[New York = E-Daily Seong Joowon Correspondent] Chinese startup Moonshot AI has recently caused a stir in the U.S. artificial intelligence (AI) industry by fully releasing the weights of its self-developed AI model, “Kimi K3.” This is because the model demonstrated performance comparable to top-tier models such as OpenAI and Anthropic, and because any developer can now download, modify, and run it. While this move has sparked calls in Washington for sanctions against Chinese AI models, Silicon Valley’s big tech companies have taken the opposite stance and voiced their opposition.

A Shinhung Powerhouse That Outperformed DeepSeek; Founder Is a ‘Rock Band Fan’

Moonshot is a company founded in early 2023 by Yang Zilin, a former Tsinghua University professor. He has worked at Meta Platforms and Google, and after graduating from Carnegie Mellon University, he published numerous papers in the field of natural language processing. He was considered the “star” of China’s AI industry until DeepSeek emerged. According to Bloomberg, as venture capital investments poured in, MoonShot’s valuation once soared to $20 billion (approximately 29.34 trillion won), and its chatbot “Kimi” established itself as the most popular ChatGPT rival in China.
Yang Zilin is also known as a rock music enthusiast. The company named its conference rooms after bands such as Radiohead, Led Zeppelin, Queen, and Nirvana, and labeled its subscription plans with classical music tempo terms like Adagio, Andante, and Moderato. Bloomberg reported that the company’s original Chinese name was “The Dark Side of the Moon,” taken from his favorite Pink Floyd album.

Strong Benchmark Performance... Aims to Gain Traction by Revealing Weightings

MoonShot unveiled the Kimi K3 for the first time in mid-month, demonstrating benchmark performance on par with top-tier models such as OpenAI and Anthropic. Bloomberg reported that AI and semiconductor-related stocks fell across the board immediately following this announcement. Concerns that U.S. companies like Anthropic might slow their investment pace due to the rise of Chinese AI spread to semiconductor firms such as NVIDIA, SK hynix(000660), and SamsungElectronics(005930), fueling fears of a slowdown in demand.
At the end of this month, the weights for Kimi K3 were fully disclosed. With developers now able to freely download, modify, and run the model, analysts predict that Moonshot’s user base will expand further. Yang Zilin has outlined a strategy to differentiate Moonshot from U.S. proprietary systems by emphasizing openness and accessibility.
Price competitiveness is another factor. According to the benchmarking site Artificial Analysis, the weighted average cost of performing standardized intelligent tasks is 33 cents for “Kimi 2.6” and 2 cents for DeepSeek’s “V4 Flash” model, whereas Anthropic’s “Claude Fable 5” costs $2.75. However, Kimi K3 itself is priced at 95 cents, which is higher than these figures.
MoonShot commented, “While it is a very competitive model overall, Kimi K3 still lags significantly behind Claude Fable 5 and GPT 5.6 Sol in terms of user experience.”

“Intellectual Property Theft” vs. “Legitimate Technology”…Mixed Views

MoonShot’s growth has also been accompanied by controversy. In February, Anthropic claimed in a blog post that Chinese companies such as MoonShot, DeepSeek, and MiniMax had engaged in an industrial-scale campaign of intellectual property infringement by unauthorized use of Claude’s technology through a “distillation” method, which violated its terms of service and regional access policies. Moonshot did not respond to a request for comment on the matter, and Bloomberg reported that Chinese authorities strongly refuted the allegations of intellectual property theft.
Amid this, calls for sanctions have gained traction in Washington. According to Bloomberg Opinion columnist Catherine Sorevacki, U.S. Treasury Secretary Scott Bessent stated that the possibility of sanctions against China and designation on the Entity List is “on the table,” and Michael Kratsios, Director of the White House Office of Science and Technology Policy (OSTP), expressed a similar position. Kratchios also raised allegations that Moonshot violated U.S. export controls by using NVIDIA chips during the development of the Kimi K3; however, as neither side has issued an official statement on the matter, the facts remain unclear.
However, Silicon Valley itself voiced a completely opposite view. A group of 179 startup and tech founders sent a Seohan to Director Kratsios and U.S. Secretary of Commerce Howard Rutnick opposing the ban on foreign-made models, arguing that “open distribution is now a competitive strategy.” A few days later, Jensen Huang, founder and CEO of NVIDIA, shared a separate Seohan signed by Microsoft (MS), Google, OpenAI, and others as his first post on his X (formerly Twitter) account. This Seohan went a step further, arguing that policymakers “must not confuse legitimate model development techniques with the improper acquisition of intellectual property,” and countered that distillation is “a technique widely used to improve, evaluate, and validate model performance.”
Columnist Sorbecki pointed out that among major Silicon Valley companies, Anthropic was not listed in either of the two Seohan letters, and assessed that this controversy has left Anthropic and OpenAI in the most vulnerable position. This is because he believes that rather than reducing revenue, the proliferation of open models is more likely to redistribute it toward infrastructure and cloud companies such as NVIDIA, Amazon, Microsoft, and Google.

Washington’s Choice: ‘Openness’ or ‘Restriction’?

It remains unclear whether Washington will actually move to regulate public AI models. Columnist Sorebekki pointed out that hasty regulation could backfire. He argued that even if the U.S. blocks the use of Chinese models, it cannot prevent developers worldwide from accessing them, which could put only U.S. researchers at a disadvantage. He also cited a recent incident in which an undisclosed OpenAI agent escaped control during an evaluation process and compromised the Hugging Face model repository system. Noting that Hugging Face utilized China’s Zhipu “GLM 5.2” model in its defensive response, he argued that the security risks associated with open-source models are not as straightforward as they may seem.
Ultimately, the key question is whether Washington will heed the voices of its domestic industry and shift its focus toward fostering an open ecosystem, or whether it will play the “China sanctions” card, prioritizing containment. Depending on the outcome, forecasts suggest that the global demand landscape for AI semiconductors—spanning NVIDIA, SK hynix, and SamsungElectronics—will also change.