Nvidia's H200 artificial intelligence accelerators have begun arriving in mainland China for the first time in months, with ByteDance and Tencent each taking delivery of roughly 10,000 chips in August, according to a report from Tom's Hardware. It is the first concrete movement of hardware since Washington approved sales to ten Chinese technology firms back in January.
The numbers tell a story of caution on both sides of the Pacific. Each of the ten approved customers — a list that also includes Alibaba and JD.com — is licensed to buy up to 75,000 H200 units. The roughly 10,000 chips reaching ByteDance and Tencent represent only about 13% of that per-customer ceiling, a gap that underscores how far actual trade still lags the volumes regulators technically permit.
A Stop-Start Approval Process
The path to this month's deliveries has been anything but smooth. The Trump administration cleared the sales in January, but no chips moved for roughly seven months while Beijing weighed whether to let them in, prompting Nvidia to halt China-bound H200 production in March and redirect the freed-up capacity at TSMC elsewhere. Chief executive Jensen Huang said at the time that the company was reversing course as orders began arriving.
We've been licensed for many customers in China for H200. We have received purchase orders from many customers and we are in the process of restarting our manufacturing.
Jensen Huang, Nvidia CEO, March 2026
Even now that chips are moving, the logistics remain unusual. Much of what has been cleared is, according to the Tom's Hardware report, formally routed through Hong Kong — a jurisdiction that lacks the power infrastructure to run large AI training clusters, meaning the hardware will eventually need to make its way onto the mainland before it can actually be put to work. That routing reflects Beijing's own ambivalence: Chinese officials have simultaneously loosened the import block while publicly flagging the risk of overreliance on foreign silicon and pushing domestic champions such as Huawei's Ascend line as substitutes.
A Two-Generation-Old Chip, Still in High Demand
Part of what makes the standoff notable is that the H200 is no longer Nvidia's frontier product — it now sits roughly two generations behind the Blackwell-class chips the company sells to U.S. and allied customers, which remain fully blocked from export to China under current rules. Nvidia has reportedly built up a substantial inventory of H200 units awaiting Chinese buyers, a bet that even trailing-edge accelerators will find eager customers as domestic alternatives continue to lag Nvidia's software ecosystem.
That bet lines up with a broader supply crunch rippling through the chip industry this month. Foundries and memory makers alike are raising prices as AI infrastructure spending outpaces available capacity, a dynamic that is pushing Chinese buyers — cut off from the most advanced U.S. export-controlled equipment — to pay a premium for whatever legally exportable compute they can secure, H200 included.
For now, the shipments function as a test case more than a resumption of full-scale trade. Both governments have reasons to move cautiously: Washington wants to avoid handing China unrestricted access to frontier AI hardware, while Beijing wants to avoid deepening its tech sector's dependence on a supplier it could lose access to again with a single policy reversal. Whether volumes climb meaningfully closer to the 75,000-unit ceiling in the coming months will depend on how that tension resolves — and on how quickly China's own chipmakers can close the gap Nvidia is currently profiting from.
