Moonshot Went From $300M to $1B ARR in Two Months
Moonshot AI told investors its annual recurring revenue passed 1 billion dollars in August, up from 300 million in June, and that it is targeting 2 billion annualized by year end. Bloomberg and TechCrunch both carried it on September 11. A run rate more than tripling in two months is the kind of number you normally see from a company that just found product-market fit in a new market, not from a two-year-old lab in Beijing.
The cause is not mysterious. Kimi K3 shipped in July, 2.8 trillion parameters, one million token context, open weights, and it went to the top of the benchmarks while costing a fraction of the US frontier models to run. Then everybody started building on it. Cognition post-trained SWE-2 on K3 and sells it at 64 percent below Fable 5.1. When your base model becomes the cheap floor other people's products stand on, inference revenue compounds in a way subscription revenue does not.
Scale check before anyone declares a winner. OpenAI is reported around 40 billion and Anthropic around 65 billion. Moonshot at 2 billion would be roughly 3 percent of Anthropic. But the derivative is the story, not the level, and the structure is different: Moonshot gives the weights away and still books the revenue, because most people would rather rent the serving than run a 2.8 trillion parameter model themselves. That is the business model the open-weight skeptics said could not exist.
The reporting also mentions a 50 billion dollar IPO ambition, which is unconfirmed and should be treated as such. What is confirmed and matters more is the second-order effect: every American company shipping on a Chinese open base is revenue for Moonshot and a data point for Garry Tan's argument that the US needs its own open-weight supply. TechCrunch's piece is at https://techcrunch.com/2026/09/11/kimi-maker-moonshot-ai-targets-2-billion-in-annual-revenue/
Related reading: https://clauday.com/article/6a998ccf-b793-4da4-9f00-97b27af81532 and https://clauday.com/article/dec052c0-f11f-424c-bc0a-b6fbe039a204
← Back to all articles
The cause is not mysterious. Kimi K3 shipped in July, 2.8 trillion parameters, one million token context, open weights, and it went to the top of the benchmarks while costing a fraction of the US frontier models to run. Then everybody started building on it. Cognition post-trained SWE-2 on K3 and sells it at 64 percent below Fable 5.1. When your base model becomes the cheap floor other people's products stand on, inference revenue compounds in a way subscription revenue does not.
Scale check before anyone declares a winner. OpenAI is reported around 40 billion and Anthropic around 65 billion. Moonshot at 2 billion would be roughly 3 percent of Anthropic. But the derivative is the story, not the level, and the structure is different: Moonshot gives the weights away and still books the revenue, because most people would rather rent the serving than run a 2.8 trillion parameter model themselves. That is the business model the open-weight skeptics said could not exist.
The reporting also mentions a 50 billion dollar IPO ambition, which is unconfirmed and should be treated as such. What is confirmed and matters more is the second-order effect: every American company shipping on a Chinese open base is revenue for Moonshot and a data point for Garry Tan's argument that the US needs its own open-weight supply. TechCrunch's piece is at https://techcrunch.com/2026/09/11/kimi-maker-moonshot-ai-targets-2-billion-in-annual-revenue/
Related reading: https://clauday.com/article/6a998ccf-b793-4da4-9f00-97b27af81532 and https://clauday.com/article/dec052c0-f11f-424c-bc0a-b6fbe039a204
Comments