How the new open-weight model is rattling Western tech markets
China’s Moonshot AI, co-founded by former Google and Meta researcher Yang Zhilin, shook up western markets last week with the launch of its impressive new model called the Kimi K3. Now the company is reportedly planning to go public at a valuation of over $43 billion.
The company says its vulnerability-heavy AI model outperforms all but the strongest competitors; Anthropic’s Claude Fable 5 and OpenAI’s GPT-5.6. Demand for the massive 2.8 trillion-parameter model was so great over the weekend that Moonshot had to temporarily pause recording. Users were pleased with its coding and natural language capabilities and compared its long context and efficiency to the most popular models developed by the US.
The global selloff announced last week, affecting chipmakers and big tech, has wiped out trillions in market value. Bloomberg and others report that Moonshot notified investors in June that it had reached US$300 million ($430 million) in annual recurring revenue and has completed a financing round that could value it at more than US$30 billion ($43 billion) before going public in as little as six months.
Although only in his early thirties, Zhilin holds a PhD in computer science from Carnegie Mellon University, is a serial startup founder, and has worked at Google, Meta, and Huawei. In 2023, he co-founded Moonshot. So far it has largely lived in the shadow of other Chinese AI startups DeepSeek and Z.AI, but long-memory chatbot Kimi started gaining traction earlier this year. With K3, the company broke into mainstream attention.
The model is open weight technology with the best performance yet to appear in China. Open weight refers to an AI model whose trained files are available for anyone to download. It is not a completely open source in the sense that users do not have insight into how the model was trained. But openweight models (like Moonshot’s K3 and Meta’s Llama) differ significantly from closed offerings from major tech companies because anyone can run, host, or customize them. OpenAI, Anthropic, and Google only offer access to their models through paid APIs.
The market reaction to Moonshot’s news mirrors the shockwaves we’ve seen before, with temporary sell-offs in Western semiconductor and AI tech stocks whenever a Chinese lab makes a major AI breakthrough, with investors reconsidering their assumptions about how broad the US lead in AI technology actually is.
But Moonshot’s action is quite different from the early DeepSeek releases that caused immediate panic. DeepSeek’s focus on ultra-low prices raises fears of democratization and rapid commoditization, while Moonshot shows the open source gap is narrowing faster than expected. Developers now have access to advanced reasoning and coding capabilities without having to lock themselves into closed APIs. The company’s reported revenue shows that businesses are willing to pay to develop it.
But that doesn’t mean Moonshot is cheaper or easier to run than other mid-tier enterprise AIs, including major western ones. The K3 is a huge model that requires serious hardware to accommodate; You cannot download and run it on a regular computer. Consumer users will use Moonshot’s Kimi chatbot to access the model, which comes with a tiered subscription. Companies and developers will rent access to servers or custom installations, and the Chinese company’s API pricing is also on par with its rivals. Of course, licensing deals stop any other company from building K3 and selling access to Moonshot without a cut.
History shows that the market will settle as investors realize that a rising tide lifts all boats and that increased competition and a multipolar market for AI could ultimately be good for Nvidia, TSMC, and even rival AI firms. However, K3 proved that open models can rival closed models when used in high-value tasks such as coding and agents.
Get news and reviews on tech, gadgets and games in our Tech newsletter every Friday. Sign up here.

