Jensen Huang, CEO of NVIDIA, has expressed opposition to U.S. efforts to restrict Chinese open-source artificial intelligence (AI) models.
In an interview on July 21 in Fort Worth, Texas, Huang described Chinese AI models as "very impressive" and stated that "high-performing open-source models should be utilized." When asked whether U.S. companies should be allowed to use Chinese AI models, he replied, "Of course, they should be allowed." This stance contrasts with concerns from the U.S. government and major AI firms about the technological competition posed by Chinese companies. OpenAI and Anthropic, both NVIDIA clients, have warned that the spread of Chinese open-source AI could threaten U.S. technological superiority.
Huang emphasized that low-cost open-source models could expand the market by increasing the number of AI users, asserting, "There is zero possibility that China will push U.S. companies out of the market."
His comments come amid heightened scrutiny in the U.S. regarding the recently unveiled Kimi K3 by Chinese company Moonshot AI. The K3 boasts performance close to cutting-edge models at a lower price, with publicly available weights for developers to download and modify.
Following the K3's release, concerns emerged that affordable and efficient AI models could reduce the need for large infrastructure investments, leading to sell-offs in semiconductor stocks, including NVIDIA. Huang remarked, "The market initially misunderstood the impact of DeepSeek, and now Wall Street is misjudging Kimi's influence. Free AI is beneficial for hardware, semiconductors, and data centers."
He dismissed fears that downloading Chinese AI models could create a "backdoor" for the Chinese government as a "misunderstanding," adding that companies can modify models to fit their environments and operate them in a secure space known as a "sandbox" to control access.
Huang also highlighted that open-source models allow external researchers to identify vulnerabilities and develop defenses, enhancing security. He stated, "If everything relies on a single model, the points of attack and failure also converge, making the world much more vulnerable."
Regarding Anthropic's restricted cybersecurity AI model, Claude Mythos, he noted, "Even without providing Mythos, other open-source models are already available. Anthropic should be allowed to compete freely; restricting their activities does not benefit U.S. interests."
Huang drew a line against viewing the AI competition as a zero-sum game between the U.S. and China, stating, "The AI race is not a competition with a finish line. Both the U.S. and China will exist for a long time, and we will continue to use AI in the future."
Huang's interview was released just hours after U.S. Treasury Secretary Scott Vessen announced an investigation into potential intellectual property theft by Chinese AI models. In an interview with Fox Business, Vessen stated, "If it is confirmed that foreign models have stolen technology from U.S. companies, we can impose sanctions," mentioning that a "watermark" from U.S. AI models was found in Chinese models.
Huang acknowledged the importance of learning through distillation and AI from other knowledge sources but asserted that companies violating privacy regulations or contracts should be held accountable.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.