The White House has leveled a new accusation against one of China’s fastest-growing AI startups, claiming Moonshot AI gained access to restricted Nvidia GB300 systems and may have used Anthropic’s technology to help build its latest model, according to a report from CNBC. The allegations land just days after Moonshot introduced Kimi K3, an open-source model that has drawn attention for narrowing the gap with leading U.S. AI systems.

If confirmed, the claims would add fresh urgency to Washington’s effort to keep advanced AI hardware out of China’s hands. They would raise fresh questions about whether export controls can keep pace with an industry where computing resources can be rented, shared, or accessed across borders.

The timing is also notable. The accusation surfaced only days after Kimi K3 climbed to the top of Arena.ai’s Frontend Code Arena leaderboard, overtaking Anthropic’s Claude Fable 5 in a benchmark widely watched by AI developers for real-world frontend development and agentic coding.

Moonshot unveiled Kimi K3 on Friday, positioning it as its most ambitious AI model yet. With 2.8 trillion parameters, the company says it is the largest open-source model released so far. Early benchmark results suggest Kimi K3 approaches the performance of frontier models from OpenAI and Anthropic, surpassing them in a handful of tests.

That performance has quickly turned Moonshot into a company to watch. It has now become part of a broader debate over how Chinese AI firms are obtaining the computing resources needed to compete with Silicon Valley.

White House alleges Moonshot accessed restricted Nvidia systems

Michael Kratsios, director of the White House Office of Science and Technology Policy, said Wednesday that the administration believes Moonshot obtained access to Nvidia’s GB300 hardware outside mainland China, CNBC reported.

The Chinese company “acquired GB300-equipped servers and has accessed GB300s in Thailand, likely to train its AI models,” Kratsios wrote in a post on X.

Nvidia, Moonshot AI, the White House, and the U.K. embassy representing the People’s Republic of China have been approached for comment.

The accusation centers on one of Nvidia’s most advanced AI platforms. GB300 systems remain subject to U.S. export restrictions that prevent direct sales to China. They sit one generation behind Nvidia’s upcoming Vera Rubin architecture, yet remain among the most sought-after AI computing platforms available today.

Export controls face growing pressure

Washington has spent years tightening restrictions on advanced AI chips destined for China. The strategy aims to slow China’s progress in training large AI models by limiting access to the enormous computing power required.

That effort has produced mixed results.

Chinese companies have increasingly relied on domestic chipmakers, overseas infrastructure, and cloud providers to keep building larger models. Reports earlier this year suggested ByteDance explored assembling AI computing capacity outside China to support its own development efforts.

The latest accusation against Moonshot fits a pattern U.S. officials have warned about for months. Instead of shipping restricted hardware directly into China, companies may seek access through foreign data centers or cloud infrastructure located elsewhere.

Congress is attempting to address that possibility through the proposed Remote Access Security Act, or RASA. The bipartisan legislation would expand export controls beyond physical shipments to include remote cloud-based access to restricted AI hardware and software. The bill passed the House earlier this year and is awaiting Senate action.

New allegations extend beyond hardware

Kratsios raised another allegation that could prove just as significant.

“We have information that Moonshot AI distilled Anthropic’s Fable for the development of its K3 model,” he said.

Model distillation allows developers to transfer knowledge from a large, highly capable AI system into another model. The process itself is widely used across the AI industry. The dispute centers on whether the underlying training methods violate another company’s terms of service or intellectual property rights.

Anthropic has previously accused Chinese organizations of conducting “industrial-scale” campaigns to extract knowledge from its models.

Treasury Secretary Scott Bessent signaled Wednesday that the U.S. government is prepared to respond more aggressively.

“When PRC firms conduct covert, industrial-scale distillation attacks that cross the line into IP theft, sanctions and Entity List designations will be on the table,” he wrote on X.

China’s AI race keeps accelerating

Moonshot’s emergence highlights how quickly China’s AI sector has advanced over the past year.

Chinese developers have released a steady stream of increasingly capable open-source models that compete with leading Western systems across coding, reasoning, mathematics, and general language tasks. That progress has fueled concern in Washington that export controls alone may no longer be enough to preserve America’s lead in AI.

Keegan McBride, director of science and technology at the Tony Blair Institute for Global Change, said the allegations reflect a broader reality shaping the global AI race.

“It is well known that China is using American compute both within the country and abroad,” McBride said. “It demonstrates just how far ahead the US is on compute. American compute is the foundation for the AI race today.”

Chinese officials have rejected accusations surrounding the country’s AI industry. A spokesperson for the U.K. embassy of the People’s Republic of China previously told CNBC that China “opposes baseless allegations and malicious smears against its AI development.”

The dispute comes at a moment when AI leadership is increasingly defined by access to computing infrastructure rather than model architecture alone. Frontier AI models demand massive clusters of advanced GPUs, making control over those systems as strategically important as the software running on them.

For Washington, the challenge extends beyond blocking chip shipments. It now includes preventing restricted computing power from being accessed anywhere in the world. For companies like Moonshot AI, that growing scrutiny is likely to become part of the cost of competing with OpenAI, Anthropic, and other U.S. AI leaders.

Moonshot AI Founder