Nvidia’s Huang warns DeepSeek running on Huawei chips would be ‘horrible’ for the US

At a glance:

• DeepSeek plans to launch V4 on Huawei’s Ascend 950PR, moving away from Nvidia’s CUDA.
• CEO Jensen Huang calls the shift a “horrible outcome” for US AI dominance.
• The move could break Nvidia’s software-hardware lock and reshape global AI supply chains.

DeepSeek’s pivot to Huawei threatens US AI leverage

Nvidia CEO Jensen Huang issued a stark warning on the Dwarkesh Podcast, framing DeepSeek’s reported migration from Nvidia GPUs to Huawei’s Ascend chips as a direct threat to American technological supremacy. If China’s most advanced AI lab succeeds in optimising its models for Huawei’s CANN framework instead of Nvidia’s CUDA, Huang said, it would create an alternative AI development stack that could erode the United States’ current dominance. The stakes are high: CUDA has functioned as a second layer of American control over AI, beyond the chips themselves, because virtually every frontier model outside China is built on it.

The V4 launch and the software-hardware dependency

DeepSeek is preparing to launch V4, a multimodal foundation model expected later this month, on Huawei’s Ascend 950PR processor. The Information reported in April that V4 would run on this latest Huawei silicon, while Reuters suggested the model had been trained on Nvidia’s Blackwell chips—a potential export-control violation. These claims are not mutually exclusive; a model can be trained on one hardware set and deployed for inference on another. What matters is the underlying software migration: DeepSeek has spent months rewriting its core code to work with Huawei’s CANN framework, breaking free from the CUDA ecosystem that Nvidia has spent two decades embedding into AI development worldwide.

Why raw performance gaps may not decide the race

On paper, Huawei’s chips lag Nvidia’s. The Ascend 910C, predecessor to the 950PR, delivers roughly 60% of the inference performance of Nvidia’s H100, a chip two generations behind Nvidia’s current best. American chips are about five times more powerful than their Chinese equivalents today, with the gap projected to widen to 17 times by 2027. Yet Huang’s concern is not about today’s performance deficit. He noted that even with inferior chips, China could catch up in AI development thanks to abundant energy, a large pool of AI researchers, and the ability to optimise software for its hardware. If V4 performs competitively on Ascend chips, it would validate an alternative path for AI development that bypasses Nvidia entirely.

The export-control paradox and Huawei’s growing role

The situation exposes a tension at the heart of US chip export policy. Nvidia restarted production of the H200 for the Chinese market, but China has blocked H200 imports to protect Huawei’s domestic chip business, and Nvidia has recorded no revenue from China H200 sales. Controls designed to limit China’s AI capabilities are instead accelerating the development of a Chinese alternative. DeepSeek’s experience with its R2 model illustrates both the promise and limits of the Huawei route: R2 was repeatedly delayed due to training failures on Huawei hardware, forcing the company to revert to Nvidia GPUs for training while using Huawei chips only for inference. The distinction matters—training is the most compute-intensive phase, and Huawei’s chips could not yet handle it reliably—but inference, where commercial value is generated, appears within reach.

US lawmakers push tighter restrictions

Meanwhile, US lawmakers are seeking to tighten the screws further. On Thursday, legislators and experts accused China of buying “what they can” and stealing “what they cannot” in the AI industry, calling for the government to evaluate placing DeepSeek, Moonshot AI, and MiniMax on the entity list for export control. The push reflects growing anxiety that export restrictions are not slowing China’s AI progress as intended, but instead spurring the creation of a parallel, self-sufficient Chinese AI stack.

Huang’s deeper warning: the erosion of CUDA’s moat

Huang’s warning is ultimately about software-hardware co-design. Nvidia’s dominance rests not just on making the fastest chips but on CUDA’s position as the default development environment for AI. When researchers write code, they write it for CUDA. When startups build products, they build them on CUDA. When governments invest in AI infrastructure, they buy Nvidia GPUs because that is what the software requires. DeepSeek’s migration to CANN threatens to create a parallel ecosystem in which none of that applies.

The scale of Nvidia’s business makes the stakes concrete. The company’s market capitalisation exceeds $3 trillion. Its data centre revenue grew 93% year over year in its most recent quarter. Its chips power the training runs for virtually every major AI model outside China. If the most capable Chinese AI lab demonstrates that competitive models can be built without Nvidia, the argument for maintaining export controls weakens, the argument for buying Nvidia weakens, and the geopolitical assumptions that have shaped AI policy for the past three years come under pressure.

What V4’s performance will signal

None of this means Huawei is about to overtake Nvidia. The performance gap is large and growing. The R2 training failures demonstrate that Chinese hardware is not yet ready for the most demanding AI workloads. But Huang is not warning about today. He is warning about a trajectory in which DeepSeek proves the concept, other labs follow, and the CUDA moat that has made Nvidia the most valuable company in the AI supply chain begins to erode.

The fact that the CEO of Nvidia is the one making this argument publicly suggests he believes the risk is no longer theoretical. DeepSeek’s V4 will be the first major test. If a multimodal foundation model runs competitively on Huawei silicon, the warning Huang issued on Wednesday will look less like corporate lobbying and more like the most consequential forecast in the AI chip war so far.