Small batches of Nvidia’s H200 AI accelerators have begun arriving in China after Washington shifted from an outright ban to case-by-case export licenses. The change followed President Trump’s December 8, 2025 announcement that approved Chinese clients could buy H200 chips if they comply with specific rules. Under the revised framework, exporters must convince US authorities that sales will not reduce supply for American customers, Chinese buyers must follow export regulations and conduct customer verification, and the chips must undergo independent testing in the United States.
Beijing, however, has imposed its own restrictions. Chinese regulators are reportedly allowing companies to fulfill only about 200,000 H200 orders, far below requested volumes, and limiting use to AI training on public data rather than inference or sensitive workloads. ByteDance and Tencent have each received roughly 10,000 H200 processors in recent weeks, while other Chinese technology companies could win approvals for similarly sized shipments. Although US licenses allow purchases of up to 100,000 H200 processors per Chinese technology group, Beijing wants most of those chips to remain outside mainland China and has told companies they can ship H200 processors to Hong Kong. That creates infrastructure challenges because Hong Kong lacks sufficient data-center capacity and power supply to absorb large volumes. A person familiar with the matter told the Financial Times: “It’s a dilemma. Everyone needs the chips but struggles to find a way to use them in Hong Kong. The hope is for the control to loosen up gradually.”
Nvidia is sitting on about 500,000 H200 chips intended largely for Chinese customers, according to people cited by the Financial Times. Lenovo and other partners have started telling Chinese customers they can place orders for H200-based AI servers, but each order still needs approval through China’s National Development and Reform Commission.
The broader market picture remains difficult for Nvidia. In the first quarter of fiscal 2027, Nvidia posted record revenue of $81.6 billion, up 85% year over year, with data-center revenue of $75.2 billion. Still, management’s next-quarter guidance of about $91 billion included no data-center compute sales to China. CEO Jensen Huang said Huawei is “very, very strong” and that Nvidia had largely conceded the Chinese market after years of tightening US restrictions.
TrendForce survey data cited in the reports indicates Chinese-made chips are expected to account for 46% of AI accelerator spending by Chinese executives over the next year, up from about 30% today. Bernstein estimates Nvidia’s share of China’s AI chip market could fall to about 8% by 2026 from nearly 40% a year earlier, while Huawei rises above 50%. China plans to invest about 2 trillion yuan, roughly $294 billion, in data centers over five years, with at least 80% of core technology including chips expected to come from local manufacturers.
A March 2026 CSIS commentary noted that US and allied export controls did slow China’s technological progress but also accelerated Beijing’s semiconductor self-reliance push and created a captive customer base for domestic chipmakers. The Stanford AI Index 2026 found the performance gap between top US and Chinese AI models has narrowed to about 2.7% as of March 2026. Chinese AI developers are increasingly using domestic chips for inference while Nvidia processors remain more common for training. Recent model releases from Moonshot, Alibaba, DeepSeek and Z.ai underline Beijing’s need to balance access to Nvidia hardware with its long-term goal of building a self-sufficient semiconductor ecosystem.
