Nvidia pushes back

Nvidia is openly warning Washington that tighter China chip bans could hurt the U.S. companies the rules are meant to protect. (economictimes.indiatimes.com) Chief executive Jensen Huang said on a podcast posted in mid-April 2026 that pushing Chinese developers onto Huawei chips would be “a horrible outcome for the United States” and argued the U.S. should keep selling into China. (msn.com) Nvidia is also changing the products themselves. Its China-only GeForce RTX 5090D keeps the RTX 5090’s 21,760 CUDA cores and 32 gigabytes of GDDR7 memory, but cuts artificial intelligence throughput from 3,352 AI TOPS to 2,375. (videocardz.com) That split is the point: gaming speed stays about 98% to 100% of the standard card, while machine-learning features are dialed back to stay under export thresholds. (kad8.com) The fight sits on top of a moving U.S. rulebook. On April 9, 2025, Nvidia disclosed that Washington would require a license for exports of its H20 chip to China, Hong Kong, Macau, and other covered destinations. (sec.gov) Nvidia said that H20 restriction would force a $5.5 billion charge in its fiscal first quarter of 2026, a sign that China was still a meaningful market even after earlier controls. (cnbc.com) Washington’s policy has also shifted between administrations. The Commerce Department said on May 13, 2025 that it was rescinding the Biden-era Artificial Intelligence Diffusion Rule while adding other semiconductor controls. (bis.gov) Nvidia’s own filings show how the company is threading that needle. In its annual report for the year ended January 26, 2025, Nvidia said China data-center revenue grew, but remained “well below” the share it had before the October 2023 export controls. (publicnow.com) The result is a new kind of chip launch: not faster versus slower, but allowed versus restricted. Huang is arguing for access to China while Nvidia’s engineers build around the limits already in place. (247wallst.com)