DeepSeek has unveiled software designed to run AI workloads on Huawei's AI chips, a move that directly challenges Nvidia's grip on the Chinese AI computing market.

The Chinese AI developer released tooling that makes Huawei's processors a practical option for training and inference, according to a report from Seeking Alpha. Until now, Nvidia's CUDA software ecosystem has been the de facto platform for AI development worldwide, and it remains the company's deepest moat — arguably more durable than its silicon lead.

The stakes are commercial, not technical vanity. Washington has progressively tightened export controls on advanced AI accelerators sold to China, cutting off Chinese buyers from Nvidia's top-tier data center GPUs. In response, Huawei has pushed its Ascend line of AI processors as a domestic substitute. The hardware exists. What has lagged is software: developers around the world have spent nearly two decades building on CUDA, and that accumulated toolchain, libraries, and developer habit has kept customers on Nvidia even where alternatives are available on paper.

DeepSeek's release targets exactly that gap. By providing software that lets AI models run on Huawei silicon, the company reduces the switching cost that has protected Nvidia's revenue in China — and potentially in other markets where US export rules or cost pressure push buyers toward Chinese hardware.

The move carries weight because of who DeepSeek is. The company has already demonstrated that it can deliver frontier-adjacent AI models at a fraction of the compute budget its Western rivals deploy, a capability it proved publicly with earlier releases. When a lab with that track record invests engineering effort in porting its stack to Huawei chips, it signals that Ascend hardware is now usable for serious work, not just a nationalist showcase.

For Nvidia, the threat is uneven. The company still holds the performance lead at the high end, and its Blackwell-generation accelerators remain unmatched for the largest training runs. Nvidia also continues to sell simplified, export-compliant variants of its GPUs into China, generating billions in revenue from a market it cannot fully serve. But that fallback business depends on Chinese buyers accepting deliberately hobbled hardware. Every improvement in the Huawei software stack narrows the gap between a sanctioned Nvidia SKU and a fully enabled domestic alternative, eroding the pricing power that compliant GPUs have enjoyed.

For Huawei, DeepSeek's software strengthens the Ascend ecosystem at a moment when Beijing is pushing hard for semiconductor self-sufficiency. Chip adoption in AI is a software decision as much as a procurement one, and a credible development stack multiplies the value of every wafer Huawei can produce.

Neither company has disclosed licensing terms, supported chip models, or performance figures for the new software in the material reported so far, so claims about parity with CUDA-based workflows remain unverified. The Seeking Alpha report does not specify whether the tooling covers training, inference, or both.

What is confirmed is the direction: China's most cost-efficient AI lab is now writing software for China's most advanced AI chips. If the tooling proves reliable in production, Nvidia's China revenue faces a structural challenge that export rules alone never managed to create.