OpenAI's frontier models and the Codex coding agent are generally available on Amazon Bedrock, letting enterprises fold OpenAI usage into existing AWS security, procurement, and billing workflows — and putting two model providers on the same substrate.
Amazon Web Services customers can now run OpenAI's latest frontier models and its Codex coding agent inside the same AWS security, compliance, and billing infrastructure they already use, without negotiating a separate OpenAI commercial relationship. OpenAI announced the general availability of its frontier models and Codex on AWS on June 1, 2026, with access delivered through Amazon Bedrock, AWS's managed service for hosting and routing between third-party foundation models.
The release lands just over a month after the formal end of OpenAI's exclusive cloud arrangement with Microsoft. CNBC reported on April 28, 2026 that OpenAI was bringing its models to AWS as that exclusivity unwound, and GeekWire and Forbes tracked the Bedrock arrival as a direct consequence. The June 1 GA post is the timestamp that matters for enterprise procurement teams, and it is the moment the model stops being a rumor and becomes a line item.
The Bedrock release is really two announcements stapled together. The first is OpenAI's frontier model family — including GPT-5.5 and GPT-5.4, per the AWS News Blog post on getting started with OpenAI models and Codex on Bedrock and the parallel AboutAmazon announcement — available through the same Bedrock API surface enterprises already use to reach Anthropic, Meta, Mistral, and AWS's own Titan and Nova families. The second is Codex, OpenAI's coding agent, on the same substrate.
That distinction matters. Model access is a procurement and governance story. Codex is a developer-workflow story: OpenAI's own announcement notes that Codex is already used by more than 5 million people every week, and putting it on Bedrock means engineering teams that have standardized their identity, logging, and code-execution guardrails on AWS can now reach that tool inside the same perimeter.
The mechanism the announcement leans on is friction reduction, not redirection. Enterprises that have already cleared OpenAI's vendor review once — security questionnaires, data handling attestations, procurement paperwork — can now route OpenAI usage through the AWS relationships they already maintain. OpenAI frames the release as removing a barrier to production AI by letting teams apply their existing AWS operational workflows to OpenAI usage, and the AWS blog makes the same case from the other side: Bedrock customers can add OpenAI to the same security, governance, and compliance posture they apply to every other model on the service.
For an enterprise that has already committed spend to AWS — through Enterprise Discount Programs, Savings Plans, or committed-use contracts — the practical effect is that OpenAI inference can now sit on the same bill and inside the same budget envelope, rather than requiring a parallel OpenAI contract with its own procurement, legal, and finance motion. That is expansion of usable spend inside an existing envelope, plus consolidation of billing, rather than a literal redirection of dollars from one provider to another.
AWS is making OpenAI available in both its Commercial and GovCloud regions, which is a meaningful detail for regulated and public-sector buyers. GovCloud is the AWS partition that holds the accreditations federal and other public-sector customers typically require, and adding OpenAI there means those buyers no longer have to choose between staying inside GovCloud and using OpenAI's latest models. The AboutAmazon release confirms GovCloud as part of the launch footprint.
The customer voice in OpenAI's announcement is Amgen, the biotechnology company, which is positioned as evidence that the friction reduction matters for regulated, scientific buyers — not as generic cheerleading. For a life-sciences organization, the relevant question is whether a model can be deployed inside the validated, audited AWS environment their researchers and clinical workflows already depend on. Bedrock on GovCloud — and on the regulated commercial partitions — is what makes that answer yes.
Putting OpenAI on Bedrock is a genuine competitive realignment, and it should be named as one rather than parked at the end. AWS already sells its own frontier models and has a deep partnership with Anthropic; both Anthropic's Claude family and AWS's Nova models are first-class citizens on Bedrock. Adding OpenAI to that lineup means Bedrock is now an explicitly multi-vendor substrate, and the decision about which model to use for a given workload is being pushed up to the application layer rather than being made at the cloud-contract layer. CIO Dive's coverage of AWS's agentic posture reads Bedrock's expansion as part of AWS's broader bet that enterprises want a single governed surface for agentic AI, regardless of which lab trained the underlying model.
For Microsoft, the symmetry is uncomfortable. Azure remains a primary OpenAI partner, and the post-exclusivity arrangement is reportedly a large, committed cloud relationship. But the message from Bedrock is that enterprises that have standardized on AWS for security, identity, and procurement reasons no longer have to go outside that perimeter to use OpenAI's best models or its coding agent.
Several things the announcement does not — and should not be assumed to — settle. Bedrock pricing for OpenAI models, regional quota, enterprise contractual mechanics such as how OpenAI usage draws down AWS commitments, and the revenue share between AWS and OpenAI are not disclosed in the captured material, and buyers will need to read the Bedrock pricing page and their own AWS contracts to understand unit economics. Independent analyst commentary on Azure-versus-Bedrock margin posture and on the implications for AWS's partnership with Anthropic is thin in the available coverage; the competitive realignment is real, but the financial shape of it is still an open question.
What is settled is the surface. OpenAI's frontier models and Codex are now reachable through the same Bedrock APIs enterprises already use, in both Commercial and GovCloud regions, billed through AWS, governed by AWS-native controls, and validated for the same compliance posture as every other model on the service. For an enterprise that has already done the work to standardize on AWS, the procurement and security review for adding OpenAI just got dramatically shorter.