OpenAI Lands on AWS in $38B Deal, Bringing GPT-5.5 to Bedrock

OpenAI and Amazon Web Services announced a multi-year strategic partnership Tuesday that puts OpenAI's frontier models on AWS infrastructure for the first time, with a $38 billion compute commitment and immediate availability of GPT-5.5, Codex, and a new Bedrock Managed Agents service. The deal is the most consequential post-Microsoft cloud move OpenAI has made — and it lands one day after the two companies dismantled their exclusivity agreement.

The Deal in Numbers

Under the partnership, AWS becomes a strategic compute provider for OpenAI's core AI workloads, with deployment beginning immediately and full capacity targeted for end of 2026, with expansion options through 2027 and beyond. AWS will provide Amazon EC2 UltraServers featuring hundreds of thousands of accelerator chips and the ability to scale to tens of millions of CPUs.

The $38 billion figure represents OpenAI's committed spend on AWS over the term of the agreement. That puts AWS in a similar tier to Oracle and SoftBank as compute partners for OpenAI's training and inference needs, though still smaller than the multi-hundred-billion-dollar commitments tied to the Stargate buildout.

OpenAI Models, Now on Bedrock

The customer-facing piece of the announcement is just as important as the compute commitment. AWS customers can now access OpenAI's frontier models — including GPT-5.5, OpenAI's latest flagship — through Amazon Bedrock, AWS's managed inference service. The Codex agent for coding is also available through Bedrock.

A new service called Amazon Bedrock Managed Agents Powered by OpenAI lets enterprises build agents that maintain context, execute multi-step workflows, use tools, and take actions across business processes — without standing up the orchestration infrastructure themselves. Bedrock customers get the same enterprise security, IAM, and operational controls AWS offers for its other model providers.

For enterprises that have standardized on AWS, this collapses one of the most awkward parts of using OpenAI: the procurement and security review for a separate vendor. OpenAI's models now sit alongside Anthropic's Claude, Meta's Llama, and Amazon's own Titan models inside the same Bedrock console.

Why Now

The timing is not accidental. On Monday, Microsoft and OpenAI announced an amended partnership that ends Microsoft's exclusive cloud arrangement with OpenAI and removes the so-called AGI trigger clauses. The amendment freed OpenAI to sell on any cloud — and within 24 hours, OpenAI had landed on the largest competing cloud.

That sequencing matters. Microsoft remains a major OpenAI investor and partner, and Azure will continue to host significant OpenAI workloads. But OpenAI now has multi-cloud distribution as a core strategy, not an exception. Google Cloud is widely expected to be next.

For AWS, the deal addresses a competitive gap that had been growing visible. Anthropic's Claude has been the headline model on Bedrock and a core part of AWS's enterprise AI pitch. But many of the largest enterprise customers also wanted access to OpenAI's models, and the Microsoft-exclusivity arrangement made that messy. AWS's response was to wait Microsoft out — and then move the moment OpenAI was free.

Industry Implications

The cloud-AI competitive dynamic has been shifting all year, and this deal reframes it. Until Tuesday, the rough alignment was: Microsoft + OpenAI vs. Google + Gemini vs. AWS + Anthropic. That tripod is now more of a marketplace, with all three hyperscalers competing to offer the same models on broadly similar terms.

For enterprises, this is good news. Vendor lock-in to a particular model provider is now easier to avoid, and Bedrock-style multi-model offerings let teams swap models without rewriting integrations. For Anthropic, AWS's biggest model partner, the practical impact is competitive pressure: Claude now shares console real estate with GPT-5.5 inside the customer accounts AWS has spent two years building.

For Microsoft, the deal removes a key reason customers had to standardize on Azure. OpenAI's models were a marquee Azure feature; they're now available on the two largest competing clouds. Microsoft's response so far has been to lean into Copilot and the application layer, where it still has structural advantages.

Expert Reactions

Industry analysts on X were quick to read the deal as a continuation of OpenAI's diversification strategy. Several pointed out that the $38 billion figure, while large, is a fraction of OpenAI's total committed compute spend across Microsoft, Oracle, and the Stargate buildout — meaning AWS becomes a meaningful but not dominant supplier.

Cloud-economics watchers noted that AWS Bedrock pricing for OpenAI models will be a key signal. If AWS prices aggressively, the move could pull workload share from Azure faster than expected. If pricing matches Microsoft's, the deal is more about enterprise distribution than capacity arbitrage.

What's Next

OpenAI says all $38 billion of capacity is targeted to be deployed before the end of 2026, with expansion options into 2027 and beyond. AWS customers can experiment with OpenAI models through Bedrock starting today, with Bedrock Managed Agents rolling into broader availability over the coming weeks.

The next domino to watch is Google Cloud. With the Microsoft exclusivity gone, Google has the same opportunity AWS just took. Whether Google moves quickly will depend in part on the politics of competing directly with Gemini in its own Vertex AI catalog.

Bottom Line

OpenAI is now a multi-cloud model provider, and the era of treating ChatGPT as an Azure-exclusive product is over. For enterprises, the practical change is immediate: GPT-5.5 is one click away inside AWS Bedrock. For the cloud business, this is the clearest signal yet that frontier-model access is becoming a commodity layer — and that the competitive battle has moved to agent infrastructure, pricing, and integration depth.