Can controls meaningfully delay competitors, or will they accelerate substitution and damage domestic firms and allies?
The October 2022 controls were the most aggressive technology denial since the Cold War. Three years later, DeepSeek and Huawei have narrowed the gap anyway — and analysts disagree about whether the controls bought time or mostly sped up Chinese substitution.
View on the map → · Open in Browse →
Both sides got stronger evidence this summer. For the controls: DeepSeek's CEO privately saying chip access is what holds China back, and on-the-ground reporting of compute shortages. Against them: a one-year swing of open-router inference from ~70% American to ~70% Chinese models, and Congress finding the real chokepoint is tool servicing, not tools. The 2026 supply picture is now quantified — Grunewald's accounting puts China's acquisition near 320,000 B300-equivalents, over half legally imported H200s — and the substitution clock has a date: Huawei meeting half of domestic demand by 2028, full self-sufficiency thinkable by 2030.
Compute denial slowed the frontier race and remains worth its costs.
Huawei's problem is still insufficient capacity
Leaked investor remarks: Liang Wenfeng says chip access is what holds China back — he wanted 200,000 Huawei 950s and could get 16,000.
One estimate suggests that OpenAI has about as much compute as the entire Chinese AI industry.
On-the-ground reporting from Chinese AI companies that the controls currently bind.
China is building a parallel stack; the controls' premises have aged.
Huawei cannot use global-standard tools or fabrication processes, it is forced to build an entirely separate, vertically integrated ecosystem.
Argues the controls are pushing China to build a self-contained parallel stack — substitution accelerated rather than progress delayed.
the policy itself is based on a hardware-centric approach to AI capabilities that the technology has since outgrown
Argues the controls rest on outdated hardware-centric assumptions and should not be traded as diplomatic leverage.
Tokens on OpenRouter were ~70% American in June 2025, [and are] ~30% in June 2026
Within one year the majority of open-router inference shifted from American to Chinese models.
By our estimates, Huawei could meet half of China's domestic compute demand as soon as 2028, with full self-sufficiency becoming thinkable by 2030.
Puts a timeline on the substitution argument: controls buy delay but Huawei's ramp makes Chinese self-sufficiency thinkable by 2030, so US leverage from chip denial is a wasting asset.
Compute stocks and shifting policy, measured.
Both estimates were roughly 2.8 million H100e, and the convergence of estimates suggests we may be on the right track.
Bottom-up estimate of China's AI compute stock — an empirical baseline for judging whether the controls bind.
Documents the January 2026 reopening of H200 sales alongside congressional moves against remote cloud access to compute.
Without the expertise of ASML and TEL engineers to replenish parts and fix tools, China's most advanced fabs may grind to a halt.
Tool servicing, not tools, identified as the real chokepoint in Congress's codification attempt.
Over half of the compute will likely be legally imported NVIDIA H200s, but other sources—domestic production, proxy fabrication, and smuggling—matter too
Quantifies China's 2026 compute acquisition (~320k B300-equivalents) across legal H200 imports, Huawei production, smuggling, and proxy fabrication — implying the controls' design, not evasion, is now the binding variable.