Moonshot AI’s Kimi K3 Cloud Talks Face a 30% Revenue Split—and Major U.S. Security Questions

Moonshot AI’s reported negotiations with Microsoft, Amazon Web Services and Google Cloud would test whether a frontier Chinese model can be distributed through leading U.S. cloud platforms. The talks are still preliminary: Moonshot is reportedly seeking as much as 30% of revenue from Kimi K3-related services, but no agreement has been confirmed. 1

The potential deal matters because K3 combines open weights and very large scale with reported performance near leading Western models. It also sits at the intersection of cloud economics, enterprise AI adoption and U.S.–China technology restrictions.

What Moonshot is reportedly asking for

Three people familiar with the discussions told Reuters that Moonshot is negotiating with Azure, AWS and Google Cloud over arrangements that would allow the platforms to host Kimi K3. The company is reportedly seeking up to 30% of revenue generated by K3-related services. If finalized, the arrangement could become the first major revenue-sharing deal between a Chinese AI company and a large U.S. cloud provider. 1

The reported 30% figure is a negotiating request, not an agreed term. The parties still need to determine which revenue would count as K3 revenue and which would belong to cloud computing, storage, orchestration, support or other services. They also reportedly have unresolved questions about data access and how token consumption would be measured and audited. 1

Those details are central to the economics of the deal. Token usage is the basis for much AI-service billing, so the measurement system would directly affect revenue allocation. Data-access rules would determine how customer information is isolated, governed and protected, particularly for enterprise and regulated users.

What remains uncertain

No agreement has been announced. Reuters reported that the discussions are at an early stage and may not result in deals; Moonshot did not respond to its request for comment, while Microsoft, Google and AWS declined to comment. 1

That leaves several important questions unanswered, including:

• Whether any provider would offer K3 globally or restrict it by geography or customer type;
• What pricing and commercial model customers would receive;
• Where customer and service data would be stored and processed;
• What security, monitoring and model-isolation controls would apply;
• Whether U.S. export-control or sanctions requirements would prevent deployment;
• How a provider could terminate or suspend the arrangement if policy conditions changed.

Moonshot has reportedly reached smaller cloud-platform arrangements, including a July revenue-sharing agreement disclosed by Chinese IT-services company Chinasoft International. However, that agreement did not disclose the revenue split and does not establish that a deal with a U.S. hyperscaler will be completed. 1

Why Kimi K3 needs cloud distribution

Moonshot unveiled Kimi K3 in July 2026 as a roughly 2.8-trillion-parameter open-weight model and described it as the largest system of its kind. The company said its performance approached Anthropic’s frontier Fable model.

Third-party assessments cited by Reuters placed K3 strongly in web-interface building and described performance comparable to leading OpenAI and Anthropic models on complex, multi-step tasks. Those results are benchmark evidence rather than proof that K3 matches every competing model across real-world workloads. 1

Open weights allow developers to download and adapt a model, but they do not remove the infrastructure burden. A model at K3’s scale requires substantial computing capacity, deployment expertise and operational tooling. For many customers, a major cloud provider could therefore be the practical route to managed access, enterprise integration and pay-per-use deployment. 1

That creates a natural commercial role for Azure, AWS and Google Cloud: they could provide the infrastructure and customer distribution, while Moonshot supplies the model and seeks a share of the resulting service revenue.

The national-security and compliance problem

The commercial talks come amid U.S.–China technology restrictions, including limits on exports of advanced AI chips to China. Hosting K3 through U.S. providers could prompt scrutiny of the model’s provenance, customer data, intended uses and whether cloud access might undermine the goals of existing controls. 1

U.S. officials have accused Moonshot of obtaining technology from Anthropic’s Fable model and of illegally acquiring Nvidia chips. Treasury Secretary Scott Bessent reportedly said Moonshot might be added to a trade blacklist. These remain allegations and policy threats in the evidence available here, not adjudicated findings. 1

Moonshot has denied that K3 was created through distillation, saying its performance came from original changes to the model’s underlying architecture. That denial does not independently settle the provenance allegations. 1

For the cloud providers, the upside would be more model choice and potentially lower-cost access for customers. The downside would include regulatory, sanctions, cybersecurity, reputational and government-contracting risks. Any serious agreement would likely require extensive due diligence, strict data-governance rules, usage records that can be audited and clear rights to suspend or exit the relationship.

What it could mean for the AI market

A completed U.S. hosting arrangement would signal that capable Chinese models can attract commercial demand outside China even as technology tensions remain high. It could increase competitive pressure on Western model providers, especially in coding, agentic workflows and high-volume inference where price and infrastructure efficiency matter. 1

It would also strengthen the role of hyperscalers as neutral—or at least broad—AI distribution platforms. Rather than promoting only their own models, Azure, AWS and Google Cloud could increasingly compete by offering customers a portfolio of models from different developers.

For Moonshot, access to U.S. cloud infrastructure could provide recurring foreign revenue, broader enterprise reach and additional credibility as it reportedly considers a potential Hong Kong listing. But the same opportunity could disappear if the company faces a blacklist designation, tighter export controls or findings that support the allegations over model technology or chip access. 1

Bottom line

The reported talks are significant, but they are not yet a deal. Moonshot is testing whether Kimi K3’s scale and reported capabilities are valuable enough for U.S. cloud providers to accept a substantial revenue share and navigate the associated compliance burden.

The decisive issues will not be the headline 30% request alone. They will be whether the parties can independently measure usage, protect customer data, establish acceptable model provenance and satisfy U.S. security requirements. Until those questions are answered, the negotiations are best understood as a high-stakes commercial possibility—not evidence that Kimi K3 has secured a route onto the three major U.S. cloud platforms.