Enterprises have a problem: they want to use frontier AI models, but they don’t want to rip out their existing AWS infrastructure, security protocols, and procurement workflows to do it. This week, OpenAI and AWS solved that problem by bringing GPT-5.5, Codex, and managed agents directly into Amazon Bedrock. Teams can now build AI applications without choosing between best-in-class models and the systems they already trust.
OpenAI’s Frontier Models Are Now Available Inside AWS Infrastructure
OpenAI and AWS announced the launch of OpenAI models, including GPT-5.5, directly on Amazon Bedrock in limited preview. The integration sounds incremental but reshapes how enterprises deploy AI. Previously, companies had to choose: use OpenAI through their public API, or stick with AWS’s own model catalog. Now they get both—within the same infrastructure, security controls, and billing systems.
Why this matters: enterprises have massive AWS commitments because AWS handles their identity systems, compliance requirements, and security protocols. If you’re a financial services firm or healthcare company with regulatory obligations, running AI workloads across multiple cloud platforms creates friction, audit burden, and operational complexity. By embedding OpenAI models directly into Bedrock, AWS customers can now apply OpenAI’s capabilities—including access to GPT-5.5—to their existing workloads without architectural gymnastics.
Imagine you’re a mid-market SaaS company building a customer support tool. Your infrastructure is on AWS, your security team has pre-approved AWS’s compliance controls, and your spending is locked into an AWS commitment. You want to use GPT-5.5 to power intelligent routing and ticket summarization. Before this announcement, you’d either spin up a separate OpenAI account and manage API costs separately, or settle for a weaker in-house model. Now you can build the feature directly in Bedrock, count it toward your AWS spend, and keep everything within your existing security perimeter.
This move solidifies AWS’s position as the enterprise AI platform. Companies won’t fragment their AI stacks across multiple cloud providers if they don’t have to—the operational overhead is punishing. Expect AWS’s Bedrock service to become the default AI deployment layer for enterprises already committed to AWS.
Codex Moves Into Bedrock—Making Code Generation Enterprise-Grade
Over 4 million people now use Codex every week, according to OpenAI’s report. The tool has expanded far beyond code completion; teams are using it to write code, explain systems, refactor applications, generate tests, and modernize legacy codebases. Now Codex is available directly on Amazon Bedrock, configured to use OpenAI models served from the same infrastructure as a customer’s other workloads.
Why this matters: developer tools usually live in the cloud as disconnected SaaS products. Your IDE extension talks to OpenAI’s servers, your Slack bot talks to Anthropic, and your code search tool talks to somewhere else entirely. This fragmentation means data leaves your infrastructure, it’s hard to enforce consistent security policies, and you’re paying multiple vendors separately. By bringing Codex into Bedrock, AWS lets enterprises run their coding AI agent inside their own security boundary—and apply it toward their existing AWS commitments.
For developers in regulated industries or companies with strict data residency requirements, this matters. If your team is working on financial software or healthcare applications, you’ve probably had to disable GitHub Copilot or other cloud-based coding tools because of compliance concerns. Codex on Bedrock gives you the same capability—4 million weekly users know it works—without the data leaving AWS. Developers can enable it in Visual Studio Code, the Codex desktop app, or the Codex CLI, all configured to use Bedrock as the backend instead of OpenAI’s public servers.
This also signals a larger shift: the best developer tools will increasingly be embedded inside enterprise platforms rather than sold as standalone SaaS products.
Managed Agents Let Enterprises Build Production AI Without Building Infrastructure
Amazon Bedrock Managed Agents, powered by OpenAI, handle the infrastructure layer that usually trips up AI projects: orchestration, tool use, state management, and governance. Instead of your team building an agent framework from scratch—wiring up retrieval, managing context windows, orchestrating multi-step workflows, and implementing safety guardrails—Bedrock Managed Agents provide all of that built-in.
Why this matters: the gap between a working prototype and production-grade AI application is massive. Prototypes usually fail when they hit real data volumes, deal with hallucinations at scale, or need to handle edge cases in complex workflows. Managed agents solve this by offering a pre-built operational layer with built-in integration across Amazon’s security and compliance controls. Tool orchestration, context management, and audit logging are already done.
Consider a team building an AI agent to automate customer support workflows. In prototype, the agent responds to simple questions. In production, it needs to handle customer authentication, check a ticket history database, escalate appropriately, maintain context across multi-turn conversations, and log every action for compliance review. Building that yourself requires deep expertise in LLM architecture, API design, and production systems. Managed Agents give you the skeleton; your team focuses on the business logic.
This is the future of enterprise AI development: managed platforms will become table stakes. Companies that force developers to build agentic orchestration from first principles will lose to teams that buy pre-built platforms. Bedrock Managed Agents, by centralizing this responsibility, will likely become the default choice for enterprises already committed to AWS.
FAQ
Q: Do I have to use Amazon Bedrock to access OpenAI models on AWS? A: Yes. Bedrock is AWS’s managed AI service, and it’s the infrastructure layer that handles model serving, security, billing, and compliance. You can’t use GPT-5.5 on AWS outside of Bedrock—but that’s actually the advantage: all your AI workloads, whether from OpenAI, Anthropic, or AWS’s own models, run through the same security and billing interface.
Q: What compliance benefits does Bedrock Managed Agents provide? A: Bedrock Managed Agents include built-in integration with AWS’s security, identity, and compliance controls. Audit logging, IAM policies, encryption, and compliance features like HIPAA or SOC 2 are already configured. You don’t have to retrofit governance onto an agent framework you built yourself—it’s part of the product.
Q: Can I migrate existing OpenAI API integrations to Bedrock Managed Agents? A: Not directly. Bedrock Managed Agents are a different architecture designed for complex, multi-step workflows; they’re not a drop-in replacement for simple OpenAI API calls. However, new agent projects will almost certainly target Bedrock Managed Agents if your infrastructure is already on AWS, because the operational burden is lower.
Key Takeaways
-
Enterprises using AWS can now deploy OpenAI’s best models—including GPT-5.5—without leaving their security perimeter or fragmenting their infrastructure. Expect this to become the standard pattern for enterprise AI deployments over the next 18 months.
-
Codex on Bedrock removes the biggest barrier to AI-powered developer tools in regulated industries: data residency and compliance. Teams in healthcare, finance, and government can now adopt the tool without compromising security requirements.
-
Bedrock Managed Agents shift the burden of production AI development away from custom engineering toward configuration and business logic. Teams that invest in learning Bedrock’s agentic layer will ship faster and with fewer operational surprises than teams building agents from scratch.
-
AWS’s position as the default enterprise AI infrastructure platform is now nearly unassailable. If you’re already committed to AWS for compute, storage, and compliance, there’s almost no reason to run your AI workloads elsewhere.
-
The shift toward managed AI infrastructure (rather than model APIs) suggests that the next wave of AI adoption will be led by platform teams configuring pre-built agentic layers inside their existing infrastructure.