NVIDIA's Jensen Huang on Open and Closed AI Models: Both Will Coexist Long-Term

Huoshan Finance reports that on July 26, NVIDIA CEO Jensen Huang, in an interview with Bloomberg’s podcast, stated that open-source and proprietary models will coexist long-term, serving different needs. Proprietary models are typically more convenient and lower in cost, but enterprises requiring control over data, specialized capabilities, service levels, or national sovereignty-related functions must still build their own AI based on open-source models. Huang emphasized that open-source models are critical for security, cybersecurity, innovation, startups, and technological sovereignty. Proprietary models are not inherently more secure due to their closed nature—they can still be jailbroken, stolen, or leaked internally; if their safeguards or sandbox designs are inadequate, they may also launch attacks on other companies’ systems. The AI industry cannot rely on a single point of defense but must build large-scale, distributed self-defense capabilities. It is incorrect to simplify AI’s future as a binary choice between open or closed models, and it is equally unfounded to assume that open-source models are inherently less secure. Huang cited Hugging Face’s security incident as an example: according to his understanding, when proprietary models failed to assist in diagnosing the issue, Hugging Face used the open-source model GLM 5.2 to identify vulnerabilities and intrusion points and rapidly resolved them. He believes this incident demonstrates that open-source models can provide essential self-defense capabilities and underscores the importance of diversity in AI technologies.