The exclusive frontier partnership died the moment enterprise procurement stepped in.
OpenAI announced that its frontier models and Codex workflows are coming to Google Cloud's Model Garden, letting corporate customers draw down their existing Google Cloud spend commitments to run GPT workloads. Six months ago, the exclusive intellectual property license with Microsoft quietly ended, followed twenty-four hours later by an agreement with Amazon Web Services. OpenAI's models now sit natively inside all three hyperscalers.
Industry coverage frames this as a distribution race, but the underlying accounting tells a more constrained story. Enterprise buyers rarely pay for AI API calls out of fresh discretionary cash. They pay out of multi-year, nine-figure minimum spend commitments negotiated with their primary cloud infrastructure provider. If a Fortune 500 bank or insurer has committed three hundred million dollars over five years to Google Cloud or AWS, every dollar routed to Azure to query an OpenAI endpoint counts as an unbudgeted cash leak. The CFO asks why the lab cannot simply burn down the committed balance sheet credits sitting on GCP.
For two years, Microsoft used exclusivity to force those buyers onto Azure, treating OpenAI's frontier weights as the sole anchor tenant capable of moving legacy workloads. That works when frontier models hold a wide, unquestioned performance lead. It breaks down once Anthropic and open-weights models close the capability gap, while enterprise CIOs refuse to take on multi-cloud egress penalties or duplicate security audits just to run inference. The cost of friction began to exceed the value of the exclusivity payout.
On the lab side, OpenAI is raising capital at a 1.4 trillion dollar valuation while managing staggering compute bills and an impending public offering. At that scale, you cannot leave two-thirds of the addressable enterprise balance sheet untouched. Frontier labs have transitioned from proprietary research institutions into software wholesalers. They need every sovereign balance sheet and corporate committed-spend bucket open for depletion.
The hyperscalers accept the arrangement because they have already won the base layer. Hosting competing frontier models keeps enterprise data pipelines, vector indexes, and compute instances inside their own data centers. When model margins inevitably compress under cross-cloud price matching, the underlying kilowatt-hours and networking egress remain high-margin annuities for the landlord.
