Kimi K3 Cloud Talks with Three US Hyperscalers
Moonshot AI is in negotiations with Microsoft, Amazon, and Google to deploy its flagship Kimi K3 model on Azure, AWS, and Google Cloud, seeking up to 30 percent of the revenue those platforms generate from K3-related services. The talks remain in early stages, and all three cloud providers and Moonshot AI have declined to comment on the negotiations' progress, according to reporting from August 26 cited in subsequent coverage. A deal would mark one of the first revenue-sharing agreements between a Chinese AI company and major US cloud providers.
The two sides have not yet reached agreement on key issues including how revenue will be calculated, what usage data Moonshot AI can access, and how tokens sold by the cloud providers will be audited. Kimi K3 is a 2.8-trillion-parameter open-weight model that is expensive to run, with Moonshot recommending clusters of 64 or more accelerators, which pushes large-scale consumption toward hosted cloud platforms.
Licensing Structure and Earlier Revenue-Sharing Deal
Kimi K3 operates under Moonshot AI's custom Kimi K3 License. The first half of the license follows a permissive approach similar to MIT, allowing use, modification, deployment, and commercialization, while adding a commercial threshold: any company and its affiliates running a model-as-a-service business generating more than $20 million in trailing 12-month revenue must reach a separate agreement before commercial use.
Moonshot AI has previously signed a similar revenue-sharing agreement with smaller platforms. Chinese IT services provider Chinasoft International announced on July 20 via a Hong Kong Stock Exchange filing that it had signed a token revenue-sharing and joint innovation cooperation agreement with Moonshot AI, under which the two parties will split token revenue generated from Kimi large models and other collaborative derivative income at an agreed ratio. K3 is already available on more than ten inference platforms, including Together AI and DigitalOcean.
Capability Benchmarks and Industry Context
Per Artificial Analysis cited in reporting, Kimi K3's capability is now comparable to OpenAI's GPT-5.5 and Anthropic's Claude Opus 4.8, at a reported cost around a third of Anthropic's Fable. The broader market is moving in a similar direction: Alibaba and MiniMax are exploring comparable revenue-sharing mechanisms, while DeepSeek and Tencent continue to favor permissive licensing. If the negotiations succeed, the talks would mark a shift in Chinese open-source model expansion from "maximizing deployment" to "monetizing at scale."
Chinese open-weight models first crossed into US enterprise channels through partners such as IBM, whose watsonx.ai directories list DeepSeek-derived models and whose recent $240 million agreement with Together AI includes DeepSeek, Kimi, and MiniMax in its model menu. Kimi's next phase is less about entry and more about how revenue is split once inside.
Policy Backdrop and Strategic Stakes
The commercial push runs alongside policy scrutiny. On July 22, US Treasury Secretary Scott Bessent said the department was considering adding Moonshot AI to a trade blacklist, while technology official Michael Kratsios accused the company of distilling Anthropic's Fable to build Kimi K3 and obtaining Nvidia GB300 servers. Moonshot denied the claims, saying K3's gains came from original architectural work. A cloud agreement would give Moonshot AI a revenue stream that scales with overseas usage, a factor that matters as it prepares for an IPO at a valuation reportedly up to $50 billion.
Open Negotiating Points
Three substantive questions remain unresolved in the talks: the revenue calculation methodology, the scope of usage data Moonshot AI can access, and the auditing mechanism for tokens sold by the cloud providers. Until these points are settled, the structure of any eventual agreement — including whether the 30 percent figure becomes a uniform rate or varies by platform — is not publicly defined.
Share this article
