Nvidia's H200 chips have started reaching mainland China, but Beijing is making sure most of them can't actually be plugged in yet.
The Financial Times reports, citing two people with knowledge of the matter, that ByteDance and Tencent have each taken delivery of roughly 10,000 H200 accelerators in recent weeks - the first meaningful movement of the chips onto the mainland since President Trump approved their export last December in exchange for a 25% cut of sales revenue to the U.S. Treasury. Around 10 companies, including Alibaba, ByteDance, Tencent, and JD.com, hold licenses for up to 100,000 chips apiece, but China's National Development and Reform Commission requires case-by-case approval for every purchase, and until now almost nothing had shipped. Nvidia is reportedly sitting on about 500,000 H200s built for Chinese buyers. Lenovo and other partners told customers last week they could resume taking H200 server orders, per the FT, though each sale still needs separate sign-off from Beijing.
Here's the catch: Beijing is steering most of each company's allowance toward Hong Kong, not the mainland. That's a problem, because a single 100,000-chip allotment draws roughly 125 megawatts at full load, and Hong Kong's entire data-center footprint totals about 581 megawatts, according to Hong Kong Free Press. There's effectively nowhere in the territory to switch these chips on.
Chinese labs still lean on Nvidia hardware to train frontier models - DeepSeek's reported struggles training its R2 model on Huawei's Ascend chips show why domestic silicon isn't a clean substitute yet. But parking the bulk of licensed H200s somewhere that can't power them looks less like a market opening and more like a holding pattern that keeps the mainland market captive for China's own chipmakers.