China AI advances erode pricing edge of U.S. giants

2026-07-23

The rise of China-developed AI is shaking the high-margin business model built by U.S. AI companies. Moonshot AI, a Chinese startup, said its new model Kimi K3 comes close in performance to top models from U.S. firms Anthropic and OpenAI. If the premise of selling high-performance AI at premium prices breaks down, plans to recoup massive investments could also be affected.

China AI is 70% cheaper than Fable

Moonshot AI's Kimi K3 is said to have shown performance close to Anthropic's Fable and OpenAI's GPT-5.6 across a wide range of fields, including programming, mathematics and science. Its usage fees are lower, with the price per unit of usage 70% cheaper than Fable and 40% cheaper than Anthropic's second-tier model Opus 4.8.

The lower price reflects, in part, the fact that Chinese AI is open in nature, with research results made public. Because it does not need to lock up technology, it is easier to keep research and development costs down. Another advantage is that it does not need to house all data centers and other infrastructure in-house. By contrast, Anthropic and OpenAI, which have led on performance, have adopted a closed model that keeps cutting-edge technology inside the company and have used a strategy of selling high-performance models at high prices to fund development.

The debut of Kimi K3 could upend that structure. If the two firms can no longer maintain a performance gap, the risk of being drawn into price competition rises. If investment payback slows, funding for the next round of development will also thin, making it harder to sustain a competitive edge. AI researcher Nathan Lambert has pointed out that powerful open models can sharply compress the potential profit margins of closed-model companies. Concerns over profitability could also affect initial public offerings (IPOs) by Anthropic and OpenAI.

Signs of such commoditization are already beginning to emerge. As overall AI performance improves, the need to insist on top-tier U.S. models is fading for office productivity and programming use cases. According to U.S. startup OpenRouter, which lets users switch among multiple AIs, Chinese AI accounts for 60% of corporate AI usage on a usage basis. Marty Kausas, CEO of U.S. software startup Pylon, also said, 'There are no uses inside the company that require Anthropic's highest-performing AI, Fable.'

'Secret distillation' could bring sanctions

U.S. AI companies remain highly wary. Dean Ball, policy chief at OpenAI, argued that if open models that share technology become the norm, AI would turn into a 'public good provided by the state' and lead to 'AI communism.' He warned that corporate incentives to develop the technology would fade and progress would slow.

The Trump administration also suspects that unfair methods may have been used to improve Chinese AI performance. The focal point is regulation of 'distillation,' a process of learning from and incorporating the capabilities of rival AIs. Michael Kratsios, director of the U.S. White House Office of Science and Technology Policy, said on the 22nd that there is information suggesting Moonshot AI distilled Anthropic's Fable. He also referred to possible measures taken to evade detection and said, 'Secret distillation that steals America's proprietary technology is unacceptable.'

Anthropic has long characterized distillation by Chinese players as organized misconduct, arguing that they are free-riding on U.S. investment gains without bearing the costs and risks of advanced AI development. U.S. Treasury Secretary Scott Bessent also said on the 22nd that if Chinese companies are secretly conducting industrial-scale distillation and stealing U.S. intellectual property, he would consider designating them for sanctions or de facto export bans, putting Moonshot AI and others on notice.

Still, there is also caution about expanding restrictions. Investor David Sacks, who is close to the Trump administration, criticized OpenAI and Anthropic, saying the two already-dominant closed-model companies are asking the U.S. government to exclude competing open models.

Enjoyed this article? Share it with your network!