China's Moonshot AI is negotiating revenue-sharing agreements with Microsoft, Amazon, and Alphabet's Google that would let the U.S. cloud giants host its Kimi K3 model on Azure, AWS, and Google Cloud, according to three sources familiar with the talks reported by Reuters.

Deal Structure and Revenue Split

The Beijing-based startup is seeking up to 30% of revenue generated from K3-related services—a cut that aligns with terms Moonshot has outlined for major customers using its open-weight model. While the discussions remain in early stages with no guarantee of a deal, any agreement would mark the first significant revenue-sharing pact between a Chinese AI firm and American hyperscalers. The 30% revenue share Moonshot is pushing for represents standard territory for enterprise AI partnerships, but negotiating that split across three competing cloud providers introduces layers of complexity. Sources indicate unresolved issues include how revenue might be divided, data access protocols, and auditing mechanisms for token usage—tokens being the text units processed by AI models that form the basis of usage-based billing calculations.

Kimi K3's Competitive Standing

Moonshot isn't coming to the table empty-handed. Kimi K3 has posted impressive third-party evaluation results that make it attractive to U.S. partners hungry for cost-competitive alternatives. Arena.ai ranked K3 first in benchmarks assessing web interface-building capabilities, while Artificial Analysis notes performance comparable to OpenAI's GPT-5.5 and Anthropic's Claude Opus 4.8—especially on tests measuring complex, multi-step reasoning tasks. For developers building applications that depend on these capabilities, a cloud-hosted option through familiar platforms could lower barriers significantly.

Political Headwinds and Trade Concerns

The negotiations are happening despite escalating friction between Beijing-based AI companies and Washington officials. Treasury Secretary Scott Bessent suggested last month that Moonshot might face addition to a trade blacklist—a threat that could complicate any U.S. hosting arrangement. Beyond revenue terms, the geopolitical backdrop adds contractual uncertainty that infrastructure buyers typically avoid. Any deal would need robust exit clauses and compliance checkpoints built into the architecture from day one.

The Accusations Looming Over Moonshot

U.S. officials have accused Moonshot of training Kimi K3 through distillation of Anthropic's Fable model—essentially using outputs from a more capable system to shortcut development—and of illegally acquiring Nvidia chips restricted from export to China. Moonshot has rejected these claims, telling China's National Business Daily that its performance gains came from original changes to underlying architecture rather than distillation techniques. For cloud providers considering hosting agreements, the intellectual property allegations introduce reputational and legal risk that standard indemnification clauses may not fully address.

Technical Considerations for Cloud Teams

For DevOps and platform engineering teams evaluating whether to deploy Kimi K3 through their cloud provider's marketplace, several practical questions emerge. How will token metering work across multi-tenant infrastructure? What data residency requirements apply when inference happens on U.S. servers but the model originated from a Chinese company under regulatory scrutiny? These aren't trivial implementation details—they're the kind of operational friction that sinks partnerships before they scale.

Key Takeaways

  • Moonshot wants 30% revenue share for hosting Kimi K3 on Azure, AWS, and Google Cloud
  • Discussions are early-stage with no guarantee of agreements being reached
  • Kimi K3 benchmarks competitive with GPT-5.5 and Claude Opus 4.8 on complex tasks
  • Treasury Secretary Bessent has suggested Moonshot could face trade blacklist designation
  • Unresolved technical issues include token auditing, data access, and compliance checkpoints

The Bottom Line

The Kimi K3 hosting talks show U.S. hyperscalers are willing to overlook serious geopolitical baggage if the model delivers on its benchmarks—but that calculus could shift fast if Moonshot lands on a trade blacklist or distillation allegations gain traction in court.