China's Moonshot AI is targeting $2B in annual revenue. Its model is free to download.

Moonshot AI, the Chinese lab behind the popular Kimi assistant, told investors it is targeting $2 billion in annualized revenue by the end of 2026, according to a Friday Bloomberg report. The target is striking for a simple reason: the model powering Kimi is freely downloadable. Anyone can run it, undercut the price, or wrap their own product around it.

The reported target roughly doubles Moonshot's August run rate. That pace requires either a fast jump in usage or a fast jump in average revenue per token, and it forces a question open-weight labs have to answer in public: where does the money come from when the weights are not a moat?

Open-weight economics work when the lab makes money on volume, on hosted API access, on enterprise contracts, and on the developer ecosystem that forms around a popular model. The model itself is a marketing surface, not the product. The product is the billable access layered on top.

K3, the model behind Kimi, is the volume engine. OpenRouter, one of several public API aggregators, shows K3-family models serving as many as 300 billion tokens per day on the platform, with K3 itself pushing into the top tier of OpenRouter's rankings over the summer. That figure has softened slightly in recent months even as the revenue target has risen, which means the company is leaning harder on the part of the business that actually bills per token: API contracts, enterprise deployments, and developer subscriptions.

OpenAI has been reported at roughly $40 billion in annual revenue, and Anthropic at roughly $65 billion, both third-party figures rather than audited numbers. Moonshot's $2 billion target is a real business but a fraction of the frontier labs' books. The open-weight strategy is not trying to match them on topline. It is trying to be profitable on a smaller top line by avoiding the cost of training a new frontier model every six months.

Then there is the allegation Anthropic has put into the public record. Anthropic says Moonshot ran a long-running model distillation campaign that routed nearly 300,000 Kimi requests directly to Claude Opus and collected more than 23 million responses from Anthropic models for use in training. Distillation is the practice of training a smaller or cheaper model against the outputs of a larger one. It is not, in itself, a violation of any rule Anthropic has published, and Moonshot has not been adjudicated as having broken any. It is, however, conduct a closed-weight lab treats as a competitive threat to the moat its investors are paying for.

Anthropic's choice to put the allegation in writing, with specific request and response counts, is a closed-weight lab telling the market what rival it is worried about. The 300,000 requests and 23 million responses are small relative to the global training corpora behind modern models, but the pattern is what a rival leaves when it leans on a closed-weight lab's outputs rather than training its own frontier model.

Moonshot filed for a $3 billion Hong Kong IPO at a $50 billion valuation earlier this month, a print that tests whether public-market investors will price an open-weight lab like a consumer-app company or like a frontier-model competitor. The $2 billion revenue target is the number the IPO will be measured against. The Anthropic allegation is the number the IPO will be argued around.

The next data points to watch are the Hong Kong IPO pricing, Moonshot's first reported quarter of revenue, and whatever Anthropic puts on the public record next. The open-weight business model becomes testable on those three prints.