NVIDIA CEO Huang Renxun boldly supports China’s AI, saying the U.S. is in “Kimi panic” and “got the priorities wrong”: free and open-source will ignite demand for NVIDIA chips

In response to the recent market panic triggered by China’s AI model Kimi K3, NVIDIA CEO Jensen Huang offered a radically different interpretation. In an exclusive interview with Axios, Huang strongly defended China’s open-source AI models, saying that U.S. companies “absolutely” should use these excellent models. He said that Wall Street and Washington’s ban-minded mindset is completely going in the wrong direction, because cheaper, open-source AI applications will significantly boost strong demand for hardware and chips.
(Background: How did Sega save NVIDIA back then? Jensen Huang has been grateful for a full 30 years.)
(Additional context: Want to get hold of Jensen Huang’s original signature leather coat? Sotheby’s held a charity auction, with the estimates revealed.)

Table of contents

Toggle

• Washington has it backward—free AI is a super bullish driver for hardware
• Oppose national security crackdowns, saying “China can’t possibly push the U.S. out of the race”
• Open source makes AI safer, urging Anthropic to unrestrict its model

The Kimi K3 model rolled out by the China startup Moonshot AI has recently shaken the market. With near state-of-the-art performance and an extremely low price, it sparked the most severe chip-stock selloff panic since the January 2025 DeepSeek incident. However, in the eyes of Jensen Huang, CEO of NVIDIA (the global AI computing powerhouse), this “Kimi panic” is entirely a misunderstanding.

Washington has it backward—free AI is a super bullish driver for hardware

Speaking at the opening ceremony of an advanced AI infrastructure factory built by the Taiwanese contract manufacturer Wistron in Fort Worth, Texas, Jensen Huang received an exclusive interview with Axios. He said bluntly that Wall Street and Washington’s fear of China’s open-source AI models is “completely backward.”

He explained that cheaper and open-source models will cause more people and companies to engage with artificial intelligence, thereby expanding the entire market; and once these applications become widespread, it will inevitably translate into strong demand for NVIDIA chips, data centers, and computing power. “Free AI should be good for hardware, good for chips, good for data centers,” Huang emphasized.

Oppose national security crackdowns, saying “China can’t possibly push the U.S. out of the race”

Recently, including U.S. Treasury Secretary Scott Bessent, officials from the Trump administration, and parts of U.S. AI labs, have repeatedly accused Chinese models of stealing U.S. technology through “distillation,” and have actively urged Washington to impose sanctions and bans.

In response, Jensen Huang made his stance clear, strongly opposing it. He pointed out that learning and distillation from other sources of knowledge are “a basic element of intelligence,” and that governments should punish specific violations rather than suppressing the model itself. Regarding concerns from the outside about “a backdoor in Beijing,” he dismissed it as a misunderstanding, because companies can fully customize and monitor how models are used within a secure “sandbox” environment. He then made a more forceful declaration: “In any situation, China cannot possibly push U.S. companies out of the race. Zero possibility.”

Open source makes AI safer, urging Anthropic to unrestrict its model

Huang further put forward a counterintuitive viewpoint: restricting open-source models on national security grounds would instead make the U.S. more vulnerable. He strongly believes that “openness” allows external researchers around the world to inspect models, identify weaknesses, and build defenses. “If the whole world has only one model, one point of attack, and one single source of failure, then the world would be much more fragile.”

He even applied this openness logic to his U.S. peers, publicly calling on AI giant Anthropic to open up its restricted-access network security model, Claude Mythos, for “everyone” to use—strengthening defenses through fast testing and fixes rather than setting limitations from the start. Huang concluded that this AI race isn’t a “race with a finish line.” The U.S. and China will both be around long term, and embracing openness is the best solution to accelerate the overall AI industry’s growth.