Moonshot released Kimi K3, a 2.8-trillion-parameter model, prompting broad reaction across global technology markets and intensifying debate over the competitive positioning of Western AI labs1.
Chip stocks entered a bear market on Friday, driven in part by questions about what the scale and pricing of Kimi K3 imply for compute economics. NVIDIA rebounded modestly on Monday, with attention shifting to the intense memory requirements of models at this parameter count. Moonshot said it is limiting new subscriptions while it works out additional capacity.
The model is described as open source and lower cost than OpenAI's closed models3. Alibaba Group Holding also unveiled a new model version with 2.4 trillion parameters, adding to a wave of large-scale Chinese AI releases.
Sameer Dholakia, a partner at Bessemer Venture Partners and an investor in Anthropic through the firm, said he views the development positively2. "I believe that the right tool for the job is the right answer for our enterprises, for customers, and they will use FrontierLab models like Anthropic for the stuff that they need to use that for, and they will use open source models from China and other places in the world over time," Dholakia said. He argued that free competition with Chinese models will push Western open-weight models to improve.
The release has prompted pointed questions about the moats of leading Western labs. Bloomberg Tech host Ed Ludlow asked Dholakia directly: if a startup of Moonshot's size and resources can produce a 2.8-trillion-parameter model at its price point, what justifies the valuations assigned to Anthropic or OpenAI?
Moonshot's founder, who is 33 years old, graduated from Carnegie Mellon and returned to China after studying in the United States. He named the company in Chinese after Pink Floyd's "Dark Side of the Moon". Bloomberg described him as a rising star in the AI universe before Deep Seek's emergence.
Moonshot is preparing for a Hong Kong IPO, reportedly within the next six months. The company's valuation is above $30 billion. Moonshot was previously reported to be planning a pre-IPO funding round targeting a pre-money valuation of up to $50 billion with talks expected to begin in August ctx. Bloomberg separately reported that Deep Seek is looking at a mainland listing sometime in 2027.
The broader open-versus-closed debate gained traction online, with a Hacker News post titled "China's open-weights AI strategy is winning" drawing 1,161 points and 869 comments.
ANALYSIS The simultaneous release of trillion-parameter-scale models by both Moonshot and Alibaba Group Holding, priced below Western closed-model equivalents, puts direct pressure on the pricing and differentiation strategies of Anthropic and OpenAI. Dholakia's framing — that enterprises will use frontier lab models for some tasks and open-source alternatives for others — suggests that even investors in Western labs see a segmented market rather than winner-take-all dynamics.
Moonshot's capacity constraints, evidenced by its decision to limit new subscriptions, indicate that demand has outpaced infrastructure, a factor that could moderate the near-term competitive impact even as the model's technical specifications draw attention.
Homin Lee, senior macro strategist at Lombard Odier, noted that the China AI space could see further positive developments this week.