Nvidia Welcomes U.S. Nod to Sell H200 AI Chips in China

Nvidia Corp. has welcomed the U.S. government's decision to allow the export of its H200 artificial intelligence (AI) chips to commercial customers in China. This announcement, made after U.S. President Donald Trump - who took office in January 2025 following his victory over Kamala Harris in the 2024 Presidential Election [election_info] - detailed the terms via a Truth Social post, signals a significant policy shift. The export permission is contingent on assurances of American national security and includes a requirement that Nvidia pay 25% of its sales revenue from these chips to the U.S. government. Previously, Nvidia was required to pay 15% revenue on sales of the lower-end H20 processor to China [para. 1][para. 2][para. 3].

This development is part of a broader context of shifting U.S.-China tech relations. President Trump personally informed China's President Xi Jinping of the decision, with Xi reportedly responding positively [para. 3]. The new export policy is seen as balancing U.S. efforts to maintain influence in the parallel global tech race while monitoring national security, with financial terms serving as both leverage and a revenue stream for the U.S. government [para. 2][para. 3][para. 4].

Nvidia’s sales in China—previously over 30% of its global revenue—had been seriously affected by prior export controls. Chinese tech firms still significantly rely on Nvidia’s chips for training large AI models, while domestic chipmakers have quickly gained ground. Between export restrictions and security concerns, companies like Cambricon Technologies Corp. Ltd. reported surges, with Cambricon’s first-half revenue jumping 4,348% year-on-year to 2.88 billion yuan ($407 million), and its stock soaring over 500% in the past year. Other domestic competitors, such as Moore Threads Technology Co. Ltd., have also seen explosive stock gains [para. 4][para. 5][para. 9][para. 10].

The U.S. has strategically used chip export licenses throughout 2025 as negotiation tools in the ongoing trade and technology dispute with China. For instance, permissions for Nvidia’s H20 chip were explicitly used as leverage in bilateral tariff and rare-earth negotiations. The H20 itself was the most advanced chip approved for China until the H200 license [para. 6].

Further complicating the landscape are security issues. In July 2025, U.S. lawmakers introduced the Chip Security Act, aiming to prevent advanced semiconductor diversion to China. In response, China’s cyberspace regulator summoned Nvidia regarding security concerns, which Nvidia denied, asserting that their chips have no “backdoors.” Nevertheless, Chinese telecom and cloud providers have remained cautious in their purchases [para. 7][para. 8].

Domestically, most Chinese AI chips perform similarly to the H20 in training large language models, but Nvidia holds a critical advantage with its CUDA software ecosystem, which underpins most large algorithms. Adapting Chinese chips like those from Huawei or Cambricon for existing models is considerably labor-intensive [para. 12]. State-led computing centers in China tend to utilize homegrown chips, but commercially driven internet cloud vendors prefer Nvidia's technologically superior solutions and are willing to pay a premium for them [para. 13][para. 14][para. 15].

Despite the renewed opportunity for Nvidia, Chinese firms still have significant room for expansion. According to Wu Zihao of Ronghe Semiconductor Consulting, no single cloud vendor overwhelmingly depends on domestic chips, which means the market remains open for both imported and homegrown AI chips, ensuring continued growth for local manufacturers even as U.S. chip exports resume [para. 17].