Nvidia H200 chips reportedly reach China as ByteDance and Tencent get 10,000 each
ByteDance and Tencent have each reportedly received about 10,000 Nvidia H200 AI chips, the first meaningful flow of export-approved H200s into mainland China since late 2025.
ByteDance and Tencent have each reportedly received about 10,000 Nvidia H200 artificial-intelligence chips in recent weeks, the first meaningful movement of export-approved H200s into mainland China since late 2025. Washington cleared the sales to vetted Chinese buyers in December 2025.
The deliveries, described by people familiar with the shipments, mark a cautious restart of high-end US chip flows into a market Nvidia had been largely shut out of. Chinese regulators are reportedly steering companies to route most of their US-licensed allowance — up to 100,000 chips per company — to Hong Kong rather than the mainland, a move that would ease pressure on domestic chipmakers while letting a smaller batch land in China proper.
The split points to Beijing’s competing priorities: access to Nvidia’s hardware now, against a longer-term push to build a homegrown alternative. ByteDance, Alibaba and Tencent have together been approved to buy more than 400,000 H200 chips, with other Chinese firms queued behind them, and every purchase requires case-by-case sign-off from China’s National Development and Reform Commission.
None of this is confirmed by Nvidia or Chinese regulators, and the reported volumes remain a fraction of what these companies would need to close the gap with US cloud providers. The chips are also a generation behind Nvidia’s newest, leaving Chinese buyers with capable but not the latest silicon.
How quickly the remaining approved chips ship — and how many stop in Hong Kong rather than the mainland — will show whether this is a genuine reopening or a controlled trickle.
More news

AWS releases six open-source Hugging Face deployment skills for SageMaker

Google Research releases MilleMiglia logistics benchmark generator

AWS launches AgentCore Runtime V2 with elastic memory and snapshot starts
