China's Emerging AI Model Rises: Jensen Huang Rarely 'Challenges Washington' - Ban Only Makes U.S. More Vulnerable

China's AI startup 'Moonshot AI' recently launched the Kimi K3 model, triggering discussions about the competitive landscape of artificial intelligence (AI) and even causing temporary sell-offs in chip stocks like Nvidia. However, Nvidia CEO Jensen Huang, in an exclusive interview with U.S. media Axios, stated that the U.S. should not fear China's open-source AI models. Instead, he warned that the growing 'ban sentiment' within Washington is the real concern.

Huang acknowledged that Chinese open-source models perform impressively, saying, 'Excellent open-source models should be used.' This statement directly challenges certain U.S. government officials and leading American AI labs advocating for blocking Chinese models.

Clients of Nvidia, including OpenAI and Anthropic, previously accused Chinese competitors of absorbing their model capabilities and warned that China's open-source AI wave could threaten U.S. leadership in AI. In response, Huang countered that closed-model companies need not fear open-source models. He believes open-source models allow more people to access AI for the first time, expanding the overall market. Many users, he argues, will eventually opt for paid closed services due to convenience, reliability, and superior performance. He emphasized that U.S. AI companies remain secure: 'There is no scenario in which Chinese companies push American firms into a corner. The possibility is zero.'

Following DeepSeek's market disruption in early 2025, Kimi K3 has reignited Wall Street's concerns about AI investment. The model combines near-frontier performance, extremely low pricing, and soon-to-be-released downloadable and customizable open-source weights, raising fears that cheap AI models could weaken demand for data centers and AI infrastructure investment.

However, Huang insists the market has fundamentally misunderstood the impact of such models. He stated that Wall Street first misunderstood DeepSeek and has now misunderstood Kimi again. Cheap, free, and open AI models, he argues, will actually encourage more businesses and individuals to adopt AI, thereby increasing demand for Nvidia chips, data centers, and computing power.

'Free AI should be very good news for hardware, chips, and data centers,' Huang said.

Addressing concerns that downloading Chinese models could introduce security backdoors, Huang offered a different perspective. He noted that enterprises can control and customize models within secure sandbox environments, meaning using Chinese open-source models does not equate to exposing systems to Beijing.

He argued that openness actually makes AI safer, as external researchers can help detect vulnerabilities, expose weaknesses, and build defenses. Relying on a single model globally would create a single point of attack and a single point of failure, making the entire system more vulnerable.

Applying the same logic, Huang urged Anthropic not to restrict its web model Claude Mythos but to open it to more users, improving security through testing and rapid patching. 'Restricting Anthropic brings no benefit to the United States,' he said.

Regarding U.S. Treasury Secretary Scott Bessent's claims that Chinese models steal American intellectual property and are considering sanctions, Huang argued that learning from other AI or knowledge sources is fundamental to intelligence development. While companies violating privacy or breaching contracts should be punished, policies should target specific violations rather than banning models outright.

On U.S.-China AI competition, Huang does not view it as a race with a finish line. He stated that the U.S. will exist long-term, China will exist long-term, and humanity will forever use AI. Therefore, if the U.S. blocks open-source models out of fear, it may not maintain its advantage and could instead make itself more vulnerable in AI development.