in this essay on selective de-risking: https://vandanagyanchandanijglsjgu.substack.com/p/selective-de-risking-without-decoupling
The servicing restriction may matter more than the hardware controls themselves. Alberto’s point is key: immersion scanners need ongoing ASML field engineers, spare parts, and software updates to sustain throughput and maintain yield. Once Chinese fabs can no longer call in vendor support, their installed DUV fleet begins degrading—and that degradation compounds quickly at high utilisation rates. The question worth tracking is whether the effective service cut-off—not the nominal hardware ban date—is the real operational cliff for Chinese fabs, and on what timeline existing capacity falls meaningfully below current production levels without regular ASML maintenance.
A national compute shortage is not necessarily a military compute shortage. If Beijing gives strategic users priority, commercial firms may absorb most of the adjustment. Controls could then impose a substantial productivity cost without an equally large reduction in military access. Scarcity has a queue, and governments can rearrange it. Which outcome should define success here: slowing frontier capability, constraining military deployment, or raising the economic cost of sustaining both?
The doubling of the installed base after 2023 is partly a design lesson. Any control that is announced but phased in invites front-loading. China made up close to half of ASML's system sales in early 2024, which is what you'd expect when buyers can see a door closing.
That leaves a question for the report's cap of about ten million chips a year. How much of it depends on the existing fleet getting serviced? Immersion scanners need vendor field engineers, spare parts and software updates to hold yield and throughput. If they can't get those, the installed base slowly loses capacity even with no new shipments. If they can, the stockpile keeps its value. Limits on servicing could matter as much as limits on new tools.
A supplier-side data point on the timing question: on ASML's Q3 call on Oct 14, the CFO kept China at about 20% of 2026 total net sales and said demand there remains very strong for 2027 and beyond. Asked about market share in China, the CEO declined to give a number but said "you can imagine it's extremely high, especially on immersion." Separately, ASML said in July it plans to lift DUV immersion capacity about 30% in 2027 and is studying another 30% for 2028. It didn't break out how much of that goes to China, but the pool of immersion tools that could ship under current rules keeps growing while the MATCH Act debate runs.
What I find interesting is how a very specific technology can become a source of strategic leverage between major powers. We often talk about the US-China AI competition in terms of models and computing power, but the machinery needed to manufacture advanced chips may be just as important. The difficult question is how long export controls can preserve that advantage before they accelerate the effort to develop alternatives.
The installed-base point makes the case for periodic review especially important: the effect of a control changes as tools accumulate, substitutes improve and diversion routes emerge. I would want the policy to state what evidence would justify tightening, narrowing or retiring it, rather than treating today’s chokepoint as a permanent category. That would also give allied exporters a clearer basis for coordination. I discuss that problem of keeping security measures tied to a defined risk in this essay on selective de-risking: https://vandanagyanchandanijglsjgu.substack.com/p/selective-de-risking-without-decoupling
The servicing restriction may matter more than the hardware controls themselves. Alberto’s point is key: immersion scanners need ongoing ASML field engineers, spare parts, and software updates to sustain throughput and maintain yield. Once Chinese fabs can no longer call in vendor support, their installed DUV fleet begins degrading—and that degradation compounds quickly at high utilisation rates. The question worth tracking is whether the effective service cut-off—not the nominal hardware ban date—is the real operational cliff for Chinese fabs, and on what timeline existing capacity falls meaningfully below current production levels without regular ASML maintenance.
A national compute shortage is not necessarily a military compute shortage. If Beijing gives strategic users priority, commercial firms may absorb most of the adjustment. Controls could then impose a substantial productivity cost without an equally large reduction in military access. Scarcity has a queue, and governments can rearrange it. Which outcome should define success here: slowing frontier capability, constraining military deployment, or raising the economic cost of sustaining both?
The 100x gap assumes China can't build DUV tools at home within a decade. What's the strongest evidence that it can't?
The doubling of the installed base after 2023 is partly a design lesson. Any control that is announced but phased in invites front-loading. China made up close to half of ASML's system sales in early 2024, which is what you'd expect when buyers can see a door closing.
That leaves a question for the report's cap of about ten million chips a year. How much of it depends on the existing fleet getting serviced? Immersion scanners need vendor field engineers, spare parts and software updates to hold yield and throughput. If they can't get those, the installed base slowly loses capacity even with no new shipments. If they can, the stockpile keeps its value. Limits on servicing could matter as much as limits on new tools.
A supplier-side data point on the timing question: on ASML's Q3 call on Oct 14, the CFO kept China at about 20% of 2026 total net sales and said demand there remains very strong for 2027 and beyond. Asked about market share in China, the CEO declined to give a number but said "you can imagine it's extremely high, especially on immersion." Separately, ASML said in July it plans to lift DUV immersion capacity about 30% in 2027 and is studying another 30% for 2028. It didn't break out how much of that goes to China, but the pool of immersion tools that could ship under current rules keeps growing while the MATCH Act debate runs.