OpenAI Models Reach AWS as Microsoft Cloud Exclusivity Ends

Azure-first replaces Azure-only

On April 27, 2026, and announced an amended agreement replacing Microsoft's cloud exclusivity with an Azure-first commitment: OpenAI products ship first on Azure unless Microsoft cannot or chooses not to support the capabilities they need, after which OpenAI is free to distribute them on any cloud provider. [2]

Microsoft's prior contract had already let some consumer products, such as ChatGPT, run on other clouds, but it kept exclusive rights over stateless API access to OpenAI's models. That API exclusivity is what the amendment removed, clearing the specific obstacle that had kept OpenAI's models off competing clouds. [3]

A license through 2032, a revenue share through 2030, and the AGI clause is gone

Microsoft's license to OpenAI's models and products becomes non-exclusive but extends through 2032. Microsoft no longer pays OpenAI a share of Azure OpenAI revenue, but OpenAI's payments to Microsoft continue through 2030 at the same rate as before, now under a total cap, and continue regardless of how OpenAI's technology progresses. [2]

Atlas interpretation: That last point mattered on its own: the original contract tied Microsoft's rights to whether or when OpenAI's board declared it had reached artificial general intelligence, a definition OpenAI controlled. Fixing the arrangement to calendar dates instead removes a standing source of leverage and dispute between the two companies, independent of the cloud exclusivity question. [4]

The Amazon side: a $50 billion investment and an exclusive agent tool

The freed-up path led to committing up to $50 billion to OpenAI, a $15 billion initial investment plus another $35 billion contingent on unspecified conditions being met. In exchange, OpenAI agreed to co-develop stateful runtime technology on Amazon Bedrock and gave AWS exclusive rights to Frontier, OpenAI's agent-building tool. This built on the $38 billion of AWS compute OpenAI had already contracted for in November 2025. [3]

OpenAI's models became available on Amazon Bedrock on April 28, 2026, one day after the Microsoft amendment was announced. Sam Altman described the AWS relationship on agent infrastructure as exclusive, telling AWS CEO Matt Garman's team that OpenAI was "doing this exclusively with Amazon," while Microsoft remained OpenAI's primary cloud partner for everything else under the Azure-first terms. [4]

A single cloud stopped being enough

Atlas interpretation: OpenAI's compute needs had grown past what a single provider's exclusive arrangement could accommodate without also blocking deals it wanted to make elsewhere. Restructuring the Microsoft agreement around dates and caps, rather than exclusivity tied to an open-ended technical milestone, is what let OpenAI sign a second cloud provider without breaching its first, and let Microsoft trade an exclusive claim it could no longer fully use for a bounded, durable revenue stream instead. [3][2]