Nvidia Halts Production of Chips Targeted at Chinese Market, Citing Compliance Concerns

Nvidia has redirected the wafer‑production capacity it reserved at Taiwan Semiconductor Manufacturing Company (TSMC) from its H200 AI accelerator to the newer Vera Rubin architecture, according to two people with knowledge of the shift. The move signals that the chipmaker no longer expects the H200, an older‑generation processor that was marketed as compliant with U.S. export controls, to generate meaningful sales in China in the near term. The decision follows months of stalled licensing talks with the U.S. State Department and a parallel signal from Beijing that it may restrict imports of the H200 to protect domestic semiconductor champions, the Financial Times reported.

The H200 was slated to become Nvidia’s workhorse for Chinese AI customers after President Donald Trump hinted in December that the United States might relax restrictions on high‑end chips destined for the market. Nvidia responded by ramping up production, with chief executive Jensen Huang telling investors in early January that “demand was very high, and we’ve fired up our supply chain and H200s are flowing through the line.” Yet the State Department subsequently tightened its stance, demanding tighter controls to prevent the chips from being used in applications deemed a national‑security risk. At the same time, Chinese customs officials have kept the H200 on a list of prohibited items unless importers secure a special approval letter, according to the same FT source.

Only about 250,000 H200 units have been fabricated so far, and Nvidia’s chief financial officer Colette Kress confirmed on an earnings call that the company has not yet booked any revenue from Chinese sales of the part. “While small amounts of H200 products for China‑based customers were approved by the U.S. government, we have yet to generate any revenue,” Kress said, adding that the firm does not know whether any imports will be allowed into China. Analysts cited by the FT note that the existing inventory could satisfy any limited orders that eventually receive dual approval, but the uncertainty makes it impractical for Nvidia to keep the line running when its advanced‑node capacity is scarce.

By shifting TSMC capacity to Vera Rubin, Nvidia hopes to accelerate delivery of its flagship AI processor, which is already in strong demand from U.S. AI leaders such as OpenAI and Google. The FT quoted an insider who said the reallocation “could in a way accelerate the Vera Rubin delivery and roll out.” Vera Rubin, unveiled earlier this year, targets more complex, next‑generation models and commands higher margins than the H200. With the global AI chip market tightening and supply constraints persisting, Nvidia appears to be prioritizing products that can be sold with certainty under the current export regime.

The broader geopolitical backdrop remains fluid. A summit between President Xi Jinping and President Trump scheduled for late March has fueled speculation that a bilateral agreement could ease chip export controls, potentially reopening a path for H200 sales. However, even if an accord were reached, the FT notes that it could take up to three months for Nvidia to re‑tool its supply chain or add capacity to resume H200 production. In the meantime, the company is betting that the Vera Rubin line will sustain its growth trajectory while the H200 remains a sidelined, compliance‑driven contingency.