ByteDance and Tencent have each taken delivery of about 10,000 Nvidia H200 chips in recent weeks, the first shipments at meaningful scale since Washington began approving sales to Chinese buyers last winter. Beijing is routing the hardware to Hong Kong rather than clearing it onto the mainland.
The volume itself is a reversal. In mid-July, a senior Commerce official told Congress that only a “trivial” number of H200s had reached China despite roughly $10 billion in approved licenses. Washington had cleared about ten Chinese firms, ByteDance, Tencent and Alibaba among them, to buy up to 75,000 chips each; Beijing’s own approval process, run case by case through the National Development and Reform Commission, was the bottleneck holding every one of those licenses on paper. Twenty thousand chips moving in a matter of weeks is Beijing choosing, for the first time, to let that bottleneck open.
It opened on Beijing’s terms. Hong Kong sits outside the mainland customs border, and its own data center capacity and power supply are limited enough that deploying chips there at real scale is its own constraint, separate from any policy one. The routing choice reflects where China still needs Nvidia most: domestic chips increasingly handle inference workloads, but training frontier models still leans on hardware China cannot yet replace at home. Chips sitting offshore can still be reached remotely by teams on the mainland, which is exactly the kind of arrangement the Commerce Department’s Bureau of Industry and Security has been building enforcement lists to cover, after a review opened this month found that current export rules do not clearly address chips reached over a network connection rather than physically imported.
The compromise suits both governments for now. Washington can point to licensed chips actually moving, supporting the administration’s argument that selling H200s to China beats losing that business to Huawei entirely. Beijing gets its national champions more training compute while keeping the hardware off the mainland, where a large deployment could undercut its push for domestic chips.
The 10,000 delivered to each company is a fraction of the 75,000-chip cap each already holds, itself a fraction of what Chinese firms have reportedly ordered for the year against Nvidia’s constrained global supply. Whether the next tranche also stays in Hong Kong, or finally crosses into the mainland, will show whether this delivery was a real loosening of the block or a limited exception to it.
Sources: Engadget, Business Standard
–
By the Control Plane Editorial Team