Analysis
University of Illinois professor Rakesh Kumar says the US bet on export controls to slow China's chip industry has backfired, and he's not shy about saying he called it: "But it really is 'I told you so,'" he told Fortune in a new interview. Kumar warned in 2022, when Washington first unveiled the controls, that cutting China off from advanced chips would simply force it to build its own supply chain — and he argues the data now backs him up.
The numbers are stark. Nvidia's share of China's data-center chip market fell to 34% in the first half of 2026, and its AI accelerator shipments there dropped 44% year over year, according to IDC data cited by Reuters. Huawei said on October 13 that it has shipped more than 100 million devices powered by its Kirin processors since 2023. In June, delivery platform Meituan said it trained a 1.6 trillion-parameter AI model entirely on domestically made chips — a scale of model many assumed required US silicon just two years ago.
“Huawei said on October 13 that it has shipped more than 100 million devices powered by its Kirin processors since 2023.”
Kumar's read: "Not only have export controls failed in that sense, but they've had the opposite effect." He also argues US policy goals keep moving, saying regulators "keep shifting the goalposts" on what the controls are supposed to achieve. Huawei's publicly stated "Tau Scaling Law" roadmap targets chip density equivalent to 1.4-nanometer by 2031, and CXMT, China's leading memory maker, says it will debut a new memory architecture by year-end that it claims could leapfrog Samsung, SK Hynix and Micron if the claim holds up.
The counterweight Kumar himself raises: China still isn't winning on cost. Most AI inference, he says, continues to run on American chips, and he doesn't expect Chinese providers to beat Western inference pricing in the near term. The real constraint isn't raw chip output, it's that the Dutch government has withheld ASML's EUV lithography export license to China since 2019, pushing fabs like SMIC toward older DUV equipment rather than the newest nodes — a gap domestic chip volume alone doesn't close.
For US chip-equipment and foundry investors, the debate isn't whether China is making chips, it clearly is, but whether cheaper, controls-restricted silicon ever gets good enough to dent demand for Nvidia, TSMC or ASML's most advanced tools. CXMT's promised memory launch by year-end is the nearest concrete test of that question.