GPT-5.5 Lands on AWS

If you work in IT and your company runs on AWS, you have probably had this awkward moment in the last year. Someone in marketing wants to use GPT for a project. Your security team wants everything to flow through your existing AWS controls. The procurement team rolls their eyes at the idea of opening a separate vendor relationship with Microsoft. And so a perfectly reasonable AI use case turns into a six week paperwork exercise.

That just got a lot easier.

What Happened This Week

On April 28, Amazon Web Services and OpenAI announced that OpenAI's flagship models, including the brand new GPT-5.5, are now available on Amazon Bedrock. ¹ Bedrock is Amazon's managed service for running large language models, the same place AWS customers have been running Anthropic's Claude, Meta's Llama, and Mistral's models for a couple of years now. OpenAI is the last of the major frontier labs to land there. ²

Three things shipped at once. The models themselves (GPT-5.5 and GPT-5.4 in limited preview), Codex (OpenAI's coding agent, the one that can read your repo and propose edits), and a new product called Bedrock Managed Agents that handles the plumbing for AI agents that need to act across your systems. ¹

If you have been waiting for "OpenAI but on the cloud my company already pays for," the wait is over.

The Seven Year Lock In That Just Quietly Ended

To understand why this matters, you have to know what changed behind the scenes.

Since 2019, Microsoft Azure was the only major cloud legally permitted to host OpenAI's proprietary models. That is why every ChatGPT message you have ever sent has run on Azure infrastructure. It is why companies that wanted to use the OpenAI API at scale ended up signing Azure contracts whether or not Azure made sense for the rest of their stack. ³

That exclusivity is now gone. Azure remains the "primary" cloud partner under a license that runs through 2032, but Microsoft no longer has the unilateral lock that defined the relationship for nearly seven years. ³ The new deal also caps Microsoft's cut of OpenAI revenue and scraps a provision that would have kicked in once artificial general intelligence was achieved. Yes, that was actually in the contract. ³

Amazon moved on this fast. The exclusivity arrangement ended on a Monday. The OpenAI models landed on Bedrock the next day. ⁴

Why This Is a Practical Win for IT and AI Teams

If you live in the day to day of building or supporting AI workloads at a company, here is what actually changes.

Procurement gets simpler. If your company has an AWS Enterprise Discount commitment, OpenAI tokens consumed through Bedrock count against it. Codex usage applies to AWS spend too. ¹ For finance teams, that is the difference between two contracts and one.

Security teams stop having to invent a parallel control plane. OpenAI inference on Bedrock inherits AWS IAM (the permission system you already use), PrivateLink (so traffic does not have to traverse the public internet), CloudTrail logging, encryption keys you manage, and the same guardrails Bedrock applies to other models. ⁵ Your CISO will not need a separate review process.

For developers, Codex now authenticates through your AWS credentials and runs through Bedrock infrastructure. ¹ Same coding agent, but it shows up as a normal AWS service rather than as a separate identity.

For teams building agents, Managed Agents is meant to be the production scaffolding around tasks that run for a long time and span many steps. The pitch is that you focus on what the agent should do, while Bedrock handles state, tool use, and orchestration. ²

One caveat worth flagging: all three are in limited preview right now. General availability is described as "within weeks," but if your project depends on hitting a specific date, watch the GA announcements before committing. ¹

The Bigger Picture: Pick Your Cloud, Pick Your Model

The deeper story here is about an assumption breaking.

For most of the last few years, the working mental model was that frontier AI labs paired with single cloud providers. OpenAI was Microsoft. Anthropic was Amazon. Google was Google. If you were on the wrong cloud, you were on the wrong model.

That is not how it works anymore. Anthropic's Claude has been on both AWS Bedrock and Google Cloud's Vertex AI for a while. OpenAI is now on both Azure and AWS. Meta's models are widely available. Even Google's Gemini, while it lives mostly on Vertex, is increasingly accessible through other paths.

The new assumption is that you pick your cloud based on the rest of your infrastructure, then pick your model based on the job. Which is how it should have always worked. The lock ins were artifacts of the early, breathless phase of the AI gold rush, not a stable end state.

For working professionals, this is a quietly big deal. It means the choice between "use the AI my company already has cloud commitments with" and "use the AI that is actually best for this task" is collapsing. You can increasingly do both.

What to Actually Do With This

If your company runs on AWS and you have been blocked from using OpenAI models because of procurement complexity, this might be the moment to revisit that conversation. The friction is real, and a lot of it just went away.

If you are an architect, you can stop designing dual integration paths for "OpenAI through Microsoft" and "everything else through AWS." One Bedrock integration covers a lot more ground now.

If you are a developer who already uses Codex elsewhere, the AWS path is worth a look. Same agent, your existing identity and billing.

And if none of this applies to your day to day, it is still worth understanding the shape of the change. The AI tools landscape is not getting more complicated. It is getting more interoperable. The days of choosing your AI ecosystem the way you used to pick a side in a console war are ending. That is a good thing for the people doing the actual work, which is most of us.