Moonshot AI’s Kimi K3 hits chip stocks as DeepSeek fears return

A new AI model from China’s Moonshot AI pressured semiconductor and AI-linked shares Friday, reviving investor concern that competitive large language models may be built with less reliance on the most expensive chip stacks. For everyday investors, the move hit a key market theme of the past year: whether AI infrastructure spending can keep justifying elevated valuations across the chip sector.

Moonshot AI launched Kimi K3 on Thursday, describing it as a 2.8 trillion-parameter open-weight model. A parameter is a learned setting inside an AI model, and open-weight means the model’s weights are made available for others to use or inspect under stated license terms.

The market reaction was broad. Taiwan’s benchmark index fell more than 6% Friday, Japan closed 4% lower, and the Nasdaq Composite dropped 1.5%, its weakest session of the week, according to Decrypt.

Semiconductor stocks also broke technical levels watched by traders. The VanEck Semiconductor ETF, ticker SMH, fell below its exponential moving average support band for the first time since April, according to Decrypt. An exponential moving average, or EMA, is a trend line that gives more weight to recent prices. Decrypt reported that SMH was more than 20% below its late-June record high.

Why Kimi K3 drew market attention

Independent benchmark group Artificial Analysis ranked Kimi K3 near top U.S. models on its Intelligence Index, a composite measure covering reasoning, knowledge, mathematics and coding. Decrypt reported that K3 scored 57, ahead of Claude Opus 4.8 and GPT-5.5, and close to Claude Fable 5 and OpenAI’s GPT-5.6 Sol.

Decrypt also reported that K3 beat those models on some benchmarks at a lower price. The full model weights are scheduled to become public by July 27 under a Modified MIT license, according to the same report. That could make the system available to smaller labs without paying to train a comparable model from scratch.

The comparison investors reached for was DeepSeek. In January 2025, DeepSeek’s R1 model challenged the market’s assumption that frontier AI required the highest levels of spending on infrastructure and chips. Nvidia lost around $590 billion in market value in a single trading session after that release, according to Decrypt.

Analysts see a trend, not a one-off

Wall Street analysts cited by Decrypt said Kimi K3 fit a broader pattern of Chinese AI labs narrowing the gap with U.S. leaders. Bernstein analyst Robin Zhu called the release “confirmatory” and said it supported the firm’s view that AI’s “state of the art” is still moving fast and that Chinese AI can keep pace with global leaders and take share over time.

Morgan Stanley analyst Gary Yu described Kimi K3 as the result of steady progress rather than a surprise, according to a Yahoo Finance report cited by Decrypt. “K3 has received positive feedback globally, signaling an all-round catch-up of Chinese LLMs with U.S. leaders in model size, performance, and pricing,” Yu wrote.

Moonshot AI has major backing. Alibaba invested $1 billion in the company in 2024 at a $2.5 billion valuation, according to Decrypt. The startup is now valued at roughly $31.5 billion, Decrypt reported.

Moonshot also has prior exposure in U.S. developer circles. Decrypt reported in May that Cursor’s Composer 2 was found to be running on Kimi K2.5 without disclosure before Cursor’s team acknowledged the open-source base.

This story draws on original reporting from Decrypt.