Chinese AI lab Moonshot aims to double revenue to $2 billion by year-end

Moonshot AI is targeting $2 billion in annualized revenue by the end of the year, Bloomberg reported Friday. That figure would double the revenue run rate the Chinese AI lab reported for August, a measure of yearly income based on a recent month’s pace. The goal reflects strong demand for K3, the open-weight model Moonshot released this summer. An open-weight model is one whose underlying parameters are published for anyone to download and run.

Usage data backs up the ambition, at least for now. OpenRouter, a marketplace that routes requests to different AI models, currently shows K3 models generating as many as 300 billion tokens a day. Tokens are the small chunks of text a model reads and writes, and daily volume at that scale points to heavy use by developers building on top of K3. Bloomberg’s report noted that K3’s usage figures have slipped slightly in recent months, even as the daily token count on OpenRouter remains high.

Moonshot’s projected revenue is still far smaller than that of the leading closed-weight labs. Recent reports put OpenAI’s revenue at $40 billion and Anthropic’s at $65 billion. Because Moonshot gives its model weights away for free, it earns far less per user than labs that keep their models closed, which limits its margins. Even so, the growth in Moonshot’s numbers suggests open-weight AI companies can still build a real business, just a smaller one than their closed-weight rivals.

A distillation dispute

Moonshot’s rise has come with controversy. Earlier this week, Anthropic accused the company of running a long-term distillation campaign against its models. Distillation is a technique where one company trains a new model by copying the outputs of a rival’s model rather than building from scratch. According to Anthropic, nearly 300,000 requests sent to Kimi, Moonshot’s chatbot, were quietly routed to Anthropic’s Claude Opus model instead, so that Opus effectively answered on Kimi’s behalf. Anthropic said more than 23 million responses were collected from its models in total and used to help train Moonshot’s own systems.

Moonshot has not issued a public response to the allegations in the source reporting reviewed here. Anthropic’s terms of service, like those of most frontier AI labs, restrict using their models’ outputs to train competing systems, and this would not be the first time a Chinese lab has faced such accusations.

What happens next will hinge on two separate tracks. One is commercial: whether Moonshot’s token usage on platforms like OpenRouter holds up or keeps declining as it chases its $2 billion target. The other is the distillation dispute, where Anthropic’s specific claims about routed requests and collected responses have yet to be independently verified or addressed by Moonshot.