NVIDIA H200 May be Approved Export to China

On November 21, sources cited by Reuters revealed that the U.S. Trump administration is weighing approval for the export of NVIDIA's H200 AI chips to China. The potential move aims to ease U.S.-China tensions while addressing NVIDIA's current zero sales of AI data center chips in the Chinese market.

Officials from the U.S. Department of Commerce, responsible for export controls, are reviewing changes to the policy restricting such chip sales to China, though the plan remains subject to adjustment. A White House official declined to comment directly but emphasized that the government is committed to maintaining U.S. leadership in global technology while safeguarding national security.

NVIDIA has not commented specifically on the ongoing review but noted that current regulations prevent the company from offering competitive AI data center chips in China, effectively leaving this large market to fast-growing foreign competitors.

In an interview with Fox Business Channel's The Claman Countdown on November 20, NVIDIA founder and CEO Jensen Huang highlighted the strategic importance of re-entering China to sustain U.S. leadership in AI:

"It's clear that the U.S. needs to compete in China. This benefits the American people, the U.S. technology sector, and strengthens our global success."

Although NVIDIA's H20 chips were previously allowed for limited exports to China, safety concerns and the chip's reduced capabilities have left it largely obsolete, pushing NVIDIA's AI chip revenue from China to zero. Huang expects this trend to continue for the next two quarters.

Huang expressed hope for a breakthrough between the U.S. and Chinese governments that would enable NVIDIA to re-enter the massive Chinese market. He estimates that China's AI chip market, currently valued at roughly $50 billion, could reach $200 billion by the end of the decade.

"It's unfortunate that U.S. companies are excluded from this important revenue stream. Access would allow us to accelerate investments and innovation," Huang said.

While the Trump administration has explicitly ruled out exporting NVIDIA's most advanced Blackwell GPUs to China, a limited export of the H200 chip—based on the previous-generation Hopper architecture—remains possible. Unlike the earlier H20, the H200 features 141GB of HBM3e memory and a bandwidth of 48TB/s, offering roughly double the performance of H20.

However, export control restrictions may still require NVIDIA to partially limit the H200's core count and bandwidth before shipment. Even so, its overall performance would substantially surpass the H20's 96GB configuration.

For Chinese AI technology companies seeking high-performance chips, this could accelerate AI model development. At the same time, domestic chip developers may face intensified competition from this enhanced foreign option.