Amazon launches OpenAI products on AWS after review of Microsoft exclusive partnership

Amazon rolled out new OpenAI products on AWS almost immediately: Amazon Bedrock received fresh OpenAI models, Codex, and the Bedrock Managed Agents service for running agent systems. The launch occurred just a day after OpenAI renegotiated its agreement with Microsoft and removed the cloud exclusivity that had kept its products locked within Azure.

What appeared in AWS

AWS opened limited preview across three directions. Amazon Bedrock now hosts the latest OpenAI models, including GPT-5.4, with GPT-5.5 promised for imminent availability. Codex is also coming — a programming agent that can be run via CLI, desktop application, and Visual Studio Code extension. The third element is Amazon Bedrock Managed Agents, a new service for building and running OpenAI agents directly within AWS infrastructure.

• Access to OpenAI models through familiar Amazon Bedrock APIs
• Codex inside AWS environment with authorization through AWS credentials
• Managed Agents for production-ready agent scenarios
• Unified corporate controls: IAM, encryption, PrivateLink, and CloudTrail

For enterprise teams this is an important detail: no need to spin up separate infrastructure for OpenAI or rebuild a security layer from scratch. Amazon is betting that companies will want to get models, a coding agent, and agent runtime in a single loop where access, logging, governance, and budget limits are already configured. Plus using OpenAI services can be counted toward existing cloud commitments to AWS. For procurement and compliance, this greatly simplifies deployment.

Why this happened now

Timing is key here. On April 27, 2026, Microsoft and OpenAI announced a renegotiation of their partnership: Microsoft remains the primary cloud partner, but its license becomes non-exclusive, and OpenAI gains the right to serve all its products through any cloud provider. Already on April 28, AWS seized this window and rolled out OpenAI in Bedrock with virtually no pause. For the market, this is de facto confirmation that the new scheme worked immediately.

"We have many developers and companies who want to run services on

OpenAI models in AWS," — Andy Jassy.

The groundwork had been prepared in advance. Back in February, Amazon and OpenAI announced a strategic partnership with Amazon investments up to $50 billion, joint development of Stateful Runtime Environment, and plans for developing an agent platform on Bedrock. At that time, the legal bottleneck was Microsoft's previous exclusivity. After its loosening, the February agreements began turning into concrete products, not just press-release promises.

What business gets

The most practical part of this announcement — not new names, but the delivery method. OpenAI comes to AWS in a form convenient for large companies: with access management, logging every agent action, running within customer environment, and without needing to teach teams a new security model. For Codex this is especially important because the agent works alongside real code bases, IDEs, and internal development services. Security teams and platform teams will have fewer new contours to coordinate.

Managed Agents are also aimed not at demos, but at long-term scenarios. AWS specifically emphasizes that each agent gets its own identity, logs actions, and uses the OpenAI agent harness for more precise control over lengthy tasks. This signals that competition between clouds is shifting from simple model hosting to a more expensive layer: who better packages agents, memory, orchestration, and enterprise security. And it's this layer that usually determines whether a pilot makes it to production.

What this means

For the market, this is further evidence that the era of "one model provider — one cloud channel" is ending. OpenAI is expanding distribution, AWS is strengthening Bedrock as an enterprise entry point into agent AI, and Microsoft is losing part of its exclusive advantage. For business, choice is becoming broader: you can take strong OpenAI models without leaving your AWS perimeter.