Nvidia CEO unsure China would take its H200 chips

Nvidia CEO Jensen Huang said he is unsure if China would accept the company’s H200 AI chips if US export restrictions were relaxed.

Huang made the comments to reporters after meeting with President Donald Trump, and before a closed-door session with the Senate Banking Committee, which oversees export controls.

He said that Nvidia cannot offer downgraded chips to China, as the country would likely reject them.

The US has imposed export restrictions on advanced AI chips since 2022 to limit China’s access to sensitive technology.

Lawmakers have discussed the GAIN AI Act, which would require US customers to get priority access to AI chips before they can be sold to countries like China, but it has not been adopted.

🔗 Source: Bloomberg

🧠 Food for thought

Implications, context, and why it matters.

Huawei Ascend gains ground in China’s AI data centers, which could reduce the impact of any relaxed US export rules

Chinese cloud and AI firms are adopting Ascend 910B plus 910C; Huawei sold 200,000 AI chips in 2024 versus Nvidia’s over 1 million H20 units 1. It has 300+ Atlas 900 A3 SuperPoDs serving over 20 customers 2. In mid-2025, China’s regulators reviewed Nvidia’s H20, and analysts said firms like Tencent pulled back while testing domestic options 3.
Huawei runs Atlas 900 A3 SuperPoD clusters with up to 384 Ascend 910C chips delivering up to 300 petaflops, the largest SuperPoD in the world as of 2025 2.
Atlas 950 targets 520,000+ Ascend chips at 524 exaFLOPS FP8 by Q4 2026, and the 960 tops 1 million chips at 2 zettaFLOPS FP8 by Q4 2027 2.

Systems integrators can earn services work by moving CUDA workloads to CANN while supply stays uncertain

CANN is Huawei’s Compute Architecture for Neural Networks. It trails CUDA, a software platform for parallel computing, with crash reports and developer reluctance to port that create a migration services opening 1.
Huawei plans to open-source CANN and related tools by end of 2025 to speed adoption 3. This gives systems integrators, IT consultancies that design and implement complex systems, time to build migration skills.
US export controls and shifting policy keep chip supply uncertain, which makes cross-platform portability valuable 1.
Domestic rivals like Cambricon, Alibaba’s T-Head, and Baidu-backed Kunlun Tech ship CUDA-compatible chips, which ease developer adoption and add migration targets 3.