NVIDIA H20 Chip Blocked by U.S. Export Ban to China
The NVIDIA H20, designed as the most powerful GPU that could still be exported to China under earlier trade limits, has now been blocked by the U.S. government’s expanded export restrictions.
Unveiled in late 2023, the H20 was a custom version of NVIDIA’s Hopper architecture, created to comply with Washington’s rules. It featured 96GB of HBM3 memory, advanced NVLink interconnect, and strong inference performance—enough to support China’s booming AI sector while staying below U.S. performance thresholds. For many Chinese data centers, it became the most important replacement for the A100 and H100 chips, which were already restricted.
That window has now closed. In 2025, the U.S. Commerce Department broadened its semiconductor controls, extending the ban to cover the H20 alongside other export-limited models like the L20 and L2. The move was aimed at tightening China’s access to advanced AI hardware seen as critical for large-scale model training and national security applications.
The decision leaves Chinese tech giants and hyperscalers scrambling for alternatives, accelerating their shift toward domestic GPUs from Huawei’s Ascend line and startups like Biren. Analysts note that while NVIDIA remains the global leader in AI chips, its direct access to the Chinese market—once a major revenue source—has now shrunk dramatically.
NVIDIA is already working on new Blackwell-based designs that could potentially comply with U.S. rules, but their availability in China remains uncertain. For now, the H20 joins the growing list of GPUs barred from one of the world’s largest markets.
