Wednesday, 30 September 2026

Where technology leaders come to think out loud

The BriefArtificial Intelligence

AWS puts Moonshot’s Kimi K3 on Bedrock with US and global routing only

The Chinese lab’s open-weight model is now generally available to Bedrock customers. For UK buyers with data residency rules, the routing options at launch matter more than the parameter count

The facts
  • Generally available on Amazon Bedrock: 18 September 2026
  • Moonshot AI claims about 2.5 times the scaling efficiency of Kimi K2
  • AWS says the global profile costs about 10% less than a geographic one
  • Moonshot’s license, changed in July, requires a commercial deal from firms offering K3 as a service with more than $20m (£15m) in annual revenue, according to Silicon UK
What it means for buyers

UK organizations with data residency obligations should treat Kimi K3 as unavailable for regulated workloads until AWS lists a European routing option. For other work, it is worth a trial on long coding and document tasks, since it runs under Bedrock controls the organization may already have approved. Before anyone builds on it, put the model’s origin and license terms through procurement and risk review, and ask AWS in writing where requests are processed under each profile.

Amazon Web Services (AWS) made Moonshot AI’s Kimi K3 generally available on Amazon Bedrock on 18 September. Moonshot, a Beijing-based AI lab, describes the model as the first open model to reach 2.8 trillion parameters, with native vision and a one-million-token context window.

AWS pitches it at long coding sessions across large repositories, analysis of many documents including scans and screenshots, and long-running agent workflows. It runs inside Bedrock’s existing controls for access, encryption and auditing, and it is the first open-weight model on Bedrock to support explicit prompt caching, which cuts input costs when the same context is reused.

Routing is the detail UK buyers will notice. AWS’s launch blog lists two inference profiles: a global one, which can send each request to any supported commercial AWS Region worldwide, and a US one that keeps processing in the US. No European profile is listed.

AWS did not disclose commercial terms. Silicon UK calls it the first major revenue-sharing agreement between a Chinese AI firm and a major US cloud company, and cites Reuters reporting that Moonshot was negotiating similar deals with Microsoft and Google.

AdvertisementZoomInfo
More from The BriefAll briefs
Scottish Government

Scotland requires impact assessments for data centers above 50MW

Orca Security

Orca Security's new partner program rewards net retention

Insight

Insight wins consultancy role in Scotland’s AI adoption program

VinciWorks

Only one in 10 sure of meeting 24-hour breach reporting, VinciWorks finds

Get the briefs every week in The VETTDD Briefing.