Illustration: AI-generated · Inside China AI
Microsoft, Amazon and Google Are Talking to Moonshot AI — and the Ask Is 30 Per Cent
If it closes, it will be a first: a Chinese AI lab earning a cut of revenue from the American cloud giants. According to a Reuters exclusive, Moonshot AI is in talks with Microsoft, Amazon and Google about hosting its Kimi K3 model — and is seeking up to a 30 per cent share of the revenue those services generate on Azure, AWS and Google Cloud. The talks are early, the sources are anonymous, and nothing is signed. But the mere fact of them tells you how far China’s models have travelled: from cheap alternative to something the hyperscalers want on their shelves — while Washington debates blacklisting the company.
What is actually on the table
Reuters reports the negotiations centre on revenue sharing for K3-related services, with unresolved questions around how revenue is split, data access, and the auditing of token usage — the practical plumbing of hosting someone else’s frontier model. For the clouds, the appeal is straightforward: customers are asking for Chinese open-weight models because of the price-performance gap we documented in our analysis of China’s AI pricing power. For Moonshot, distribution through the world’s three largest clouds would be a bigger prize than any marketing budget could buy.
Who Moonshot is
Founded in 2023 by Yang Zhilin, a Carnegie-Mellon-trained researcher, Moonshot AI became known for long-context models and the strong open-weight Kimi line. It is backed by Chinese tech heavyweights including Alibaba, raised more than $2 billion in May, and is preparing a potential Hong Kong listing — following Zhipu and MiniMax onto the public markets. A revenue stream from American clouds would be a compelling line in any IPO prospectus.
The political fault line
The timing is awkward, and that is the point. US Treasury Secretary Scott Bessent said last month he might add Moonshot to a trade blacklist — while three American technology companies are reportedly negotiating to distribute its model. That contradiction is the clearest illustration yet of the gap between Washington’s China policy and the commercial pull of Chinese AI: export controls restrict chips going in, but nothing stops capable, cheap models coming out. It is the same asymmetry we described in AI Without Nvidia, seen from the software side.
Why it matters for Europe
This is the development European IT departments should watch most closely — because it would change how Chinese AI reaches them. Today, using Kimi or Qwen means a deliberate decision: download the weights, or contract with a Chinese provider. If Azure, AWS and Google Cloud host K3, Chinese frontier models arrive through the procurement contracts European companies already have, with familiar billing, familiar data-processing terms, and no new vendor relationship. Adoption stops being a strategic choice and becomes a dropdown menu.
That raises questions worth answering before the option appears rather than after: which model families does your AI governance actually permit, and on what grounds? How do the EU AI Act’s transparency duties apply when the model is Chinese, the host is American, and the deployer is European? And, more bluntly: if the cheapest capable model on your existing cloud contract is Chinese, what is your policy — and is it written down? These talks may collapse. The question they raise will not.
No responses yet