NVIDIA has reportedly stopped production of its China-focused H200 'Hopper' GPU at TSMC's facilities, according to multiple reports. The company built up an inventory of 250,000 H200 GPUs for the Chinese market, intended for select applications that don't compromise U.S. national security. These GPUs are now stored in warehouses awaiting export and import approvals.

After receiving export rights for H200 accelerators, NVIDIA stockpiled the chips to supply Chinese AI labs. However, China's own restrictions on imports have complicated matters, requiring companies to obtain exemption letters from Beijing. The H200 GPUs use TSMC's 5 nm node with CoWoS-S packaging technology.

NVIDIA will now reallocate production capacity to its upcoming 'Rubin' GPU, though these use different manufacturing processes - a 3 nm node with CoWoS-L packaging. Reports suggest this reallocation involves packaging capacity rather than semiconductor node conversion, as switching between 5 nm and 3 nm production would require significant equipment changes. As NVIDIA's largest customer, the company has secured production capacity at TSMC for specific wafers on designated nodes over set timeframes.