OpenAI announced general availability of its frontier models and Codex on AWS as of June 1, ending a distribution arrangement that had made Microsoft Azure the exclusive cloud home for the company’s APIs. The move lets enterprises access GPT-class models without leaving the security and procurement workflows they already run on AWS.
The practical lift for enterprise buyers is real. AWS customers can now route OpenAI model calls through the same IAM roles, KMS encryption keys, and CloudWatch logging they use for every other workload. The security review that previously blocked many procurement teams from approving OpenAI tools collapses to near-zero when the infrastructure is already approved. That is the core buying-friction argument OpenAI is solving with this deal.
For Microsoft Azure, the announcement is a meaningful loss. OpenAI’s APIs had been an Azure-exclusive offering for years, a fact that gave Azure a concrete competitive advantage in AI infrastructure sales to large enterprises. That exclusivity is now gone. AWS Bedrock had already added Anthropic, Meta, Mistral, Cohere, and Stability AI as hosted model providers; OpenAI was the conspicuous gap in that catalog. The gap is closed.
OpenAI published a coordinated cookbook entry for Bedrock the same day, signaling that the AWS availability is a full distribution push and not a limited preview. The timing matters: enterprises evaluating inference providers before committing to a cloud contract now have OpenAI as a live option on both major hyperscalers simultaneously.
The announcement itself is thin on specifics. Pricing parity between the AWS integration and the direct OpenAI API is not confirmed in the release. Latency characteristics for AWS regions outside of Azure’s footprint are not disclosed. OpenAI stated the availability; an independent AWS-side confirmation through Bedrock’s standard provider announcement channel typically accompanies these launches, but the details of the technical integration remain limited to what the company released.
Codex, OpenAI’s code-execution model, is also included in the AWS availability, which extends the reach of the coding-agent workloads that enterprises have been testing against internal repositories. Routing Codex through AWS means audit logs and access controls can be applied at the infrastructure layer rather than managed separately in OpenAI’s dashboard.
The strategic read for the cloud market is that multi-cloud distribution is now table stakes for frontier-model vendors. OpenAI extending to AWS follows the same pattern Anthropic took when its Claude APIs landed on Bedrock and on Google Cloud Vertex AI. Distribution exclusivity, which had been Azure’s structural moat in AI, no longer constrains where the dominant models can be consumed.
Any enterprise currently running AI workloads on AWS that has held off on OpenAI integrations due to procurement or security concerns should revisit those evaluations this month; the primary objection that kept GPT models off AWS-native stacks no longer applies.
Source: OpenAI announcement published June 1, 2026, on openai.com.
