BEIJING/SHANGHAI/NEW YORK – Nvidia (NVDA.O) has placed orders for 300,000 H20 AI chipsets with Taiwan’s TSMC (2330.TW), according to two sources familiar with the matter. One source cited robust demand from China as a key factor in Nvidia’s decision to ramp up production, moving beyond its existing stockpile of 600,000 to 700,000 chips. The orders come after the Trump administration recently allowed Nvidia to resume H20 GPU sales to China, reversing an April ban that had blocked shipments over national security concerns.
The H20, a downgraded version of Nvidia’s flagship AI chips, was specifically developed for the Chinese market following U.S. export restrictions in late 2023. While it lacks the computing power of Nvidia’s H100 or Blackwell series, strong demand has prompted the company to secure additional supply. Nvidia CEO Jensen Huang, during a recent Beijing visit, indicated that production restarts would depend on order volumes, warning that reviving the supply chain could take nine months.
However, uncertainties remain. Nvidia still requires U.S. export licenses to ship the H20, and the Commerce Department has yet to approve them, according to sources. The company has reportedly asked Chinese buyers to submit detailed order forecasts, suggesting cautious supply chain planning. Meanwhile, U.S. lawmakers have criticized the decision to resume H20 sales, fearing it could undermine America’s AI edge, while industry players argue that maintaining Chinese market share is crucial to prevent clients from switching to rivals like Huawei.
The H20’s approval was tied to U.S.-China negotiations over rare earth magnet exports, highlighting the chip’s geopolitical significance. Analysts note that while Nvidia seeks to balance regulatory constraints with market demands, its ability to sustain Chinese interest in its ecosystem—including software tools—could be pivotal in fending off competition. TSMC and Nvidia declined to comment, while the U.S. Commerce Department did not immediately respond to requests for clarification.