Moonshot AI is targeting $2 billion in annualized revenue by the end of 2026, roughly double the run rate the Chinese lab reported for August. The goal, first reported Friday by Bloomberg, hinges on continued adoption of the company's open-weight K3 model, which launched over the summer and has become one of the most-used open models on third-party inference platforms. It is an aggressive number for a lab whose weights are freely downloadable and whose margins are structurally thinner than those of closed-weight rivals.
K3's usage on OpenRouter currently runs as high as 300 billion tokens generated per day, even after a slight decline in recent months. That volume puts Moonshot in the same conversational tier as the frontier US labs on distribution, if not on price capture. Token throughput on a routing marketplace is not the same as booked revenue, but it is the closest public proxy for how much real inference demand K3 is absorbing.
The $2 billion target still sits far below the reported run rates of the US frontier leaders. Recent figures put OpenAI at $40 billion annualized and Anthropic at $65 billion — 20 to 30 times what Moonshot is aiming for. The gap reflects both scale of paid deployment and the pricing power that closed-weight APIs command over open-weight alternatives that customers can self-host.
Key facts
- 01Moonshot AI is targeting $2 billion in annualized revenue by the end of 2026, double its reported August run rate.
- 02Kimi's K3 model generates as many as 300 billion tokens per day on OpenRouter.
- 03Recent reports put OpenAI at $40 billion and [Anthropic](/claude) at $65 billion in annualized revenue, dwarfing Moonshot's target.
- 04Anthropic alleges Moonshot routed nearly 300,000 Kimi requests to Claude Opus and collected 23 million Anthropic responses for training.
Open weights cut both ways for Moonshot. Free distribution has clearly seeded adoption fast, giving Kimi the kind of developer footprint that closed labs spend years and hundreds of millions in sales motion to build. But every enterprise that downloads K3 and runs it on its own GPUs is a customer Moonshot does not directly bill. Monetization has to come from hosted inference, enterprise contracts, and premium tiers rather than from the model itself.
The revenue projection lands in a week that has been considerably less flattering to Moonshot on the research-ethics side. Anthropic published findings this week alleging that Moonshot, Alibaba, and DeepSeek all ran distillation campaigns against Claude — the practice of using a competitor's model outputs to train your own. Moonshot has not publicly responded to the specific allegations.
The Anthropic filing is unusually specific about the Kimi case.
The allegation, if accurate, complicates the story Moonshot is telling investors. A $2 billion revenue lab whose product quality depends materially on Anthropic outputs is exposed to any technical or legal move Anthropic makes to shut that pipeline down. Anthropic has already said it is blocking the identified traffic patterns, and the company's terms of service explicitly prohibit using Claude outputs to train competing models.
The broader signal for the AI market is that open-weight economics are not as bleak as the frontier labs sometimes suggest. Moonshot going from launch to a $2 billion revenue target inside a year — even at open-weight margins — shows there is real willingness to pay for hosted inference of a strong open model, particularly outside the US where OpenAI and Anthropic distribution is thinner. Whether Moonshot can defend that revenue while resolving the training-data questions raised this week is the harder call, and one investors will price into any next round.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
