Nvidia CEO Calls AI Model Distillation Competition, Not Theft

Nvidia CEO Jensen Huang said AI model distillation is “competition”, rejecting claims from US officials that the practice amounts to intellectual-property theft. His comments differ from Treasury Secretary Scott Bessent, who warned in July that unauthorised distillation could trigger sanctions, and from CISA, NSA and FBI assessments that Chinese AI companies conducted industrial-scale extraction of models from Anthropic, OpenAI, Google and xAI. AI model distillation allows a smaller system to learn from the outputs of a larger model. The technique is widely used legitimately, but the dispute concerns unauthorised access, fake accounts and the use of API responses to train rival systems. Anthropic reported nearly 200 million exchanges across five suspected operations. It said Alibaba generated more than 151 million exchanges between May and July 2026 through over 3,500 accounts flagged as fraudulent to train the Qwen model family. Moonshot reportedly used 5,380 fake accounts over 10 days to route almost 300,000 Kimi user requests to Claude. Huang compared model distillation with reverse-engineering Nvidia hardware and argued that companies can stop suspected abuse by identifying customers and terminating service. Critics say this is difficult because operators use third-party aggregators, proxy services and grey-market “relay” platforms to conceal their identities. For traders, the dispute raises longer-term risks around AI regulation, US-China technology restrictions, API controls and cybersecurity, but has no direct impact on cryptocurrency prices.