Nvidia Corp. plans to restart shipments of its H20 artificial intelligence chip to China after securing U.S. government assurances for export approvals, the company said Tuesday, reversing April restrictions that cost it $5.5 billion in inventory write-downs and marking a shift in the Trump administration’s tech trade stance. The move, coming days after CEO Jensen Huang met with President Donald Trump, revives access to a market accounting for 13% of Nvidia’s revenue. That comes to about $17 billion last fiscal year.
The news comes as concerns grow over accelerating China’s domestic AI capabilities and potential military applications amid ongoing U.S.-China tensions. As Huang noted in May: “50% of the world’s AI researchers are Chinese.”
Nvidia, which briefly hit a $4 trillion market valuation last week as the first company to do so, designed the H20 to comply with earlier curbs, but experts warn the sales could boost rivals like Huawei while exposing U.S. firms to supply chain vulnerabilities in global R&D.
Nvidia shares responded positively to the news, climbing $6.39 (3.90%) to close at $170.46 on Tuesday, reflecting investor optimism about the company’s renewed access to the Chinese market. The stock rally suggests Wall Street views the H20 sales resumption as a significant revenue opportunity, potentially offsetting the billions in write-downs Nvidia took earlier this year when the chips were initially restricted.
The H20 processor represents a deliberately weakened version of Nvidia’s flagship H100 and H200 chips, engineered specifically to fall below U.S. export control thresholds that limit computing power and chip interconnect speeds sold to Chinese entities. While the H100 delivers up to 80 gigabytes of high bandwidth memory and peak performance exceeding 2,000 teraflops for AI training workloads, the H20 offers roughly half the computing capability and restricted networking features that prevent large scale clustering.
This performance gap initially made the chip less appealing compared to Nvidia’s flagship products, but Chinese buyers still purchased H20s heavily as they represented the most advanced AI chips legally available. The April export ban forced a shift toward domestic alternatives like Huawei’s Ascend series, though Nvidia maintains the H20 still provides superior software compatibility with existing AI frameworks crucial for enterprise deployments
U.S. officials and industry experts remain divided on the policy shift, with Commerce Secretary Howard Lutnick emphasizing that China will only access Nvidia’s “fourth-best” AI chip to minimize national security risks, while critics warn it could inadvertently bolster Beijing’s military AI programs despite the H20’s downgraded specs. The resumption aligns with signs of thawing U.S.-China trade tensions, including Beijing’s recent easing of rare earth export rules in exchange for chip access, but analysts say it exposes American firms to heightened supply chain dependencies and potential vulnerabilities in collaborative ventures. For Nvidia, the move could recover up to $5 billion in stalled inventory while funding U.S.-based innovation, though it risks accelerating Huawei’s Ascend chips as a viable alternative in China’s domestic market, with the Ascend 910C already claiming to reach 60% of Nvidia’s H100 performance in certain applications.
