US Lifts Nvidia H200 Export Ban to Huawei, Puts China Back in CUDA Loop

As the US lifts export restrictions on Nvidia's H200 AI accelerator, a new tech battle emerges.

The US has lifted export restrictions on Nvidia’s H200 AI accelerator for China, attaching a 25% fee. This decision is part of the ongoing tech rivalry between American and Chinese firms.

📌 Key Takeaways

✓US lifts ban on Nvidia H200 exports with a 25% fee
✓Decision influenced by Huawei’s advances in Ascend chips
✓Restrictions still apply to Nvidia’s latest architectures
✓Many Chinese firms prefer Nvidia CUDA over domestic alternatives

Decision Behind the Scenes

The move aims to maintain the dominance of the American tech stack while preventing significant setbacks in national security. Huawei’s advancements with its CloudMatrix 384 system, featuring Ascend chips, influenced this decision as they pose a competitive challenge to Nvidia’s offerings.

Despite China’s efforts towards developing an independent instruction set via open-source CANN, many AI firms still prefer Nvidia’s CUDA-based systems for training advanced models. This preference is evident with Deepseek choosing Nvidia over domestic alternatives.

Nvidia’s Strategic Move

Nvidia CEO Jensen Huang had previously expressed uncertainty about the H200 sales to China post-restriction changes. With this policy shift, Chinese companies now have access to a wider array of AI accelerators but are still restricted from acquiring Nvidia’s latest Blackwell architectures.

The decision reflects a strategic balance between maintaining tech leadership and avoiding complete exclusion that could backfire by accelerating domestic technology growth in rival nations like China.

Future Outlook

As Huawei scales up production of its Ascend 910C chips, the US faces a critical period where technological superiority might be at stake. By allowing H200 sales but blocking Blackwell architectures, the US hopes to keep Western AI models ahead in performance.