OpenAI ends Microsoft exclusivity, expands to AWS and Google Cloud; Dell becomes on-prem channel
The end of the Microsoft–OpenAI exclusive cloud arrangement is the most consequential enterprise-AI restructuring of the year. Under the new terms, OpenAI can sell its frontier models through AWS Bedrock and Google Cloud Vertex, dramatically widening enterprise access for customers who refused to standardize on Azure. A former Microsoft VP told Windows Latest the company "missed the AI wave" — and Copilot is reportedly being scaled back inside Windows 11.
Dell's new role as OpenAI's on-prem channel for frontier models is the parallel story: regulated industries (finance, defense, healthcare) that cannot or will not run inference in the public cloud now have a sanctioned air-gapped path to GPT-5.5 and successor models. That directly attacks Microsoft's Azure Local / Foundry Local hybrid pitch, which Microsoft is hardening this week by extending support for OpenAI's open-weight gpt-oss models.
Guaranteed Capacity is the customer-facing financial instrument: Pro and enterprise customers can pre-purchase compute allocations against future scarcity. Sam Altman called it a "win-win for planning" — OpenAI gets demand-side commitments to underwrite buildout, customers lock in capacity ahead of the next training run that will tighten supply. This mirrors what airlines and chip foundries do, and it formalizes the new reality that compute is the binding constraint, not model quality.
The White House angle adds policy weight: an executive order in preparation would seek early government access to frontier models — implicitly favoring vendors with multi-cloud and on-prem distribution. Reading the moves together: OpenAI is converting cloud-distribution leverage into a Switzerland-style position (sell everywhere, lock in capacity, partner with Dell for on-prem) while Microsoft pivots toward owning the open-weight and hybrid layer. The Karpathy departure to Anthropic and Musk's failed lawsuit (1,067 HN points) frame the talent and legal context.