White House Accuses Moonshot AI of Violating US Export Controls

Export Control Allegations and Nvidia Hardware

The White House Office of Science and Technology Policy has publicly challenged Moonshot AI’s development practices. Director Michael Kratsios alleged in a social media post Wednesday that the Chinese firm bypassed U.S. restrictions to build the Kimi K3 system. Kratsios stated that Moonshot acquired GB300-equipped servers and has accessed GB300s in Thailand, likely to train its AI models, making the case that the company violated U.S. export control rules and companies’ terms of service to build its product.

These allegations strike at the core of U.S. efforts to limit China’s access to advanced semiconductor technology. The accusation suggests that Moonshot utilized infrastructure outside of China to circumvent export control rules and service agreements. The model’s release has reignited the debate over who leads the global AI race, drawing comparisons to when the DeepSeek Chinese model jolted Silicon Valley last year.

Kimi K3 Performance and Market Disruption

Despite the regulatory controversy, the release of Kimi K3 has disrupted the AI market. The Beijing-based company unveiled the model as the largest open-weight model released to date. It features a 1-million-token context window—enough to take in hundreds of pages at once—and handles both text and images natively. In overall rankings, the model sits below Anthropic’s Claude Fable 5 and OpenAI’s GPT 5.6 Sol, trailing the former in several areas of evaluation. However, it beat Claude Opus 4.8 and GPT 5.5 on coding and general agent tests.

Arena.ai, a platform that ranks models on blind human comparisons, placed Kimi K3 ahead of Claude Fable 5 on its Frontend Code Arena leaderboard. In these blind tests, developers picked K3 over every leading US model on front-end coding tasks, and it matched GPT 5.6 Sol in the general text category. Vercel CEO Guillermo Rauch noted on social media that it was the first time that an open model is ahead of all proprietary ones for this comprehensive web engineering benchmark, while warning that benchmarks don’t always tell the full story. Wharton professor Ethan Mollick called the model closest to the frontier yet while also urging caution regarding benchmark scores.

For more on this story, see Anthropic Accuses Alibaba of Stealing AI Model Data.

Pricing Strategies and Investor Reaction

Moonshot has adopted an aggressive pricing model to compete with U.S. labs, positioning Kimi K3 as one of the cheaper frontier models. The Kimi K3 API costs $3 per a million input tokens and $15 per a million output tokens, with cheaper rates for cached inputs. By comparison, OpenAI charges $5 and $30 for GPT-5.6 Sol, while Anthropic charges about $10 and $50 for Claude Fable 5.

The financial impact of the release was immediate in Asian markets. Z.ai fell as much as 30% in Hong Kong trading, its worst single-day drop since listing in January. MiniMax Group dropped as much as 16%, while Alibaba, one of Moonshot’s backers, fell 4%. Nasdaq 100 futures also fell 2%. Bank of America analyst Alex Liu noted that the model raises the capability ceiling for China AI models, shifting the burden of proof to other independent AI labs. David Sacks, a tech advisor to the Trump administration, called K3’s capabilities concerning and cautioned that heavy AI regulation could cost the U.S. ground.

Regulatory Tensions and Future Outlook

President Xi Jinping recently made his first appearance at China’s top AI summit, telling nations that the technology should not be a solo performance by a single country. U.S. policy remains distinct: OpenAI recently widened public access to GPT-5.6 Sol after the Trump administration cleared it, while Anthropic’s Mythos 5 remains limited to a small set of American organizations under a Commerce Department export control order.

Founded in 2023, Moonshot raised $2 billion in May at a valuation above $20 billion. The next critical juncture arrives on July 27, when the company plans to release the full model code for public download and self-hosting. The public and developers are now waiting to see if the Kimi K3 model holds up according to reported benchmarks.