Moonshot AI is targeting $2 billion in annualized revenue by the end of 2026, representing a significant acceleration from $1 billion in annual recurring revenue recorded in August and roughly $100 million a year earlier.
The Beijing-based company shared the figures with investors, noting that annual recurring revenue had grown from approximately $300 million in June to above $1 billion by August. The company expects revenue to double again before December as consumer subscriptions to Kimi and enterprise adoption of its hosted AI models expand.
Kimi K3 Drives Revenue Growth
Most of Moonshot's recent growth followed the July launch of Kimi K3, a model with 2.8 trillion parameters. The model performed near the top of benchmark tests while requiring significantly lower computational costs to operate compared to major U.S. models. The launch expanded Moonshot's reach beyond China, with demand accelerating internationally as well.
Revenue streams include paid Kimi subscription plans for individual users and commercial licensing of its AI models to enterprises.
Hong Kong IPO Plans and Valuation
Moonshot is raising additional capital ahead of a planned Hong Kong IPO this year at a valuation of approximately $50 billion, positioning it near the current market value of comparable AI companies.
Anthropic Alleges Unauthorized Model Use
Anthropic reported discovering unauthorized use of its Claude model by several Chinese AI companies, including Moonshot, DeepSeek, and Alibaba. Anthropic characterized the practice as "illicit distillation," in which one AI system extracts outputs from a stronger model to train its own systems.
Anthropic stated that Moonshot transmitted prompts from Kimi users to Claude, returned Claude's responses to customers, and retained exchanges for training purposes. During one 10-day period, approximately 300,000 customer prompts were sent to Claude through accounts Anthropic identified as fraudulent, originating primarily from Singapore and Japan. Over the May to July period, more than 23 million exchanges were retained by Moonshot for training.
Anthropic noted that some transmitted exchanges contained sensitive information from individual users, multinational corporations, and state-affiliated actors, and stated uncertainty about whether Moonshot disclosed the data sharing to its customers. Anthropic indicated these practices may conflict with privacy regulations and terms of service.
Anthropic documented similar activity from DeepSeek, recording more than 12 million suspected distillation attacks over a 14-day period in July. The report covered disrupted activity from December 2025 through August 2026 involving multiple Chinese AI developers.
