Amid renewed market anxiety over AI triggered by the Kimi K3 model and pressure on chip stocks, NVIDIA CEO Jensen Huang has publicly defended Chinese open-source models, stating that the market has once again misjudged the situation.
In an exclusive video interview on July 21st, Huang argued that the market first misinterpreted DeepSeek and is now misinterpreting Kimi. His core thesis is that free, open-source models lower barriers to entry, allowing more users to access AI for the first time, which in turn boosts demand for chips, data centers, and computing power. As he put it, "free AI should benefit hardware, benefit chips, benefit data centers."
Addressing fears that Chinese models could push American companies out of the market, Huang directly dismissed the possibility, calling it "zero chance." He believes open-source models will expand the overall market, as many users will still pay for the convenience, reliability, and superior performance of closed-source services.
On the policy front, Huang warned that restricting open-source models on national security grounds could actually make the United States more vulnerable. He countered the argument that downloading a Chinese model equates to giving Beijing a backdoor, noting that companies can adapt and control access within isolated sandbox environments. In contrast, he suggested the world relying on a single closed-source model poses a greater "single point of failure" risk.
Notably, he applied the same logic to American companies, unusually calling on his own client, Anthropic, to open its restricted cybersecurity model, Claude Mythos, to everyone rather than limiting access from the start. He stated that "Mythos should be offered as a service" and that "restricting Anthropic is not in the interest of America."
Huang's comments also directly countered the stance of the U.S. Treasury Department. Just a day earlier, Treasury officials stated they were reviewing whether Chinese models were stealing U.S. intellectual property and considering sanctions. Huang offered a contrary conclusion: learning from other sources of knowledge is fundamental to intelligence. He argued that companies should face consequences for violating privacy or contracts, but the focus should be on specific misconduct, not the models themselves.
The Kimi K3 model, released by Chinese AI firm Moonshot AI on July 16th, is the largest open-source model with 2.8 trillion parameters and recently topped the Arena front-end programming leaderboard, marking a first for a Chinese model. Its release and low-cost, open-source nature contributed to the Philadelphia Semiconductor Index falling about 10% that week, its worst performance since April 2025, with NVIDIA briefly losing its position as the world's most valuable company. Huang's perspective finds some echo among sell-side analysts; Morgan Stanley believes the global AI computing demand growth cycle is far from over, while Bernstein calculates that even if K3 reduces training costs by about 40%, competition driving firms to double training frequency could still increase total power demand by roughly 20%.
Comments