Back to News
PolicyAI Understanding briefing

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

Nvidia chief Jensen Huang told CNBC that training AI models on the outputs of competitors’ systems – known as model distillation – is a form of competition, pushing back against U.S. officials who have labeled the practice as theft and are considering sanctions against Chinese firms.

4 min readRead the original reporting
Source-page capture accompanying Nvidia CEO Jensen Huang says AI distillation is competition, not theft
Attributed reportingSource recorded
Publisher
cnbc.com
Source link
cnbc.comhttps://www.cnbc.com/2026/09/28/nvidias-jensen-huang-ai-distillation-china.html
Source type
Reporting by a news outlet — not a first-party document.

What we could not confirm independently: This claim is attributed to the named outlet. We did not verify it against a first-party document. (cnbc.com)

ContextUnderstand this in 60 seconds

Start here

Key terms

Distillation
Compressing knowledge from a large teacher model into a smaller student model.
API (Application Programming Interface)
A structured way for one software system to send requests to and receive responses from another system.
Knowledge Distillation
Training a smaller model to imitate the outputs of a larger model.
Test yourselfAI Models Explained Quiz

What happened

Jensen Huang appeared on CNBC’s Squawk Box and rejected the U.S. Treasury’s characterization of AI model as “theft.” He said the practice of training models on other models’ outputs is simply competition, noting that Nvidia’s own products are often dissected by rivals. The remarks come amid accusations from U.S. officials that Chinese AI companies are using large‑scale distillation to extract capabilities from U.S. models, prompting discussions of possible sanctions.

During a live interview on Monday, Sep 28, 2026, Nvidia CEO Jensen Huang was asked whether AI model —training a new model on the outputs of an existing one—constitutes theft. Huang responded that it is "competition" and that companies are free to test each other's products.

He acknowledged that Nvidia’s own hardware and software are sometimes stripped down to their core components by competitors seeking to understand how they work, adding that while he would prefer others not learn from Nvidia’s products, competition ultimately benefits the industry.

Huang’s comments directly counter remarks made earlier by Treasury Secretary Scott Bessent, who in July described as "theft" and warned of potential sanctions against overseas firms that use the technique to extract capabilities from U.S. models.

The interview also referenced recent accusations from the U.S. Cybersecurity and Infrastructure Security Agency that Chinese AI firms are conducting "industrial‑scale campaigns" in violation of U.S. terms of use. Anthropic has publicly named Alibaba’s Qwen models and DeepSeek as engaging in illicit .

The White House did not immediately comment on Huang’s statements, leaving the policy debate open.

Source details: cnbc.com ↗

Why it matters

The interview spotlights a growing policy clash over AI intellectual property and national security. Model can accelerate AI development, but U.S. officials argue it violates terms of service and may transfer critical technology to adversaries. Huang’s framing of the practice as competition challenges the narrative that such activity is illicit, potentially influencing future regulatory approaches and industry norms. The dispute also underscores broader U.S.–China tensions in the race for AI supremacy, with possible repercussions for global AI supply chains and the enforcement of export controls.

Model is a powerful method for rapidly building high‑performing AI systems, especially for firms lacking the data or compute to train models from scratch. If U.S. regulators treat it as theft, they may impose export controls or sanctions that could limit the flow of AI technology and affect global competition.

Huang’s framing of as competition challenges the premise that the practice is inherently illicit, suggesting that industry norms may evolve toward more open sharing of model outputs, provided appropriate licensing is in place.

The dispute reflects a broader strategic rivalry between the United States and China over AI leadership. Policy decisions made now could set precedents for how emerging AI techniques are regulated internationally, influencing trade, security, and innovation ecosystems.

For Nvidia, the stance has practical implications: if the company chooses to add technical barriers—such as watermarking or usage restrictions—to its models, it could affect how partners and competitors interact with its technology.

Interactive Mechanism

Interactive Mechanism: How It Actually Works

Explore the underlying technology behind this development interactively.

Model Parameter Size:8B Parameters
VRAM Required5.5 GBGPU memory footprint
Target HardwareMacBook / Single GPUDeployment tier
Privacy100% Air-GappedLocal device capability
Core takeaway: Small, quantized models (3B–8B) now run directly inside smartphones and laptops with complete data privacy, while mammoth 400B+ models remain the domain of datacenter clusters.
Interactive Concept Check+10 Points
AI Models Explained Quiz

In AI, what are a model's "parameters"?

What to watch next

Watch for any formal policy actions from the White House or Treasury Department targeting AI , especially sanctions against Chinese firms. Monitor Nvidia’s response—whether it tightens licensing or adds technical safeguards to limit model extraction. Follow statements from other AI developers, such as Anthropic, which have already accused specific Chinese companies of illicit distillation, to gauge industry consensus. Finally, observe how the debate shapes broader international discussions on AI governance and intellectual‑property enforcement.

Potential regulatory moves: The Treasury and the White House may issue new guidance or sanctions targeting AI , especially if evidence of large‑scale extraction by Chinese firms grows.

Nvidia’s product strategy: The company might introduce licensing terms, technical safeguards, or API controls to limit how its models can be queried and reverse‑engineered.

Industry response: Other AI developers, including Anthropic, OpenAI, and emerging Chinese firms, may issue statements or adjust their own policies on model sharing and licensing, shaping a de‑facto standard for the practice.

International dialogue: Multilateral bodies such as the OECD or the G7 could convene to discuss AI intellectual‑property norms, potentially leading to coordinated policy frameworks.

Related guides & quizzes

AI Models ExplainedAI EthicsFuture of AITest what you know — try a free AI quizLook up an AI term in our glossaryFollow the AI regulation tracker
Found this useful?