OpenAI just made it a lot easier to ignore the question of which cloud you’re on. As of April 28, 2026, OpenAI’s GPT models, Codex, and Managed Agents are officially available on AWS — meaning enterprises that have built their entire infrastructure on Amazon’s cloud no longer need to route traffic through a separate API or wrangle cross-cloud data agreements. OpenAI on AWS is here, and the implications go well beyond a simple distribution deal.
How We Got Here
For most of OpenAI’s existence, accessing its models meant calling the OpenAI API directly, or — if you were a Microsoft shop — going through Azure OpenAI Service, which has been available since 2023. That arrangement made sense given Microsoft’s deep investment in OpenAI and the exclusive cloud partnership that came with it. But enterprise IT doesn’t work in neat loyalty lines. A massive chunk of the world’s production workloads live on AWS. Always have.
The result was an awkward gap. Companies running on AWS who wanted GPT-4o or o3 had two options: call the OpenAI API over the public internet (acceptable for prototypes, uncomfortable for regulated industries) or spin up Azure infrastructure just for AI inference. Neither option was clean. Security teams hated the first one. Finance teams hated the second.
Something had to give. The renegotiated Microsoft deal, which loosened some of the exclusivity constraints, opened the door. AWS walked through it.
What’s Actually Available — and How It Works
This isn’t just an API endpoint hosted on different hardware. The integration is deeper than that, and the three components work differently enough that it’s worth separating them.
GPT Models via AWS
OpenAI’s flagship GPT models — currently including GPT-4o and the o-series reasoning models — are now accessible directly within AWS accounts. Enterprises can call these models without data leaving their AWS environment in ways that violate their compliance posture. That matters enormously for healthcare, financial services, and government contractors who have strict data residency and egress requirements.
Billing gets consolidated into AWS invoices, which sounds boring but is genuinely significant. Getting a new SaaS vendor approved through procurement is a months-long process at large companies. If the charges show up on the existing AWS bill, that friction largely disappears.
Codex on AWS
OpenAI Codex, the AI software engineering agent that can read codebases, write code, run tests, and fix bugs autonomously, is now available in AWS environments. If you’ve been following what Codex can actually do, you know this is a bigger deal than it sounds. Codex isn’t a code autocomplete tool anymore — it’s closer to an asynchronous junior engineer you can assign tasks to.
Running Codex inside AWS means it can interact with your existing AWS services: CodeCommit repositories, Lambda functions, RDS databases, S3 buckets. The security model stays within your VPC. That’s a fundamentally different security posture than having an external agent that needs to authenticate into your infrastructure from outside.
Managed Agents
This is the most forward-looking piece. Managed Agents are pre-built, OpenAI-hosted agent configurations that enterprises can deploy and orchestrate through AWS. Think of them as agent templates with defined capabilities, guardrails, and tool access — ready to drop into a workflow without building the scaffolding yourself.
Key features across the full offering include:
- Native AWS IAM integration for access control and permissions
- VPC support, keeping inference traffic off the public internet
- AWS CloudTrail logging for audit trails on every model call
- Consolidated billing through existing AWS accounts
- Support for AWS PrivateLink for fully private connectivity
- Codex agent access to AWS developer services (CodeCommit, Lambda, etc.)
- Managed Agent deployment through AWS console and CLI
Pricing follows OpenAI’s existing token-based model, applied through AWS marketplace billing. There’s no markup disclosed in the announcement, but it’s worth watching whether AWS adds a platform fee over time — that’s their standard playbook with marketplace products.
Who This Actually Helps — and Who It Disrupts
The Enterprise IT Angle
The immediate winners are enterprise security and compliance teams. These are the people who’ve been blocking or delaying AI adoption not because they distrust the models, but because they couldn’t sign off on data handling when inference happened outside their controlled environment. AWS-native deployment gives them the audit logs, the network controls, and the contractual framework they need to say yes.
I wouldn’t be surprised if this unlocks a significant wave of enterprise deployments that have been stuck in security review limbo for months. The technology has been ready. The paperwork hasn’t been.
What This Means for Amazon Bedrock
Here’s the thing: Amazon runs its own AI model service called Amazon Bedrock, which offers Anthropic’s Claude models, Meta’s Llama, Mistral, and others. OpenAI on AWS is now a direct competitor to Claude on Bedrock — both available in the same console, on the same bill, from the same cloud provider.
Amazon is presumably fine with this, since they make money either way. But it does create a genuinely interesting dynamic where AWS customers can now run A/B evaluations between GPT-4o and Claude Sonnet on identical infrastructure with identical billing. That kind of direct comparison, at scale, is going to produce a lot of data about which models actually perform better for specific enterprise tasks. Anthropic should be watching this closely. So should Google, whose Gemini models are available through Google Cloud Vertex AI but not through AWS.
The Microsoft Question
The elephant in the room is Azure. Microsoft has spent years building Azure OpenAI Service as the enterprise-grade access point for OpenAI models. That’s still a strong offering with deep integrations into Microsoft 365, GitHub Copilot, and the broader Microsoft stack. But for companies that aren’t Microsoft shops, Azure OpenAI always required a degree of platform commitment that felt uncomfortable.
AWS availability removes the pressure to adopt Azure just for AI access. Whether that meaningfully dents Azure’s AI revenue is an open question — but it removes one of the cleaner arguments for enterprises to expand their Azure footprint. OpenAI is clearly making a deliberate choice to distribute more broadly rather than stay anchored to any single cloud partner.
Developers and Platform Teams
For engineering teams, the practical difference is significant. Running Codex inside your AWS environment means tighter integration with the tools developers already use. CI/CD pipelines running in AWS can now trigger Codex agents natively. Infrastructure-as-code workflows can incorporate AI-driven code review without external API calls. Managed Agents can be embedded directly into Step Functions workflows.
This is what “enterprise-ready AI” actually looks like in practice — not a demo, but something that fits into existing deployment patterns without requiring developers to learn a new platform or security teams to approve a new vendor.
Key Takeaways
- OpenAI GPT models, Codex, and Managed Agents are now available on AWS as of April 28, 2026
- Data stays within AWS VPCs — no public internet inference for enterprises using PrivateLink
- Billing consolidates into existing AWS accounts, cutting procurement friction significantly
- Codex can now directly access AWS developer services inside secured environments
- This creates direct competition with Claude on Amazon Bedrock — same console, same bill
- Azure OpenAI Service loses its position as the only enterprise-grade cloud deployment option for GPT models
- Managed Agents lower the barrier for enterprises to deploy multi-step AI workflows without custom scaffolding
Frequently Asked Questions
What OpenAI models are available on AWS?
The initial availability includes OpenAI’s GPT-4o and o-series reasoning models, along with Codex and Managed Agents. The exact model roster is expected to stay current with OpenAI’s general availability lineup, similar to how Azure OpenAI Service has historically rolled out new models within weeks of their general release.
How is this different from just calling the OpenAI API?
The key differences are network security and billing. With AWS-native deployment, inference traffic can stay entirely within your VPC using AWS PrivateLink, meaning it never traverses the public internet. Billing is consolidated into your AWS invoice, which removes the need to onboard OpenAI as a separate vendor through procurement — a meaningful operational difference at large enterprises.
Does this replace Azure OpenAI Service?
No — Azure OpenAI Service remains fully operational and is still the better choice for organizations deeply embedded in the Microsoft stack, using GitHub Copilot, Microsoft 365 Copilot, or other Azure-native services. What AWS availability does is give non-Microsoft enterprises an equivalent option without requiring any Azure footprint at all.
Can Codex on AWS access my existing codebase and services?
Yes, that’s one of the more compelling parts of this announcement. Codex running in an AWS environment can interact with AWS developer services including CodeCommit, Lambda, and others, subject to IAM permissions you define. This makes it possible to run Codex as part of existing CI/CD pipelines without external API calls or cross-environment authentication headaches.
The broader pattern here is worth watching: OpenAI is clearly building toward a world where its models and agents are available wherever enterprises already operate, rather than requiring enterprises to come to a specific platform. For AI agents in particular — where the tooling is evolving fast — that kind of infrastructure neutrality could matter as much as model quality when enterprises make long-term bets. The cloud wars are no longer just about compute. They’re about who controls the AI runtime.
