Nvidia warns on export rules

Nvidia chief executive Jensen Huang said U.S. export curbs on China risk helping Chinese companies build an artificial intelligence stack that no longer depends on American chips or software. (enterpriseai.economictimes.indiatimes.com) Huang made the warning in an interview published April 15 on the Dwarkesh Podcast, where he argued Washington should let Nvidia “win around the world” instead of giving up large markets. He said a shift by Chinese developers from Nvidia’s CUDA software to Huawei’s CANN tools would weaken the U.S. hardware-and-software position in AI. (dwarkesh.com, enterpriseai.economictimes.indiatimes.com) The immediate backdrop is a U.S. licensing rule disclosed by Nvidia on April 9, 2025. The company said Washington now requires a license to export its H20 chips to China, Hong Kong, Macau and certain related parties, and Nvidia said the move could produce about $5.5 billion in charges in its first quarter of fiscal 2026. (sec.gov) AI chips are the processors used to train and run large models, and the software layer on top decides which hardware developers build for. Huang’s argument is that if Chinese labs spend years tuning models for domestic chips, those habits can harden into a separate ecosystem. (dwarkesh.com, enterpriseai.economictimes.indiatimes.com) That debate has been building for more than a year inside Washington. The Commerce Department published a “Framework for Artificial Intelligence Diffusion” on January 15, 2025, then said on May 12, 2025 that it was rescinding that Biden-era rule while still tightening some semiconductor controls and warning industry about Huawei Ascend chips. (federalregister.gov, media.bis.gov) Huang has been making the same case in public since at least May 21, 2025, when he called the export controls a “failure” at Computex in Taipei. He said then that China could represent a $50 billion AI market in 2026 and argued that, if U.S. suppliers are blocked, Chinese customers will buy from domestic rivals instead. (scmp.com) Nvidia’s own filings show why the company keeps pressing the issue. In its annual report for the year ended January 26, 2025, Nvidia said data-center revenue in China grew in fiscal 2025 but remained well below the share seen before export controls imposed in October 2023. (sec.gov) U.S. officials frame the policy differently. Nvidia’s April 2025 filing said the license requirement was meant to address the risk that covered products could be used in or diverted to a supercomputer in China, and the Commerce Department said its later actions were designed to keep advanced AI technology out of adversaries’ hands. (sec.gov, media.bis.gov) For cloud providers and model companies, the practical issue is portability. If customers may need to run one model on Nvidia hardware in one market and on other accelerators in another, the value shifts toward serving software, compression methods and quantization techniques that can move models across different chips with less rewriting. (dwarkesh.com, enterpriseai.economictimes.indiatimes.com) Huang’s warning is not that Nvidia is losing today’s market alone. It is that export rules meant to slow China could also finance the engineering work needed to make American AI infrastructure easier to replace. (enterpriseai.economictimes.indiatimes.com, scmp.com)