Nvidia's Jensen Huang Warns Against Selling AI Chips to China

Nvidia CEO Jensen Huang recently engaged in a heated debate on the Dwarkesh Podcast, discussing the implications of the United States selling artificial intelligence (AI) chips to China. This conversation comes at a time when the geopolitical stakes surrounding AI technologies are higher than ever, with concerns about national security and technological dominance at the forefront.

During the podcast, Huang was pressed by host Dwarkesh Patel on whether providing China access to AI chips poses a threat to American companies and national security. Patel highlighted the potential for Chinese entities to develop cyber-offensive capabilities, citing Anthropic’s Claude Mythos, which reportedly revealed vulnerabilities in major operating systems and web browsers. He argued that if China gains access to Nvidia's advanced computing power, it could bolster its capabilities in ways that threaten U.S. security.

In response, Huang articulated a more complex view. He pointed out that China already possesses substantial compute power, asserting that restricting Nvidia's chip sales would not halt China's AI advancements. Instead, he warned that it would push Chinese developers to create AI models using non-American technologies. "You’re not talking to someone who woke up a loser," Huang stated, emphasizing the importance of maintaining an American tech ecosystem for AI development.

Huang elaborated on the dangers of fostering two separate AI ecosystems—one that is open-source and based on American technology, and another that is closed and reliant on foreign tech. He argued that such a split would be detrimental to U.S. interests. "It would be extremely foolish to create two ecosystems: the open-source ecosystem, and it only runs on a foreign tech stack, and a closed ecosystem that runs on the American tech stack," he said. "I think that would be a horrible outcome for the United States." This sentiment reflects a growing concern among tech leaders about the implications of U.S.-China tech competition.

As the conversation progressed, Huang noted that the AI industry comprises five layers: energy, chips, infrastructure, models, and applications. He stressed that neglecting any one of these layers could jeopardize the entire industry. "Why are you causing one layer of the AI industry to lose an entire market so that you could benefit from another layer of the AI industry?" Huang asked, indicating that a balanced approach is necessary for the success of the entire ecosystem.

This debate has become even more pressing with the upcoming launch of DeepSeek's V4 multimodal foundation model, which is set to run on Huawei’s Ascend 950PR processor later this month. Huang warned that if DeepSeek optimizes its AI models for Huawei chips rather than American hardware, it could signify a shift that undermines U.S. technological leverage in AI. He cautioned that if Chinese AI labs start developing their models using Huawei’s CANN framework instead of Nvidia’s CUDA, it could lead to a scenario where China becomes superior to the U.S. in AI development.

Huang's concerns are underscored by the fact that DeepSeek's previous models, including V3, were trained on 2,048 Nvidia H800 GPUs—chips that were banned from sale to China in 2023. The shift to Huawei's technology breaks the dependency on American hardware and poses a direct challenge to the dominance that Nvidia has enjoyed in the AI sector.

In the podcast, Huang explained that even though Huawei's chips currently lag behind Nvidia's in performance—American chips are approximately five times more powerful—China’s vast resources, including abundant energy and a large pool of AI researchers, could enable it to catch up. This potential for progress raises alarms about the future of American AI leadership.

Huang's remarks come at a time when U.S. lawmakers are considering tightening restrictions on Chinese tech firms. Recently, experts have accused China of acquiring technology through both legal means and espionage, advocating for the inclusion of companies like DeepSeek on the U.S. entity list for export controls. Such measures are intended to curb China's technological advancements, but Huang suggests that they may inadvertently accelerate the development of competitive alternatives.

The tension between maintaining U.S. technological supremacy and the realities of global competition is evident in Huang's warnings. He highlighted that Nvidia's dominance is not solely based on its chip performance; it is also tied to CUDA's status as the default development environment for AI. As researchers and startups build their products around CUDA, the ecosystem becomes increasingly entrenched, making it difficult for alternatives to gain traction.

Huang's concerns about DeepSeek's transition to Huawei's CANN framework represent a broader fear that the U.S. could lose its competitive edge in AI if alternative ecosystems emerge. The stakes are high, as Nvidia's market capitalization exceeds $3 trillion and its data center revenue has grown significantly over the past year. The potential for a leading Chinese AI lab to demonstrate that competitive models can be built without reliance on Nvidia could weaken the argument for maintaining export controls and diminish Nvidia's market position.

Looking ahead, the launch of DeepSeek's V4 model will serve as a key test of Huang's predictions. If the model performs competitively on Huawei silicon, it could validate a new path for AI development that does not depend on American technology, challenging the assumptions that have shaped U.S. AI policy for years. Huang’s warnings, framed as corporate advocacy, may instead be a reflection of a rapidly changing technological battlefield.

As the global tech community watches closely, the implications of this debate extend beyond corporate interests, touching on national security, economic competitiveness, and the future of innovation in AI. The outcome of the U.S.-China tech rivalry could very well redefine the global AI industry in the years to come.