Nvidia CEO Jensen Huang Warns Against DeepSeek Running on Huawei Chips

Nvidia CEO Jensen Huang issued a stark warning on the Dwarkesh Podcast, stating that DeepSeek's strategic shift to optimize its AI models for Huawei's Ascend chips instead of American hardware would be 'a horrible outcome' for the United States. The Chinese AI lab is preparing to launch its V4 foundation model on Huawei's Ascend 950PR processor, marking a significant migration from Nvidia's CUDA software ecosystem to Huawei's CANN framework. This move threatens to break the software-hardware dependency that has underpinned American AI dominance for the past decade, where virtually every frontier model has been trained on Nvidia GPUs using CUDA.

DeepSeek has spent months rewriting its core code for Huawei's CANN framework, creating a critical vulnerability in US technological control. While export restrictions can limit Nvidia hardware reaching China, Chinese labs remained dependent on the CUDA ecosystem even when using alternative processors. DeepSeek's V4 model—reportedly trained on Nvidia's Blackwell chips (potentially violating export controls) but deployed on Huawei hardware—demonstrates China's ability to produce competitive models with fewer resources. Huawei's chips currently deliver about 60% of the inference performance of Nvidia's H100, but Huang acknowledges China's 'abundant energy' and 'large pool of AI researchers' could compensate for hardware disadvantages.

The partnership between China's most capable AI lab and its most advanced chipmaker represents a direct challenge to American technological leverage. If V4 performs well on Ascend chips, it validates an alternative AI development path that doesn't depend on Nvidia at any point, potentially allowing China to 'become superior to' the US in AI as the technology diffuses globally with Chinese standards. This comes as US lawmakers consider placing DeepSeek on the entity list for export control, highlighting the escalating tech war between the two superpowers.