China's Moonshot AI Kimi K3 Shakes US Markets
2026-07-18
US stocks fall on Kimi K3
On the 17th in the US stock market, Moonshot AI's new model Kimi K3 triggered declines, sending Alphabet, the parent of Google, down 2% from the previous day and Meta down 3%. The Nasdaq Composite, which has a heavy weighting of tech shares, fell 1.4%, while the Philadelphia Semiconductor Index (SOX), made up of chip-related stocks, dropped more than 20% from its June 22 peak, entering a bear market. With the view that low-cost AI from China could rattle markets, some are warning of a repeat of the DeepSeek shock that moved tech stocks in January 2025.
Ranks in top three
When US research firm Artificial Analysis ranked models across areas including programming, mathematics and science, Kimi K3 came in third behind Anthropic's latest model Fabble and OpenAI's GPT-5.6. Ethan Mollick, professor at the Wharton School of the University of Pennsylvania, said on X, formerly Twitter, that while it does not match the top models from OpenAI or Anthropic, it stands out as an open model that makes its technology publicly available. Kyle Chan of the Brookings Institution said Chinese AI models had significantly narrowed the technology gap with the United States and noted that China's AI development has not slowed despite US semiconductor export controls. David Sacks, who has influence over AI policy in the Trump administration, also warned that the model had scored at or near frontier levels and said it was a matter of concern.
The gap and the cost
In the United States, the prevailing view had been that Chinese AI companies would catch up with the US in six to 12 months. The arrival of Kimi K3 suggests the development gap may be smaller than US expectations and indicates that Chinese players are becoming more competitive. Even so, US companies still hold the edge in overall performance, and Kimi K3 outperformed Fabble only in a limited number of benchmarks, such as programming metrics that measure development capability for websites and other applications. If Kimi K3 truly matches Anthropic's Opus 4.8, released in May, or OpenAI's GPT-5.5, released in April, then China would have caught up with the US in about three months, at least for publicly available AI models.
Low-cost claims and distillation doubts
On the low-cost advantage often cited by Chinese players, some say Kimi K3 is not cheap to use. Prominent tech investor Gavin Baker said Kimi K3 costs more than GPT-5.6. Although the price per token is far lower than that of US AI models, it reportedly consumes more tokens for the same task. In February, Anthropic said Moonshot AI used a method known as distillation, in which data is extracted from another high-performance AI, in developing Kimi. Moonshot AI allegedly created hundreds of fake accounts and interacted with Anthropic's Claude more than 3.4 million times. While the US seeks to wall off frontier AI over fears of cyberattacks, China is aiming to build open models at low cost and spread domestically developed AI overseas. As the US tries to slow Chinese developers through measures such as bans on advanced semiconductors, China is countering by promoting domestic chip production. Semiconductor analyst Jim Handy said that even if Chinese AI models become more competitive, they will not be able to maintain the pace unless they catch up with the US in advanced semiconductors, and he added that the US can also adopt Chinese development methods.
Enjoyed this article? Share it with your network!