ChinaModelAPI

News / Model Watch · Business / Licensing

Reported (Reuters) Kimi K3 30% revenue share Azure / AWS / GCP
2026-09-01 · Model Watch · Reuters reported Aug 26, early-stage negotiations

Moonshot's 30% Ask: Kimi K3 Open-Source Model Seeks Hyperscaler Revenue Share

Chinese media calls it 开源模型「海外收租」 — overseas rent collection on open-source models. Per Reuters (Aug 26, three sources), Moonshot AI is negotiating with Microsoft Azure, Amazon AWS, and Google Cloud to host Kimi K3, seeking up to 30% of the revenue from K3-related services. Nothing is signed; nothing is confirmed. But if it lands, it would be the first major revenue-sharing deal between a Chinese AI lab and a US hyperscaler — and a new playbook for how open-weight AI gets monetized.

Direct answer

Moonshot AI is reportedly seeking up to 30% revenue share from Azure, AWS, and Google Cloud for hosting Kimi K3, per Reuters (Aug 26, three sources). Negotiations are early-stage with unresolved issues on revenue distribution, data access, and token-usage audits. No party has commented. The leverage: the Kimi K3 License requires companies with $20M+ annual revenue to sign a separate agreement; the 2.8T-parameter model is too large for most companies to self-host. Alibaba is reportedly pursuing similar terms for Qwen.

What's actually on the table

  • The ask: Up to 30% of revenue from K3-related services on the three hyperscalers — reportedly similar to terms Moonshot has discussed with other large customers. One smaller deal (with China's Chinasoft International) has reportedly been reached.
  • The leverage: The Kimi K3 License (not Apache/MIT) requires a $20M+ revenue threshold company to obtain a separate commercial agreement. The model is 2.8T parameters / ~1.56TB — only hyperscalers can serve it at scale.
  • The friction: Per one source — revenue distribution, data access, and token-usage audits remain unresolved. All parties (Moonshot, Microsoft, Amazon, Google) declined or did not respond to comment requests.
  • The context: Alibaba is separately seeking revenue-sharing for its own open-source model (Qwen3.8-Max uses a custom license). Harvey (legal-tech) already built Harvey Tenet on the K3 base — the ecosystem is real.
  • The risk (per analyst commentary): US export/trade restrictions or blacklisting could block the entire arrangement; political friction around Chinese AI in US enterprise channels.

Why this matters for the open-weight model

If a deal closes, it would reverse the traditional direction of software licensing: instead of US companies charging Chinese users for software, a Chinese lab would be collecting rent from US cloud giants for the privilege of serving its model. Chinese media's 「海外收租」 framing captures it — and with Alibaba reportedly pursuing the same for Qwen, this could become the standard monetization template for the entire Chinese open-weight frontier.

Primary sources

FAQ (2026)

What's the ask?

Up to 30% of K3-related service revenue on Azure/AWS/GCP, per Reuters (Aug 26, early-stage talks).

Why pay for 'open source'?

The K3 License has a $20M revenue threshold clause; the 2.8T model is too large for most companies to self-host anyway.

Alibaba doing the same?

Reportedly yes — separately seeking revenue-sharing for Qwen open-source models.

Sticking points?

Revenue distribution, data access, and token-usage audits. No public comments from any party.

Anything signed?

No hyperscaler deal. One smaller agreement with Chinasoft International reportedly reached.

Why it matters?

First Chinese-AI-lab-to-US-hyperscaler revenue share would reverse the software-licensing flow — a new open-weight monetization playbook.