The Self-Defeating Logic of AI Chip Controls: How U.S. Export Restrictions Built China’s AI Industry

Gillian Tett

Six years of escalating U.S. chip export controls have produced a result Washington did not intend: a Chinese AI industry that is more domestically self-sufficient, more technically creative, and more commercially competitive than it was before the restrictions began. That is not an argument against export controls. It is a description of what they have actually accomplished, which is materially different from what proponents claimed. YourDailyAnalysis dissects the gap between stated objective and actual outcome as the central analytical problem with how the U.S. frames its chip restriction policy.

Start with the hardware. In 2023, Nvidia held more than 90% of China’s AI chip market. By 2025, Chinese domestic chips had captured roughly 41% of that market, with approximately half of those sales coming from Huawei, according to IDC data. Moore Threads guided that 2025 revenue would be between 1.45 and 1.52 billion yuan, a 231% to 247% year-on-year increase. SMIC, China’s leading foundry, saw revenue projections topping $11 billion in 2026. These are not the numbers of an industry being strangled.

The performance gap is real. Kyle Chan, a fellow at the Brookings Institution who testified before the House Select Committee on U.S.-China Strategic Competition in April 2026, said Chinese domestic chips are likely to remain significantly behind Nvidia’s Blackwell and Rubin GPUs on key performance metrics. Huawei would still produce only roughly 5% of Nvidia’s aggregate AI computing power in 2026, even under very aggressive assumptions. The best U.S. AI chips are currently about five times more powerful than Huawei’s best offerings, and that gap is widening. YourDailyAnalysis catches the paradox: Chinese AI companies are making commercial progress on inferior hardware, which reveals more about how Chinese developers optimize models than about whether export controls are working.

DeepSeek is the clearest illustration. The Chinese AI lab released models in early 2026 that matched or rivaled leading U.S. systems at a fraction of the training cost, using chips that fell below the threshold requiring export licenses. DeepSeek did not close the gap by getting access to better chips. It closed the gap by finding better ways to use worse ones.

The distillation concern adds another dimension. White House chief science and technology adviser Michael Kratsios issued a memo in April 2026 accusing foreign entities principally based in China of engaging in industrial-scale campaigns to extract capabilities from leading U.S. AI systems. If Chinese AI labs can systematically extract capability from U.S. frontier models without acquiring the compute that produced them, then the hardware restriction framework addresses the wrong chokepoint entirely.

There is a counter-argument worth taking seriously. Chan testified that export controls can slow China down in the near term but are unlikely to halt China’s AI progress in the long run. That is not a concession that controls are ineffective – it is a calibration of their purpose. Even a two-year delay in China’s ability to deploy frontier-class AI at scale in military applications is a genuine strategic benefit. YourDailyAnalysis argues the controls remain strategically justified on this narrower basis, and that the political argument around them fails because it overstates their effectiveness.

The January 2026 revision to export control rules changed H200 and AMD MI325X exports from a presumption of denial to case-by-case licensing. So far, that relaxation has produced a trickle rather than a flow. Chinese authorities reportedly warned domestic firms against buying U.S. AI technology unless unavoidable. Beijing’s active encouragement of domestic alternatives has given Chinese AI developers a political reason, not just a technical one, to optimize for Huawei hardware.

The uncomfortable implication is that export controls have accelerated precisely the outcome they were designed to prevent: a Chinese AI supply chain that does not depend on U.S. companies. SMIC at 7nm is not a ceiling that export controls imposed. It is a floor that Chinese engineers built after export controls forced them to try. Your Daily Analysis forecasts that the policy debate will continue to improve the controls’ effectiveness at the margin while the structural dynamic they created continues to run in China’s favor at the base.

Share This Article