One of China’s leading artificial intelligence research labs, Moonshot AI, is aiming to achieve $2 billion in annualized revenue by the end of this year, according to Bloomberg. This would nearly double the company’s current revenue run rate, which was reported in August. The ambitious goal is largely driven by the success of Moonshot’s K3 AI model, which was launched this summer. Although the usage of K3 has seen a slight decline in recent months, data from OpenRouter indicates that K3 models are still generating up to 300 billion tokens per day — a measure of the amount of text processed by the AI. Despite this growth, Moonshot’s projected revenue remains significantly lower than that of its major competitors, such as OpenAI and Anthropic. Recent reports estimate OpenAI’s annual revenue at $40 billion and Anthropic’s at $65 billion. One key reason for this gap is that Moonshot makes its AI model weights freely available, allowing other developers to use and modify them. This openness results in lower profit margins compared to companies like OpenAI and Anthropic, which keep their model weights proprietary. The rising revenue projections suggest that there is still potential for profit in open-weight AI models, even if they are not as lucrative as closed-weight models. However, recent developments have cast a shadow over Moonshot’s progress. Earlier this week, Anthropic, another major AI company, accused Moonshot of running a long-term campaign to distill, or refine, its AI models using data from Anthropic’s own systems. According to Anthropic, Moonshot routed nearly 300,000 requests from its Kimi AI model to Anthropic’s Claude Opus model, effectively using Opus to generate responses instead of Kimi. Over time, more than 23 million responses were collected from Anthropic’s models, which the company claims were used to train Moonshot’s AI.