Industry·3 min read·Reuters

Moonshot Wants 30% of Cloud Revenue From Kimi K3

Reuters reports early-stage talks with Microsoft, Amazon and Google that would make Kimi K3 a first-party offering on Azure, AWS and Google Cloud. It would be the first major revenue-sharing pact between a Chinese AI lab and a US hyperscaler — and the weights are already public, which is the awkward part.

MOONSHOT WANTS 30% of cloud revenue from Kimi K3 BITSMINDS.COM
Share:

Moonshot AI is in early talks with Microsoft, Amazon and Google about deals that would put its Kimi K3 model on their clouds, Reuters reported on August 26, citing people familiar with the discussions. The Chinese startup is asking for as much as 30% of the revenue the three earn from K3-related services on Azure, AWS and Google Cloud. Any agreement would be the first significant revenue-sharing arrangement between a Chinese AI developer and a major American cloud provider.

None of it is settled. The sources describe the talks as early-stage and possibly fruitless, with the revenue split itself unresolved along with two thornier mechanics: what data access the arrangement grants, and how token consumption would be audited. That second item is the crux of the commercial design. A revenue share on a hosted model is only enforceable if both sides trust the meter, and the meter sits inside the cloud provider's infrastructure. All three companies declined to comment; Moonshot did not respond to Reuters.

There is an obvious puzzle in the arrangement, and the report does not resolve it. Kimi K3 is open-weight — Moonshot published the 2.8-trillion-parameter weights publicly in July, and any of the three clouds could serve them today without asking permission or paying anyone. What a formal deal would buy is everything the weights alone do not carry: official support, a version pipeline, tuning help, indemnity, and the difference between an unsupported community upload and a first-party managed endpoint an enterprise buyer will actually sign for. Whether that bundle is worth 30% of the revenue is exactly the question the negotiation appears to be stuck on.

The model gives Moonshot something to bargain with. K3 runs to 2.8 trillion parameters and performs comparably to OpenAI's GPT-5.5 and Anthropic's Claude Opus 4.8, and it beat Fable 5 on frontend coding when it launched. Commercially the company is on a steep curve: annual recurring revenue reached about $300 million in June, up from $200 million in April, and it is raising at a reported $50 billion valuation ahead of a possible Hong Kong listing, having targeted $30 billion as recently as June. Founded in 2023 and backed by Alibaba among others, it took in more than $2 billion in May.

The politics are the part no spreadsheet resolves. Treasury Secretary Scott Bessent has suggested adding Moonshot to US trade blacklists, and officials have alleged the company trained K3 using banned Nvidia chips and distilled the model from Anthropic's technology — the same accusation Anthropic levelled at Alibaba in June, and a category of transfer that no export control currently covers. Moonshot says its gains come from original changes to the underlying architecture.

For the hyperscalers the calculation is uncomfortable in both directions. Customers are asking for the model, and a cloud that cannot serve the strongest open-weight system on the market is selling an incomplete menu. But a formal revenue-sharing contract with a company a sitting Treasury Secretary has floated for a blacklist is a different exposure from quietly hosting weights someone else published. That the talks are happening at all says more about demand for K3 than about anyone's appetite for the risk.

Want AI news before everyone else?

The morning's most important AI stories, straight to your inbox. No fluff.

Related Articles