Moonshot AI Kimi K3 sparks market shock

Moonshot AI’s Kimi K3, a 2.8‑trillion‑parameter mixture‑of‑experts multimodal model with a 1‑million‑token context window, was unveiled on 16 July 2026 and entered public testing on 18 July. The company confirmed that full model weights will be released on 27 July, enabling developers to run the system locally on at least 64 high‑end GPUs (≈45 kW) and requiring roughly 1.5 TB of storage.

Independent benchmarks placed Kimi K3 at the top of the Frontend Code Arena leaderboard, surpassing U.S. models such as Anthropic’s Claude Fable 5 and OpenAI’s GPT‑5.6 on coding tasks, and it scored 57 on the Artificial Analysis Index. Pricing is set at $3 per million input tokens and $15 per million output tokens, about 40 % cheaper to operate than comparable proprietary models, while still matching top‑tier U.S. pricing.

The launch triggered a sharp market reaction: the Nasdaq opened about 1 % lower, the Dow Jones fell roughly 400 points, and semiconductor‑related indices slumped, with the Bloomberg Asia‑Pacific Semiconductor Index down 6 % and other chipmakers slipping 2‑5 %. Stock declines were observed across major AI‑infrastructure firms, including Nvidia, Intel, TSMC and Chinese rivals such as MiniMax. Moonshot’s own Kimi subscription service briefly halted new consumer sign‑ups because of computing‑capacity strain.

Analysts dubbed the event a “Sputnik moment” for Chinese AI, noting that an open‑weight, high‑performance model sidesteps U.S. export controls and could reshape investment patterns in AI hardware and cloud services. Observers highlighted the model’s potential to narrow the AI frontier gap between China and the United States and to reshape global pricing dynamics for AI services. Chinese leadership framed the model as a public‑good initiative, emphasizing benefits for enterprises, universities and governments worldwide while warning of security risks from unauthorised modifications.