Microsoft weighs lawsuit

Microsoft and OpenAI published a joint statement on Feb. 27 saying the new Amazon partnership “does not change the terms of our long‑standing partnership” and that their commercial and revenue‑share relationship remains unchanged. (blogs.microsoft.com) OpenAI and Amazon’s Feb. 27 corporate releases name AWS as the “exclusive third‑party cloud distribution provider” for OpenAI’s Frontier product and commit OpenAI to consume roughly 2 gigawatts of AWS Trainium capacity to support a forthcoming stateful runtime on Amazon Bedrock. (openai.com) The Financial Times reported on Mar. 18 that some Microsoft executives view AWS hosting of Frontier as potentially breaching the existing contract—sources told FT the dispute centers on whether the deal breaks the exclusivity terms in spirit if not in strict wording, and that talks are ongoing to resolve distinctions before any litigation. (ft.com) Technical coverage from InfoQ and others frames the core operational distinction as “stateless” vs “stateful” hosting: Azure is said to retain stateless API exclusivity while AWS will host a stateful runtime for agents, which shifts where session state, memory, and identity live during agent orchestrations. (infoq.com) A filings analysis published by GeekWire on Mar. 1 describes the $50B commitment as part of a larger funding and infrastructure arrangement with multiple performance triggers and Trainium supply commitments, meaning financial obligations and compute delivery are tied to measurable capacity and usage milestones. (geekwire.com) That architectural and contractual split implies enterprise deployments of Frontier‑style agents will change telemetry and failure‑mode boundaries: stateful runtimes on Bedrock will centralize long‑running memory/identity, affecting where logs, traces and durable checkpoints must be collected for reliable agent debugging. (openai.com)