Nvidia’s H200 AI chips have begun reaching Chinese companies again in limited batches, with ByteDance and Tencent each receiving roughly 10,000 units in recent weeks, according to Financial Times reporting relayed by Benzinga and corroborated by CNBC’s earlier coverage of the export approval process. No primary statement from Nvidia or the US or Chinese governments is public yet; this account relies on FT’s sourcing as relayed by the outlets above.
What’s actually happening
The H200 is an older Nvidia chip generation, not the company’s current flagship silicon, which Chinese customers still cannot buy under standing US export controls. US officials had approved limited H200 exports to China back in July, per CNBC’s reporting from that period, but a US trade official said at the time that “very few” chips had actually shipped. This week’s reporting describes the first meaningful volume actually arriving.
The more unusual detail in the FT’s reporting: Beijing, not Washington, now appears to be the party trying to limit the chips’ use inside mainland China. Chinese firms have reportedly been told they may route the processors through Hong Kong, which sits outside the mainland customs boundary, and run workloads from there instead of importing the hardware directly onto the mainland.
Why it matters
A restriction moving from the exporting country to the importing one is not the dynamic either side of the US-China chip disputes has been signaling publicly, and it suggests Beijing’s own calculus about foreign AI hardware dependence is more complicated than a simple “get the chips at any cost” position. Routing through Hong Kong specifically looks like an attempt to get access to the compute without fully accepting a foreign hardware dependency onto the mainland’s own infrastructure.
For US export-control policy, a limited, monitored flow of an older chip generation reads as a pressure-release valve, not a reversal. It likely lets both governments claim a version of their existing position is still intact while real hardware still moves.
What to watch: whether shipment volumes grow beyond this initial batch, and whether Nvidia’s own current-generation chips ever get a similar limited-export treatment, or stay off the table entirely while H200 becomes the accepted middle ground.
