Jensen Huang Responds to US Chip Ban on China with Outburst: "Too Stupid, a Completely Losing Mentality"
"You're not talking to someone who wakes up thinking they're a loser!" Jensen Huang, CEO of NVIDIA, rarely loses his composure, but did so when questioned about selling chips to China. Recently, Huang was a guest on a podcast where he engaged in a heated debate with host Dwarkesh Patel regarding the controversy over whether the US should sell AI chips to China. Faced with repeated questioning and doubts, Huang nearly lost control of his emotions, bluntly calling the advocacy for a chip ban on China extremely foolish and a typical loser's mentality.
The dialogue took place on Patel's personal podcast program. Patel consistently plays the role of devil's advocate in his interviews, posing questions from a position directly opposite Huang's. His core questioning centered on whether opening up AI chip sales to China would threaten the interests and national security of US companies.
Patel first presented arguments from a national security perspective. He cited Anthropic's Claude Mythos large model as an example, stating that the model can uncover thousands of zero-day vulnerabilities in major operating systems and web browsers. If China gains access to the massive computing power provided by NVIDIA, it could potentially develop cyberattack capabilities, directly threatening US national security.
In response to this concern, Huang immediately replied that the training of Claude Mythos used relatively ordinary computing power and the scale was insignificant.
He further pointed out that China already possesses a large amount of computing resources. Even if NVIDIA holds the world's most advanced and energy-efficient chips, China can still build advanced AI models by stacking scale. Huawei's AICloud Matrix cluster is a prime example. Blocking NVIDIA from the Chinese market will not prevent China from developing cutting-edge AI models; it will only result in China's AI training system becoming completely detached from the US technology stack.
“We need to ensure that all AI developers globally are developing on the US technology stack. Technological breakthroughs in the AI field, especially advancements in open-source technology, will ultimately benefit the US industrial ecosystem,” Huang stated directly. “Artificially creating two separate ecosystems – one open-source ecosystem running on foreign technology stacks and one closed ecosystem running on the US technology stack – is extremely foolish for the US and will have disastrous consequences.”
In addition to national security, Patel also raised concerns at the industrial level. He believes that selling advanced AI chips to China could repeat the fate of the iPhone and Tesla in the Chinese market. These two products are still leaders in their respective fields, but Chinese domestic manufacturers have already created products that can compete comprehensively in terms of price, function, and quality. The AI chip industry could see the same scenario unfold.
Once China creates AI chips that match the performance of NVIDIA's latest products, Chinese AI companies will either proactively switch to domestic chips or complete the switch under policy requirements, ultimately causing NVIDIA to lose the Chinese market completely.
Faced with this concern, Huang's emotions clearly became agitated.
“What we need to do is continue to innovate. You should also be clear that our market share has been growing, not shrinking,” he directly refuted the assumption that entering the market would inevitably lead to its loss. “You're not talking to someone who wakes up thinking they're a loser. This loser mentality, this loser assumption, makes no sense to me.”
He further explained that AI chips are completely different from cars. Consumers can freely change car brands every day, but computing power ecosystems cannot. “The long-term dominance of the x86 architecture and the strong user stickiness of the ARM ecosystem are not accidental. These ecosystems are extremely difficult to replace, switching ecosystems requires enormous time and effort, and most people are unwilling to do so.” Huang stated that NVIDIA's core work is to continuously cultivate this ecosystem, continuously advance technological progress, and maintain competitiveness in the market.
Throughout the debate, Huang also presented a core judgment. AI technology is divided into five layers: energy, chips, infrastructure, models, and applications. One should not ignore the interests of all other layers for the sake of just one layer.
He repeatedly asked in return: “Why should the AI industry sacrifice one layer to benefit another? The AI industry has five layers, and each layer needs to succeed. The application layer should actually be the most successful. Why are you so fixated on AI models and a single company? What is the purpose of this?”
At the end of the conversation, Huang once again reiterated his position.
“I completely disagree with abandoning a market based on the assumptions you've made. It makes no sense,” he said. “Because I don't believe America is a loser, and our industry isn't either. This losing proposition, this loser mentality, is completely illogical to me.”