Nvidia's Jensen Huang defends Chinese AI amid Kimi panic

Summary

In this Axios interview, Jensen Huang pushes back on fears that Chinese open AI models are inherently unsafe, saying users can download them, fine-tune them, add enhancements, and apply guardrails. He argues that closed models should be used wherever possible, but open models are still essential for science, cybersecurity, national security, and economic security. Huang specifically says Chinese models are excellent and says the market misunderstood the impact of DeepSeek and Kimi. He frames open models as complementary to closed models rather than adversarial, and says stronger AI of any kind will increase usage, which in turn drives demand for Nvidia computers, data centers, services, and broader technology diffusion.

Notes

• Huang argues for a mixed ecosystem: use closed models where possible, but keep open models available.
• He says open models matter for science use.
• He says open models matter for cybersecurity.
• He ties open models to national security.
• He ties open models to economic security.
• He rejects the idea that Chinese models are inherently backdoored.
• He says downloaded models can be fine-tuned by the user.
• He says users can enhance downloaded models.
• He says users can add guardrails to downloaded models.
• He says the world needs both open and closed models.
• He says Chinese open-source models that are excellent should be used.
• He names DeepSeek as a case the market got wrong.
• He says the market again misunderstood Kimi.
• He says great open models are good for the entire industry.
• He says AI quality drives usage regardless of where the model comes from.
• He says more AI usage means more Nvidia computer sales.
• He says more usage implies more data-center construction.
• He says more usage expands Nvidia services.
• He says better models diffuse technology into more industries.
• He summarizes the loop as models, then use, then growth.
• He rejects the idea that open models are hostile to closed models.
• He frames open and closed models as complementary, not zero-sum.
• He argues model choice should be pragmatic rather than ideological.
• He implies open-source access can be controlled locally without relying on the original vendor.
• He suggests open models can be made safer through post-download controls.
• He says the open-model issue is also relevant to the United States and the broader world.
• He presents open models as useful infrastructure for research and safety-sensitive work.
• He argues the debate over Chinese models is being overstated.
• He implies broader AI adoption is a demand catalyst for Nvidia's compute platform.
• He links model quality to technology diffusion across sectors.
• He says the first misunderstanding was DeepSeek.
• He says the second misunderstanding was Kimi.
• He treats open models as part of overall industry growth rather than a threat to it.
• He says there is no inherent connection between downloading a model and a China-linked security risk.
• He says users can apply their own safety policies after downloading.
• He places open models in a national-security context.
• He places the open-model issue in an economic-security context.
• He emphasizes that stronger models increase demand for compute-heavy infrastructure.
• He says open models can accelerate adoption rather than cannibalize it.
• He argues the market has not fully priced the significance of open Chinese models.
• He suggests the same pattern repeated with Kimi.
• He centers model quality as the driver of downstream growth.
• He argues users should retain the freedom to modify models after download.
• He suggests the industry should not treat open models as a strategic mistake.
• He indicates that both enterprise software and open research models have roles to play.
• He argues the open-model debate spans the United States and the world.
• He says excellent open-source models should be adopted, regardless of origin.
• He says industry benefits when the best models are widely used.
• He links greater AI use to broader service expansion.
• He says the growth mechanism starts with model quality.
• He argues closed models remain valuable even if open models improve.
• He portrays open models as a necessary complement to closed products, not a substitute.
• He says open models can be used safely after local adaptation.
• He argues the full AI stack benefits when model performance improves.