Beijing startup's 2.8-trillion-parameter Kimi K3 model matched or beat top US AI systems, triggering a broad selloff across tech stocks and crypto
If you felt a tremor in your portfolio this week, you can trace it back to Shanghai. Moonshot AI, a Beijing-based startup that most retail investors couldn’t name two weeks ago, just released an AI model large enough to make Wall Street question everything it thought it knew about who’s winning the AI race.
The company unveiled its Kimi K3 model on July 16 at the World Artificial Intelligence Conference in Shanghai. The model features a 2.8-trillion-parameter open-weight architecture and a one-million-token context window, making it the largest open-weight model in the world. It reportedly matched or outperformed top US AI systems from OpenAI and Anthropic on multiple coding benchmarks, while offering significant cost advantages and free access for modifications.
The Nasdaq dropped approximately 1% in the wake of the launch. Shares of Nvidia and Intel both declined as investors suddenly started doing math on whether US companies are overspending on AI infrastructure.
The damage wasn’t limited to American companies either. Chinese AI firms Zhipu and MiniMax saw their stocks drop around 27% and 16%, respectively.
Bitcoin declined alongside the semiconductor stock downturn, in what market coverage has dubbed a “Kimi moment.” The pattern will look familiar to anyone who lived through the DeepSeek shock earlier in 2025, when another Chinese AI breakthrough sent both tech stocks and digital assets tumbling in tandem.
Moonshot AI was established in 2023, which means it went from founding to releasing the world’s largest open-weight model in roughly three years.
The company raised $2 billion in May 2026, achieving a valuation exceeding $20 billion in anticipation of a potential IPO in Hong Kong. Kimi K3 reportedly outperformed older models like Opus 4.8 and GPT-5.5/5.6 Sol on coding benchmarks. The fact that it’s open-weight, meaning anyone can access and modify it, adds another layer of competitive pressure. US AI labs have largely kept their most powerful models proprietary. Moonshot is giving its best work away.
This approach echoes Chinese leadership’s broader calls for international collaboration in artificial intelligence, though the competitive implications cut both ways. Open models democratize access, but they also make it harder for any single company to maintain pricing power.