Moonshot AI Targets $2 Billion Annualized Revenue Amid Anthropic Claims
Moonshot AI targets $2 billion in annualized revenue by year-end as K3 usage surges, while Anthropic alleges mass extraction of Claude responses for training.
Summary
Moonshot AI on September 11, 2026, set a target of $2 billion in annualized revenue by year-end, twice its reported August run rate. The Chinese AI lab’s aggressive goal follows K3’s summer 2026 release. Although usage has declined slightly in recent months, OpenRouter records as many as 300 billion K3 tokens generated daily.
Moonshot remains far behind OpenAI and Anthropic, whose projected revenues are $40 billion and $65 billion, respectively. K3’s freely available weights give Moonshot far lower margins than closed-weight competitors, though its rising sales target indicates open-weight models can still make money. Anthropic this week accused Moonshot of a long-running distillation campaign that routed nearly 300,000 Kimi requests directly to Claude Opus, effectively substituting Opus for Moonshot’s models. Anthropic alleges Moonshot collected more than 23 million responses from its models for training, raising questions about potentially illegal development practices.
Positives
- $2 billion in annualized revenue by year-end would double Moonshot AI’s reported August run rate.
- Up to 300 billion K3 tokens are generated daily on OpenRouter despite the recent usage decline.
- K3’s summer 2026 release has powered Moonshot AI’s more aggressive sales expectations.
- Freely available K3 weights indicate open-weight AI models can support substantial revenue, although margins trail closed-weight systems.
Risks & concerns
- K3 usage has declined slightly in recent months, even as daily OpenRouter volume remains high.
- Moonshot’s $2 billion target remains dwarfed by OpenAI’s $40 billion and Anthropic’s $65 billion projected revenue.
- Freely available weights leave Moonshot AI with far lower margins than closed-weight competitors.
- Anthropic alleges nearly 300,000 Kimi requests were routed directly to Claude Opus, effectively substituting Opus for Moonshot’s models.
- More than 23 million Anthropic responses were allegedly collected for Moonshot training, raising questions about potentially illegal practices.