Moonshot AI wants $2 billion in revenue by year-end, even as it faces theft allegations
The Chinese lab behind the Kimi chatbot is chasing a revenue target double its August run rate. The ambition arrives the same week Anthropic accused it of secretly feeding millions of user queries through Claude.
Key points
- Moonshot AI, the Chinese lab that makes the Kimi chatbot, is targeting $2 billion in annualised revenue by the end of 2026, double its reported run rate for August 2026.
- Its K3 model, an open-weight model (meaning its underlying code is freely available for anyone to download and use), generates up to 300 billion tokens per day on the OpenRouter platform.
- Anthropic alleged this week that Moonshot routed nearly 300,000 Kimi user requests directly to Claude Opus, secretly using Anthropic's answers to train its own models.
- Moonshot's $2 billion target is still far below OpenAI's reported $40 billion and Anthropic's reported $65 billion in annualised revenue.
Moonshot AI is chasing $2 billion in annualised revenue by the end of 2026, double what it was bringing in as recently as August, Bloomberg reported on Friday. The engine behind that ambition is K3, an open-weight model: anyone can download and run its code without paying Moonshot a licence fee.
Open-weight models are a bet that popularity converts to revenue through cloud services and API access (a way for other software to connect to and use the model). It's a harder road than selling a proprietary model. Moonshot's margins are far thinner than those of OpenAI or Anthropic, where every query is a paid transaction.
How does K3 actually stack up?
Usage has slipped slightly since K3 launched this summer, but the numbers remain large. OpenRouter, a platform that lets developers compare and switch between AI models, currently shows K3 generating up to 300 billion tokens a day. A token is roughly three-quarters of a word.
For scale, Moonshot's target sits well below OpenAI's $40 billion and Anthropic's $65 billion in annualised revenue. We covered the open-weight economics driving deals worth over $26 billion in our 28 August report: popularity is real; margins aren't.
What are the allegations against Moonshot?
The revenue story has an uncomfortable backdrop. Anthropic this week accused Moonshot of a prolonged scheme to steal its model's behaviour. Moonshot, Anthropic alleged, routed nearly 300,000 Kimi user requests directly to Claude Opus, Anthropic's most capable model at the time, so that Opus's answers flowed back to Kimi users unmodified. Across the full campaign, more than 23 million Claude responses were collected and fed into Moonshot's training data.
This is model distillation: using a powerful model's outputs to teach a cheaper one to behave similarly. It's banned by most AI companies in their terms of service. Our 9 September story reported that US agencies had named six Chinese AI firms accused of extracting capabilities from American frontier models at industrial scale; Anthropic's complaint arrives days later and adds a named company, a named victim model and a specific count.
Moonshot hasn't publicly responded to the allegation.
The revenue ambition is real and the growth trajectory is genuinely fast. But whether K3's progress was built cleanly is now an open question, and it's one that any developer or business that has built on top of it will need to watch closely.
