DeepSeek Withholds V4 Model from Nvidia and AMD — The AI Cold War Just Got Real

Two AI chip processors facing each other across a digital divide, representing US-China tech rivalry

In a move that breaks with standard industry practice, Chinese AI lab DeepSeek has reportedly withheld its upcoming flagship V4 model from US chipmakers Nvidia and AMD — while granting early access to domestic Chinese suppliers including Huawei. According to Reuters, the decision marks a significant escalation in the growing tech rivalry between the US and China.

What Happened

AI developers typically share pre-release versions of major models with leading chipmakers to ensure software runs efficiently on widely used hardware. DeepSeek has previously worked closely with Nvidia's technical staff. But for V4, the company gave Chinese chipmakers a several-week head start to optimize the software for their processors, while leaving Nvidia and AMD out in the cold.

Neither Nvidia nor AMD commented on the situation. DeepSeek and Huawei did not respond to Reuters' requests for comment.

The Bigger Story

Here's where things get really interesting. A senior Trump administration official told Reuters that DeepSeek's latest AI model was trained on Nvidia's most advanced Blackwell chip using a cluster in mainland China — a move that appears to violate US export controls. Even more provocatively, DeepSeek may seek to remove technical indicators revealing its use of American chips and plans to publicly claim it used Huawei's chips for training instead.

If true, this is a remarkable piece of geopolitical theater: allegedly using American chips to train your model while simultaneously shutting American companies out of the optimization process, then claiming you used domestic chips all along.

Impact on US Chipmakers

Research firm Creative Strategies CEO Ben Bajarin downplayed the immediate technical impact, noting that "most enterprises are not running DeepSeek" and that new AI coding tools are reducing optimization timelines "from months to weeks." But the strategic implications are harder to dismiss.

The move is likely part of a broader Chinese government strategy "to try to keep US hardware and models disadvantaged" in China, according to Bajarin. DeepSeek's models have been downloaded more than 75 million times on Hugging Face since the company's emergence in January 2025, and Chinese open-source model downloads have surpassed those from any other country.

The Export Control Dilemma

The situation highlights the ongoing tension in Washington over AI chip exports to China. US authorities allowed Nvidia's H20 and AMD's MI308 inference chips to resume shipments to China, even as licenses for more advanced processors remain restricted. AMD reported $390 million in MI308 sales in its most recent quarter alone — evidence that these chips are in significant demand.

The Bottom Line

DeepSeek withholding V4 from US chipmakers is a shot across the bow in the AI cold war. Whether it's a strategic government-directed move or a company-level decision, the message is clear: China's AI ecosystem is actively working to reduce its dependence on American hardware — even if it's still secretly using it behind the scenes. The irony is thick, but the geopolitical implications are dead serious.