China’s Moonshot AI is in early talks with Microsoft, Amazon, and Google about revenue-sharing agreements that would let the three companies offer its Kimi K3 model through Azure, Amazon Web Services, and Google Cloud, according to Reuters. Moonshot is reportedly seeking as much as 30% of the revenue generated by K3-related services.
No agreement is assured. The discussions remain preliminary, and unresolved questions include the final revenue split, access to data, and how token usage would be audited. Moonshot did not comment to Reuters, while Microsoft, Google, and AWS declined to comment.
Distribution is the real enterprise moat
Kimi K3 is open-weight, so organizations can download and modify it. In theory, that makes the model broadly available already. In practice, a 2.8-trillion-parameter system is too costly and operationally demanding for many companies to run on their own infrastructure.
Cloud availability changes that equation. A managed version can place the model inside environments where enterprise customers already control identities, billing, security policies, observability, and procurement. It also lets developers compare K3 with other hosted models without building a separate infrastructure stack.
That is why these talks matter beyond one model. Frontier performance attracts attention, but distribution turns capability into recurring use. Azure, AWS, and Google Cloud are not merely hosting venues; they are the primary storefronts through which many businesses encounter and adopt AI models.
A commercial bridge across a geopolitical divide
If completed, a deal could become the first major revenue-sharing agreement between a Chinese AI developer and a leading U.S. cloud provider. It would also expose a tension at the center of the AI market: commercial demand for capable, lower-cost models can cross national boundaries even as technology policy pushes in the opposite direction.
Moonshot has faced scrutiny from U.S. officials, while export restrictions continue to limit China’s access to advanced AI chips. Those pressures have not stopped K3 from attracting attention in independent evaluations or from becoming a potential addition to American cloud catalogs.
The cloud companies would gain another competitive model for customers. Moonshot would gain global distribution, enterprise credibility, and a monetization path that does not require it to supply all of the serving capacity itself. The proposed percentage suggests Moonshot sees K3 not as inexpensive catalog filler but as a product with meaningful bargaining power.
Demand has already tested Moonshot’s capacity
K3’s launch demonstrated both its appeal and its infrastructure problem. Moonshot temporarily paused new consumer subscriptions after demand pushed its computing capacity close to its limits. Hosting through the largest cloud platforms could help absorb that demand while shifting part of the operational burden to providers experienced at serving enterprise workloads at scale.
Cloud deployment would not remove every constraint. Providers would still need to settle data governance, usage measurement, model updates, regional availability, safety controls, and the effect of any future U.S. restrictions. Enterprise buyers would also need clear answers about where data is processed and how the model is operated.
What product teams should watch
For developers, K3’s possible arrival on major clouds is another reason to design AI products around model portability. Quality, latency, price, policy, and availability can all change quickly. A product that can evaluate and route across models is better positioned than one tied deeply to a single provider.
The negotiations may still produce nothing. But the fact that they are happening shows how the market is evolving: the competitive unit is no longer just the model. It is the combination of capability, price, distribution, governance, and integration. Kimi K3 has already proved it can enter the performance conversation. The cloud talks will test whether it can enter the enterprise stack.