Moonshot AI targets $2B annualized revenue by year-end, up from $1B in August

Moonshot AI wants to push its annualized revenue to $2 billion by the end of 2026, putting the Chinese startup on course for a huge increase from roughly $100 million a year earlier.

The Beijing-based company told investors that annual recurring revenue had already moved above $1 billion in August, compared with about $300 million in June.

The data were allegedly shared privately, and said that Moonshot expects another doubling before December as more individual customers pay for Kimi subscriptions and businesses use the AI models it hosts online.

Most of Moonshot’s latest growth came after Kimi K3 arrived in July. The model has 2.8 trillion parameters, basically the values it works with when reading information and coming up with answers.

K3 finished near the top on benchmark tests while costing far less to run than big US models. The launch also got Moonshot noticed outside China, with demand picking up fast there too. It is trying to win users in a packed AI market, where OpenAI charges more for its products and Chinese companies like DeepSeek compete with lower prices.

Moonshot is also raising fresh money at a valuation of about $50 billion before a possible Hong Kong IPO this year. That puts it close to the current market value of Z.ai (HKEX: 2513).

Kimi K3 brings in more money as Moonshot gets ready for a Hong Kong IPO

Moonshot wants its yearly sales rate to reach $2 billion before the end of the year. Revenue is coming from paid Kimi plans for regular users and online models sold to companies. It is also raising more cash before its planned stock market debut.

At the same time, Anthropic has accused Chinese AI companies of using Claude without permission while training their own systems. Anthropic said Thursday that it found and blocked activity tied to Moonshot, DeepSeek and Alibaba Group Holding Ltd. (NYSE: BABA; HKEX: 9988).

Anthropic called the practice “illicit distillation.” That means one AI system takes answers from a stronger model and uses them to help train another one. Anthropic said the companies did this without approval.

Anthropic also said people linked to Alibaba used Claude answers while building Qwen models. It tracked more than 151 million exchanges connected to Alibaba between May and July, the biggest operation of this kind it said it had seen.

At its busiest point, the operation generated almost 3 million exchanges each day through more than 3,500 accounts Anthropic classified as fraudulent. Alibaba-linked researchers also used Claude for other AI work, including reinforcement learning and research into model architecture, the report said.

Anthropic says Moonshot sent Kimi customer requests to Claude and reused some exchanges for training

Anthropic also gave separate numbers for Moonshot. It said the company secretly passed some prompts submitted by Kimi users to Claude before returning Claude’s answers to customers who believed they were interacting with Kimi.

In one 10-day span, close to 300,000 customer prompts were transmitted to Anthropic. Majority of these were submitted to Claude Opus models. The data went through 5,380 accounts, which Anthropic referred to as fraudulent, mostly originating from Singapore and Japan.

According to the report, Anthropic added that Moonshot retained some of these discussions and gathered Claude’s transcripts for the purpose of training. In total, between May and July, more than 23 million discussions were credited to Moonshot.

Some of the prompts contained sensitive data. Anthropic said that it could not ascertain if Moonshot informed their clients about their data sharing with an external AI provider.

“Some of these exchanges included sensitive information, including from individual users, major multinational companies, and state-affiliated actors … These practices are likely inconsistent with privacy laws and the labs’ own terms of service,” Anthropic said in the report.

DeepSeek was accused of using a similar setup. Anthropic said the company transferred user exchanges to Claude without telling those customers. Over 14 days in July 2026, Anthropic recorded more than 12 million suspected distillation attacks linked to DeepSeek.

Anthropic’s report covered activity it said it disrupted from December 2025 through August 2026 and named several other large Chinese AI developers.

The cases were divided across seven categories: cyber operations, influence campaigns, surveillance, scams and fraud, biological misuse, conventional weapons development and AI model distillation.

If you're reading this, you’re already ahead. Stay there with our newsletter.