Moonshot AI, one of China’s leading artificial intelligence laboratories and the developer of the Kimi model, is targeting an annualized revenue run rate of $2 billion by the end of 2026, according to Bloomberg, as reported by TechCrunch on September 11. This would double the annualized revenue run rate the company reported for August, in a bet that depends heavily on the commercial success of the K3 model it launched during the summer of this year.
High Usage Despite the Recent Decline
OpenRouter data indicates that K3 models currently generate as many as 300 billion tokens per day on the platform. At the same time, however, the report states that the model’s usage indicators have declined slightly in recent months, making the revenue target an ambitious challenge rather than a certain outcome based on its current usage trajectory.
Moonshot AI is betting on a model with freely available weights, giving developers access to and use of the model without the closed distribution model followed by other companies. According to the report, however, this approach is associated with lower profit margins than those generated by competing closed models.
A Commercial Opportunity with an Open-Weights Model
The importance of the target figure lies in the indication it provides that significant revenue can be generated from AI models with available weights, even if their returns are lower than those of companies that keep their weights closed. Nevertheless, Moonshot AI’s expectations remain far smaller than the revenue figures attributed in recent reports to OpenAI and Anthropic, amounting to $40 billion and $65 billion, respectively.
What matters in practice? The development shows that a model’s reach is not measured by free usage alone, but by the company’s ability to convert that usage into revenue. In Moonshot AI’s case, high daily generation volumes are not enough to prove that the target will be achieved; the actual sources of income, their margins, and the sustainability of demand after the slight decline in usage must be determined.
Accusations Concerning Model Training
The financial target comes amid controversy over Moonshot AI’s model-development practices. Earlier in the same week, Anthropic accused the company of running an ongoing campaign to extract knowledge from its models by directing nearly 300,000 requests from Kimi straight to Claude Opus, causing the latter model to provide the answers instead of Kimi models, according to Anthropic’s allegation.
Anthropic said that more than 23 million responses were collected from its models for use in training Moonshot AI’s models. The source does not provide a response from Moonshot AI to these accusations or a legal finding concerning them, so these points remain allegations attributed to Anthropic rather than established facts. The case could affect assessments of the sustainability of Moonshot’s commercial model as much as usage and revenue figures themselves.