As of 1 June 2026, AWS customers can run GPT-5.5, GPT-5.4, and Codex directly inside Amazon Bedrock, billed at OpenAI's standard per-token rates with no additional cloud markup. The AWS Machine Learning Blog states that more than five million people use Codex every week (vendor-stated figure; a companion About Amazon article puts the count at more than four million). Either way, that user base now has a path to deploy Codex inside an AWS governance envelope.

What Went Live on 1 June

Amazon Web Services flipped the general availability switch for three OpenAI models on Bedrock: GPT-5.5, positioned as OpenAI's most capable frontier model for demanding agentic and reasoning workloads; GPT-5.4, aimed at price-conscious production use cases; and Codex, the coding agent available through the Codex API, CLI, and IDE plugins for Visual Studio Code, JetBrains, and Xcode. All three sit alongside Anthropic, Meta, Mistral, and Amazon's own models in Bedrock's unified API — the same call pattern, the same billing console.

Pricing matches OpenAI's first-party rates with no additional AWS fees. Usage also counts toward existing AWS cloud spend commitments, which matters for enterprises already managing large cloud contracts.

The $38 Billion Infrastructure Bet Behind It

The GA launch is the visible edge of a larger deal. AWS and OpenAI signed a multi-year infrastructure contract under which AWS committed US$38 billion — hundreds of thousands of NVIDIA GPUs — to power OpenAI's workloads over seven years. Amazon made the expanded Bedrock integration public on 28 April 2026; the 1 June date marks the move from limited preview to full production availability.

The deal arrived weeks after OpenAI formally restructured its relationship with Microsoft. On 27 April 2026, OpenAI and Microsoft amended their agreement: Azure retains a non-exclusive licence to OpenAI's models and intellectual property through 2032 and first-deployment rights for new products, but OpenAI can now serve its full product catalogue across any cloud provider. According to TechCrunch's contemporaneous reporting on the April 2026 amendment, Microsoft remains the "primary cloud partner" while the unilateral exclusivity that had defined the arrangement is gone. (Separately, Amazon announced an up-to-$50 billion equity investment in OpenAI in February 2026 — a distinct instrument from the $38 billion infrastructure contract.) AWS is the first hyperscaler to benefit from that opening.

Governance Controls Enterprises Already Know

Every API call to OpenAI models on Bedrock inherits the AWS governance stack: IAM permissions, VPC and PrivateLink network isolation, KMS encryption, and CloudTrail audit logging. AWS states that customer prompts and responses are not used to train models — a point enterprises in regulated industries will scrutinise closely before production deployment, since this is an AWS assurance, not an independently audited certification.

For security teams, the practical upshot is that OpenAI's models can now sit behind the same access policies and audit trails already applied to internal AWS workloads, rather than requiring a separate OpenAI API key management layer.

What's Coming: Managed Agents

One addition is in the pipeline. According to the AWS Machine Learning Blog, Bedrock Managed Agents — an agentic orchestration layer powered by OpenAI's agent framework, with isolated identity and full action logging — is coming shortly. It does not yet carry a firm release date.

What This Changes for the Cloud AI Market

AWS holds the largest share of global enterprise cloud infrastructure by revenue. Adding OpenAI's frontier models to Bedrock gives those customers a single pane of glass across the major model families — without forcing a workload migration to Azure or a separate OpenAI API contract. That is a direct competitive challenge to Microsoft, which built a substantial enterprise AI business around the Azure OpenAI Service as the historically sole distribution channel.

The $38 billion infrastructure commitment also confirms that AI compute has become a primary revenue driver for cloud providers, not a peripheral add-on. Both AWS and Google Cloud have now made nine- to ten-figure bets on AI model access, and Microsoft's response — if any — will likely define the next phase of the hyperscaler rivalry.