Nvidia's Jensen Huang warns DeepSeek running on Huawei chips would be 'horrible outcome' for America

Nvidia CEO Jensen Huang warned on the Dwarkesh Podcast that DeepSeek optimizing its AI models for Huawei's Ascend chips instead of American hardware would be 'a horrible outcome' for the United States. DeepSeek is preparing to launch its V4 multimodal foundation model on Huawei's Ascend 950PR processor, requiring a migration from Nvidia's CUDA framework to Huawei's CANN framework. This software migration is the core concern: CUDA's dominance has functioned as a second layer of American control over AI development, and breaking that dependency could erode Nvidia's moat. While Huawei chips currently deliver only ~60% of H100 inference performance and failed to handle DeepSeek's R2 training workloads reliably, Huang warns that software optimization, researcher talent, and energy availability could compensate for hardware gaps over time. US lawmakers are simultaneously pushing to place DeepSeek on the export control entity list, while the existing controls may be paradoxically accelerating China's development of a domestic AI hardware ecosystem.