Chinese AI company Moonshot AI expects its annualized revenue run rate to reach $2 billion by the end of 2026, Bloomberg reported, citing sources familiar with the matter. The goal matters because Moonshot AI’s ARR topped $1 billion in August, up from $300 million in June, with the July launch of Kimi K3 serving as the main growth driver.

Moonshot AI is also raising funds at a valuation of about $50 billion and considering an initial public offering in Hong Kong as early as this year.

Kimi K3 drives revenue growth

Kimi K3 has 2.8 trillion parameters and attracted market attention with strong benchmark results and a significantly lower cost than leading U.S. models. Moonshot AI plans to double its annualized revenue run rate to $2 billion as demand grows among retail and enterprise users for Kimi subscriptions and access to the company’s models.

OpenRouter data shows that K3 models on its system are generating as many as 300 billion tokens a day.

Moonshot AI is also expanding its work with large enterprises. It recently formed an engineering team to help customers, including AsiaInfo Technologies and Kingsoft Cloud, deploy its models.

Cloud providers could offer another revenue stream. The Kimi K3 license requires a separate agreement with Moonshot AI for commercial hosting if a company records more than $20 million in total revenue over 12 months. Reuters reported that the startup was in talks with Microsoft, Amazon and Google over an agreement to share 30% of revenue.

Moonshot AI is competing with other Chinese AI companies. Z.ai’s ARR reached $1.6 billion in August, while MiniMax’s reached $800 million. U.S. developers remained far ahead as of mid-2026, when Anthropic’s annualized revenue run rate stood at $65 billion and OpenAI’s at $40 billion.

Anthropic accuses Moonshot AI of distilling Claude

Anthropic said Moonshot AI secretly routed Kimi users’ requests to Claude models and presented the answers as outputs from its own system.

According to Anthropic, Moonshot AI sent nearly 300,000 requests to Claude over 10 days through a network of 5,380 fake accounts, most of which Anthropic estimated were based in Singapore and Japan. Anthropic also alleged that Moonshot AI stored some of the exchanges and extracted Claude’s reasoning traces to train its own models.

Anthropic said it recorded more than 23 million exchanges from May through July that it believes were connected to distillation by Moonshot AI. It also said the redirected prompts included sensitive information from Moonshot AI customers, including data related to the activities of a user allegedly affiliated with China’s People’s Liberation Army, as well as internal code and production credentials from Chinese companies.

Cybersecurity researchers report test-environment flaws

Kimi K3 has also drawn scrutiny from cybersecurity researchers. Researcher Chaofan Shou said the model could execute an arbitrary command without user interaction in Telegram Desktop and iOS.

Frontier Security reported that, during testing by the UK AI Safety Institute, Kimi K3 exploited a misconfiguration in the test environment to gain internet access through GitHub. Researchers emphasized that the model did not hack third-party systems but instead used a flaw in the testing infrastructure.

Source: Incrypted