Four Chinese artificial intelligence startups are threatening the dominance of U.S. AI companies, the Nihon Keizai Shimbun (Nikkei) reported on the 24th.
According to Nikkei, Moonshot AI, DeepSeek, MiniMax, and Z.AI (Zhipu AI) are rapidly gaining users by leveraging open-source strategies and low usage fees. The combined valuation of the four companies reaches a total of 30 trillion yen (about 270 trillion won).
"Kimi K3," recently unveiled by Moonshot AI, received high marks in the comprehensive performance evaluation by U.S. AI assessment firm Artificial Analysis, ranking behind Anthropic's "Fable 5" and OpenAI's "GPT-5.6 Sol," and was recognized as the top-level model among Chinese AI systems. Ichiro Sato, a professor at Japan's National Institute of Informatics (NII), said, "Kimi K3 appears to have boosted performance by combining multiple small AI models to minimize the use of high-performance chips."
Moonshot AI was founded in 2023 by CEO Yang Zhilin. Born in 1992, Yang graduated from China's prestigious Tsinghua University before earning a doctorate in computer science at Carnegie Mellon University in the United States. He subsequently handled AI research and development at Apple, Meta, and Google DeepMind. The company's name was taken from an album by the British rock band Pink Floyd, and it is known that he is such a rock music enthusiast that all of the company's meeting rooms are named after rock bands.
Moonshot AI applied a new AI training technology to Kimi K3. Rather than computing all data in bulk, the method selects only important information to improve learning efficiency. When this technology was unveiled in March, Tesla CEO Elon Musk called it an "impressive achievement" on his X (formerly Twitter).
In an interview with the Chinese Communist Party's official newspaper People's Daily in May, Yang stressed, "Unlike the closed models of U.S. companies, Chinese AI can build an ecosystem through open source."
In fact, most Chinese AI startups have commonly adopted open-source and low-cost strategies. The leading company that drove this trend is DeepSeek. In January last year, the company surprised the global AI industry by unveiling a high-performance model without using large quantities of expensive AI chips. The usage fee for DeepSeek's latest model is only about one-twentieth that of Anthropic.
DeepSeek CEO Liang Wenfeng explained at an investor briefing in May, "Low costs are not simply a management strategy but the result of optimizing the AI model structure," adding, "We can train large-scale models with only limited computing resources." Born in 1985, Liang earned a master's degree from Zhejiang University and ran a quant investment firm before founding DeepSeek after seeing the growth potential of the AI industry.
Since DeepSeek's emergence, Chinese AI companies have successively adopted open-source and low-cost strategies. In January this year, MiniMax and Z.AI listed on the Hong Kong stock exchange. MiniMax is a company that develops AI models that generate text into voice and video, and it is focusing on expanding into overseas markets amid intensifying domestic competition in China. About 70% of its total revenue is generated overseas.
MiniMax founder and CEO Yan Junjie, a native of Henan Province, earned degrees from Southeast University and the Chinese Academy of Sciences before serving as a vice president at AI company SenseTime. He became convinced of the arrival of the AI era after seeing OpenAI defeat a gaming world champion with AI in 2019, and founded MiniMax in 2022.
Z.AI is a company founded by Tsinghua University faculty and focuses on enterprise AI services. Recognized for its technological capabilities, it received investments from Alibaba and a Beijing municipal state-owned fund, among others, and is called the "Chinese version of Anthropic." However, in January the U.S. Department of Commerce placed Z.AI on its list of restricted entities, citing ties to the Chinese military.
Experts assess that while the United States still leads in cutting-edge AI technology, the performance gap felt by general consumers is not large. Professor Sato said, "Chinese AI is widely used around the world, including in Japan," adding, "In general usage environments, users do not feel a significant performance difference from U.S. AI." Swiss financial firm UBS also forecast in a recent report that "Chinese AI companies, whose strength lies in low costs, will be at an advantage in expanding their global market share over the long term."
U.S. containment is cited as a problem that Chinese AI companies must overcome going forward. Nikkei noted, "The fact that securing the latest AI chips remains difficult due to U.S. chip export restrictions remains a challenge for Chinese AI companies," and pointed out, "In the case of Kimi K3, the per-token usage fee is cheaper than that of U.S. competitors, but the amount of tokens required for processing is large, so there are also points raised that the actual usage cost is not as low as expected."
