Amazon Web Services has hosted Moonshot AI’s Kimi K3 on Bedrock, creating a new commercial route for a Chinese open-weight model and testing whether cloud platforms can monetise openness without swallowing it whole. The practical shift is bigger than another model appearing in a catalogue: AWS and Moonshot are reported to be sharing revenue, while Microsoft Azure and Google Cloud are also discussing Kimi partnerships.
AWS AI Watch analysis
What happened
A report published by 36Kr says AWS announced on 18 September that Kimi K3 was available through Bedrock. It says the arrangement includes revenue sharing, with Reuters cited for a possible 30% share for Moonshot, although the financial terms have not been publicly disclosed by the companies.
Kimi K3 launched as an open-source model on 16 July, according to the report, which describes a 1-million-token context window, native visual capabilities and a 3-trillion-parameter scale. The same account says Moonshot is pursuing an ecosystem model for enterprise delivery: it supplies the model and agent capabilities, while cloud vendors, developers and implementation partners handle deployment and industry integration.
Why it matters
Open-weight models are often praised for being downloadable, adaptable and cheaper to run. The awkward commercial question is how their creators get paid when customers can modify the software and other companies provide the infrastructure. A revenue-sharing deal through a major cloud platform offers one answer, while keeping the model accessible to enterprise buyers.
It also gives AWS a way to broaden its model shop beyond its close relationships with other frontier laboratories. For customers, the attraction is choice and potentially lower-cost deployment. For Moonshot, the prize is distribution, billing and a path into overseas enterprise work without building every sales and engineering operation itself.
Our read
This is a meaningful business test, not proof that open-weight AI has solved its revenue problem. The reported 30% figure is not a disclosed contract term, and the claims about performance, downloads and enterprise adoption come from the 36Kr account rather than an independently supplied benchmark or filing. Still, putting Kimi K3 on Bedrock makes the argument tangible: “open” can describe the model’s availability while the surrounding commercial machinery remains decidedly pay-to-play.
What to watch
- Contract terms:
whether AWS or Moonshot publishes the revenue split, pricing and licensing conditions. - Enterprise uptake:
whether named customers begin using Kimi K3 through Bedrock. - Cloud competition:
whether Azure and Google Cloud announce comparable hosting deals. - Implementation model:
whether Moonshot’s partner-led approach produces deployments beyond the model catalogue.
Discussion spark: If open-weight models can earn revenue through cloud hosting and implementation partners, does that strengthen the case for openness, or merely recreate a toll system around supposedly open software?
Sources and evidence
- Domestic Open-Source AI Models: The Emerging New Business Opportunity for Silicon Valley Cloud Vendors – eu.36kr.com (24 September 2026, 03:30 UTC)
not affiliated with or endorsed by Amazon Web Services (AWS)