Nvidia AI Chips for China
China Allows Limited Nvidia H200 AI Chip Imports for ByteDance and Tencent
Key Points
Nvidia's H200 chip is specified with 141GB of HBM3e memory and 4.8TB/s memory bandwidth, underscoring its suitability for demanding large-model training and inference tasks.
China’s current import framework allows H200s to enter the mainland in small batches while directing most of the hardware outside mainland China, with Hong Kong treated as an outside-customs area to facilitate deployment.
Nvidia partners, including Lenovo, have resumed selling systems containing H200 chips to select Chinese customers, but such purchases still require regulatory clearance from authorities (e.g., NDRC approval).
Industry observers expect other major Chinese tech firms to acquire similar quantities (around 10,000 chips each) in the near term, signaling a staged rollout rather than a single large import.
Nvidia's H200 AI chips have begun entering mainland China in small batches, with ByteDance and Tencent each receiving about 10,000 units in recent weeks, Financial Times reported. The move marks a significant shift in how the U.S. and China manage the flow of high-performance AI hardware between the two countries.
The H200 packs 141GB of HBM3e memory and delivers 4.8TB/s of memory bandwidth — making it one of the most capable chips available for training and running large AI models. Despite not being Nvidia's newest accelerator, demand for it inside China remains strong, according to Seeking Alpha.
How the U.S. Framework Is Letting Chips Flow
The current U.S. export framework allows H200 chips to enter mainland China in limited quantities. Each Chinese buyer can receive up to 100,000 chips under the arrangement, per TradingKey. ByteDance and Tencent are the first to receive shipments, with each getting roughly 10,000 units.
Most H200 hardware is still being directed outside mainland China. Hong Kong is treated as an outside-customs zone, which makes it easier to deploy chips there. Buyers who want to bring chips into the mainland still need regulatory clearance, including approval from China's NDRC.
Lenovo and Other Partners Resume H200 Sales to Chinese Firms
Nvidia's distribution partners, including Lenovo, have restarted selling H200-based systems to select Chinese customers, according to Market Screener. These sales still require sign-off from regulators before chips can be delivered and used inside China.
Industry observers expect other major Chinese tech firms to receive similar quantities — around 10,000 chips each — in the near term. This points to a staged rollout rather than a single large import event. Nvidia currently holds around 500,000 H200 chips in stock, a surplus built up during past restrictions.
China Buys Foreign Chips While Building Its Own
China is walking a careful line. It is letting in limited batches of Nvidia chips while at the same time pushing hard to grow its own semiconductor industry. Chinese firms like Huawei have been developing domestic AI chips to reduce long-term reliance on U.S. suppliers, per The Edge Malaysia.
The tension is clear. Chinese AI companies want access to the best hardware available. But U.S. export rules keep tightening, especially around Nvidia's newest chips. The H200, based on Nvidia's older Hopper architecture, sits in a gray zone — capable enough to be useful, but not restricted as heavily as newer models.
Data Center Capacity Could Slow How Fast Chips Get Used
Even as chips arrive, there are practical limits. Data center space in both mainland China and Hong Kong is constrained. Observers note this could slow how quickly ByteDance, Tencent, and others can actually put the H200s to work, according to Market Screener.
The broader picture is one of cautious progress on both sides. The U.S. is allowing limited chip flows without opening the floodgates. China is accepting what it can get while racing to close the technology gap. How this balance shifts in the coming months will shape the global AI race.
Sentiment
Mostly PositivePublishers
33
Articles
106
Reach
139
