AWS hosts OpenAI for GPT‑5.5

OpenAI is now selling its newest models through Amazon’s cloud — and that is the real story here. On April 28, OpenAI and AWS said GPT‑5.5, Codex, and a new Bedrock Managed Agents offering are coming to Amazon Bedrock in limited preview. (openai.com) That sounds like another model distribution deal. But basically it is a power shift. For years, OpenAI was tied to Microsoft’s cloud in a way that made Azure the default home for serious OpenAI deployment. Now AWS gets to host OpenAI’s flagship stack too — not just a model endpoint, but coding tools and agent infrastructure. (openai.com) ### (openai.com)I models including GPT‑5.5 are available on Bedrock in preview. Second, Codex comes with them, so AWS customers can use OpenAI’s coding assistant through Amazon’s platform. Third, AWS launched Bedrock Managed Agents powered by OpenAI, which wraps OpenAI’s agent harness in AWS runtime, security, identity, and logging controls. (openai.com) ### Why is Bedrock the important part? Because Bedrock is AWS’s control plane for enterprise AI. Big companies already use it to mix models, permissions, procurement, and governance in one place. So this is not just “OpenAI runs on AWS hardware.” It means buyers can consume OpenAI through the same APIs, security model, and purchasing machinery they already use for Anthropic, Amazon, and other models on Bedrock. (aws.amazon.com) ### Why does that change the market? The catch with enterprise AI was never only model quality. It was where the model lived. If a company standardized on AWS but wanted frontier OpenAI systems, Azure often sat in the middle. This deal removes a lot of that friction. AWS even says OpenAI model and Codex usage can(aws.amazon.com)aws.amazon.com) ### Did something else change first? Yes — and this is why the timing matters. OpenAI posted “the next phase” of its Microsoft partnership on April 27, one day before the AWS launch. The new setup keeps Microsoft access to OpenAI tech, but it no longer looks like the old exclusive-cloud arrangement. Then AWS immediately announced OpenAI on Bedrock. That sequence makes the strategic break pretty hard to miss. (openai.com) ### Is GPT‑5.5 itself a big enough draw? Probably yes. OpenAI calls GPT‑5.5 its “best frontier model,” and the model is positioned for coding, research, tool use, and longer real-world tasks. That matters because Bedrock is strongest when customers want one managed place to run serious production workloads, not just demo chatbots. Putting OpenAI’s newest flagship there gives AWS a much(openai.com)the top closed model. (openai.com) ### What about lock-in? It cuts one kind of lock-in and creates another. Customers no longer need Azure to get OpenAI. But if they adopt OpenAI models, Codex, and Managed Agents deeply inside Bedrock — with AWS identities, logs, security controls, and cloud-commit discounts — switching later gets harder. It is less a pure model choice and more a full-stack infrastructure choice. That is the real leverage AWS bought here. (aws.amazon.com) ### Does this kill multi-model strategies? Not at all. Bedrock’s whole pitch is model choice, and Anthropic remains deeply embedded in AWS’s AI story. So enterprises can still compare OpenAI for frontier general-purpose work, Claude for coding-heavy or enterprise workflows, and Amazon’s own stack for cost or inte(aws.amazon.com)more central. (aws.amazon.com) The bottom line is simple. OpenAI did not just add another reseller. It crossed a strategic boundary. AWS is now a first-class home for OpenAI’s newest models and agent tools — and that reshapes who controls enterprise AI deployment from here.