Moonshot AI is negotiating with Microsoft, Amazon, and Google to host its open model Kimi K3 on their clouds, while it has suspended new subscriptions due to GPU shortages. The talks remain unresolved over revenue sharing, data access, and token auditing.
Kimi K3 is Moonshot AI's open model, released amid a capacity crunch. The company has suspended new subscriptions because of GPU shortages, even as it pushes the model onto major US cloud platforms. Moonshot AI is in talks with Microsoft, Amazon, and Google to host Kimi K3, seeking up to 30% revenue share from those providers. The negotiations have not closed, with unresolved issues over revenue sharing, data access, and token auditing.
Separately, Bloomberg reported that Moonshot AI signed a contract with Alibaba to use roughly 20,000 NVIDIA chips for its Kimi series models. Those chips are reportedly older Hopper architecture, possibly H200, though Alibaba denied supplying H200s. The report follows Kimi K3's release, which scored higher than GPT-5.6 Sol and the model 5 in some benchmarks. The rental arrangement illustrates how Chinese AI firms gain access to restricted NVIDIA hardware, complicating US export controls.
Synthesized by Yomimono from the cited Yomimono stories below, each itself sourced, then editorially reviewed. Every fact links the story it came from.
Structured graph also available as JSON at /public/entities/kimi-k3.
CC BY 4.0.
Chinese AI firm Moonshot AI is negotiating with Microsoft, Amazon, and Google to host its open model Kimi K3 on their cloud services, seeking up to 30% revenue share. The talks face unresolved issues over revenue sharing, data access, and token auditing. Moonshot AI has suspended new subscriptions due to GPU shortages.
Bloomberg reported that Moonshot AI signed a contract with Alibaba to use approximately 20,000 NVIDIA chips for its Kimi series AI models. The chips are reportedly older Hopper architecture, possibly H200, though Alibaba denied supplying H200s. The news follows Kimi K3's release, which scored higher than GPT-5.6 Sol and Claude Fable 5 in some benchmarks.