One of the leading AI laboratories in China, Moonshot AI, is optimistic about converting its well-regarded open-weight model into significant revenue growth. On Friday, Bloomberg revealed that the lab aims for an annual revenue of $2 billion by year’s end, which is double the reported revenue rate for August. This ambitious target highlights the success of the company’s K3 model following its launch this summer.

Although the usage statistics for K3 have experienced a slight decline in recent months, OpenRouter data reveals that up to 300 billion tokens are currently being generated daily by K3 models operating on the system.

Moonshot’s anticipated revenue still pales in comparison to that of OpenAI and Anthropic, with recent reports estimating their figures at $40 billion and $65 billion, respectively. Given that Moonshot’s model weights are publicly accessible, the company enjoys significantly lower margins compared to its closed-weight rivals. The increasing forecasts indicate that there remains a financial opportunity in open-weight AI models, despite them being less profitable than their closed-weight counterparts.

Nevertheless, the model development practices of Moonshot continue to stir controversy — if not raise legal concerns. Earlier this week, Anthropic accused the firm of an ongoing model distillation scheme that redirected nearly 300,000 requests from Kimi directly to Claude Opus, effectively utilizing Opus instead of Kimi’s own models. In total, the company claimed that over 23 million responses were gathered from Anthropic models for incorporation into Moonshot’s training.