AWS is adding OpenAI models, Codex, and OpenAI-powered managed agents to Amazon Bedrock in limited preview, turning OpenAI's new cloud flexibility into an enterprise product less than a week after the Microsoft partnership reset became public.

AWS announced on April 28 that Bedrock is offering OpenAI's GPT-5.5 models, Codex, and Bedrock Managed Agents powered by OpenAI in limited preview. TechCrunch framed the timing plainly: Amazon is already offering new OpenAI products on AWS after OpenAI and Microsoft loosened their exclusive arrangement.

What AWS Shipped

The limited preview has three parts. First, OpenAI's GPT-5.5 model family is available in Amazon Bedrock for building generative AI applications. Second, OpenAI Codex is available through Bedrock for coding and software-engineering workflows. Third, AWS is offering Bedrock Managed Agents powered by OpenAI, so teams can run agentic workflows with OpenAI models inside the AWS environment.

Amazon's own explainer says customers can access OpenAI models in Bedrock while using AWS security, privacy, and governance features, including private connectivity and enterprise controls. Amazon described the launch as part of its broader effort to let customers choose models from multiple providers without rebuilding their infrastructure around each one.

The Bedrock Control Plane

The important layer is not just model availability. It is where the model is being sold. Bedrock is AWS's managed foundation-model platform, which means OpenAI is arriving inside the same control plane many enterprises already use for model access, permissions, logging, evaluation, and deployment.

That matters because large buyers rarely want a new AI supplier to become a new operational island. They want model choice to sit inside existing identity systems, procurement paths, network controls, and compliance workflows. Reuters reported that AWS CEO Matt Garman said customers' production applications and data already run in AWS, which is exactly the leverage Bedrock is using.

Bedrock Surface What AWS Added Why It Matters
Models OpenAI GPT-5.5 family Enterprise teams can use OpenAI models through AWS infrastructure and governance
Coding OpenAI Codex on Bedrock Software-agent workflows move closer to cloud-native developer operations
Agents Managed Agents powered by OpenAI AWS can package OpenAI as an agent runtime, not just a model endpoint
Governance Bedrock security and control features OpenAI becomes easier to adopt for companies already standardized on AWS

Why Codex Matters

Codex is the sharper signal because it is not just another model name in a catalog. It is OpenAI's software-work layer, aimed at coding, repository tasks, and delegated engineering workflows. Bringing Codex into Bedrock gives AWS a way to sell coding agents through the same enterprise environment used for broader AI applications.

That links directly to the agent market. Buyers are moving from chatbots toward systems that can take actions, call tools, and operate across business processes. Bedrock Managed Agents powered by OpenAI gives AWS a credible answer: use OpenAI where it is strong, but keep the orchestration, permissions, and deployment surface on AWS.

The Microsoft Reset Becomes Real

This is why the timing matters. On April 27, Microsoft and OpenAI announced a new phase of their partnership that preserved Microsoft's access to OpenAI technology while making the arrangement less exclusive. OpenAI said it can now work with third-party cloud providers, while Microsoft said it retains important product and IP rights.

AWS is the first clear proof of what that means in practice. The Microsoft-OpenAI reset could have sounded like legal housekeeping. Bedrock turns it into product distribution. OpenAI gets another enterprise channel. AWS gets the model brand it did not own. Customers get OpenAI without leaving the Amazon cloud stack.

The Platform Signal

The broader signal is that frontier AI is becoming less vertically tidy. One company may build the model, another may sell the cloud control plane, a third may own the customer relationship, and all of them may still compete. The old question was which lab had the best model. The new question is which platform can make the best model usable, governable, and purchasable at enterprise scale.

That is why AWS moving quickly matters. Amazon does not need to own OpenAI to profit from OpenAI demand. It needs to make OpenAI feel native inside AWS. If Bedrock becomes the place where enterprises compare, govern, and deploy models from many labs, model choice becomes a feature of the cloud platform, not an escape from it.