Nvidia CEO Jensen Huang says AI model distillation is competition, not theft

17:30

By: Azzat Manal

Nvidia CEO Jensen Huang says AI model distillation is competition, not theft

Nvidia CEO Jensen Huang has pushed back against describing artificial intelligence model distillation as theft, arguing that companies should be able to study and compete with rival technologies as the AI industry enters an increasingly intense phase of competition.

Speaking to CNBC, Huang said he viewed model distillation as part of normal competition in the technology sector. The technique involves using the outputs of a more capable “teacher” model to help train a smaller or more specialized “student” model, potentially allowing developers to reproduce some capabilities while using fewer computing resources.

The issue has become increasingly contentious as companies and governments debate where legitimate competitive research ends and unauthorized use of proprietary technology begins. Distillation itself is a widely used technical method, but concerns arise when developers use large quantities of outputs from closed commercial models without authorization.

Huang argued that companies routinely test and analyze competing products to understand their performance. He acknowledged that Nvidia would not necessarily welcome competitors examining its own technologies, but said competition can encourage companies to improve their products and develop new capabilities.

His comments come amid growing tensions between the United States and Chinese AI companies. In September, U.S. officials accused several Chinese firms, including DeepSeek, Moonshot AI and Alibaba, of using distillation to copy or extract capabilities from advanced American AI systems. The accusations have been rejected by China, which has argued that distillation is a neutral technology used widely in AI development.

The controversy has also involved allegations concerning models developed by major U.S. AI companies. American officials and technology firms have raised concerns that proprietary outputs from systems developed by companies such as Anthropic and OpenAI could be used to accelerate the development of competing models without authorization.

Reuters has reported that the dispute has become part of the broader U.S.-China technology rivalry, with Washington considering restrictions or sanctions against some Chinese AI companies over alleged intellectual property violations. Beijing has criticized such measures and accused the United States of using technology policy to limit competition.

Huang's position highlights a fundamental question facing the rapidly developing AI industry: whether the ability to learn from a competing model should be treated as a normal feature of technological competition or as an unauthorized transfer of valuable intellectual property. The answer may depend on how models are accessed, the terms governing their use and the amount and nature of information extracted from them.

Companies seeking to prevent their systems from being used for unauthorized distillation can impose contractual restrictions, monitor usage and limit access to their services. Such measures are increasingly important as advanced AI models become more capable and their outputs become more valuable for training other systems.

The debate is likely to intensify as smaller and more efficient AI models become increasingly important to the industry. Distillation can reduce computing requirements and make advanced capabilities easier to deploy, but disputes over ownership and permissible use could also influence how companies design access policies and how governments approach AI regulation.

For now, Huang's comments add a prominent industry perspective to a dispute that extends beyond technology. The debate over AI distillation is increasingly tied to questions of intellectual property, market competition, national security and the future rules governing access to advanced artificial intelligence.