Every week an enterprise AI pilot dies in procurement, not because the technology failed, but because legal won’t approve another vendor portal.

The bottleneck was never the model, it was the contract

Enterprise teams evaluating OpenAI models have been forced to manage a separate vendor relationship, separate billing, and separate compliance reviews outside their existing cloud agreements. That friction alone has killed more pilots than any technical limitation.

OpenAI’s best models now live inside the AWS console

OpenAI frontier models and Codex are now available on AWS, giving engineering and procurement teams access to frontier models including Codex directly through AWS Bedrock, using the IAM controls, VPC configurations, and consolidated billing they already have in place. You open the AWS console, select the model through Bedrock’s API, and your existing cloud security posture applies from the first call. The output is full access to OpenAI’s production-grade models without a separate OpenAI account or contract.

The teams feeling this most are the ones who stopped waiting

• Enterprise architects who have been holding OpenAI evaluations until security sign-off finally cleared AWS-native deployments
• Software engineering leads who need Codex for internal tooling but could not get a standalone OpenAI API agreement through legal in time
• Cloud procurement officers who manage a single AWS Enterprise Discount Program agreement and refuse to add net-new SaaS vendors mid-year

The teams listed above were not blocked by capability gaps. They were blocked by process.

AWS just became the fastest path from OpenAI evaluation to production

AWS already hosts more than a third of the world’s cloud workloads, and Anthropic’s Claude models have been available on Bedrock for over a year, making it the reference point enterprises compare against when choosing where to run foundation models. OpenAI’s arrival on the same platform means the build-versus-buy conversation inside large organizations just shifted significantly toward build.

What you can do with it starting today

• Run OpenAI frontier models through AWS Bedrock API calls inside existing applications
• Deploy Codex for internal code generation without a separate OpenAI contract
• Apply existing AWS IAM roles and VPC policies to all OpenAI model requests
• Consolidate OpenAI usage into AWS billing for unified cost reporting

Pricing follows AWS Bedrock consumption rates, billed through your existing AWS account.

The honest tradeoff: teams that need OpenAI’s newest experimental models or API features the moment they ship will still hit a lag, since Bedrock availability trails direct OpenAI API releases.

If your team is already on Azure, Microsoft’s OpenAI Service has offered this integration longer and with deeper Active Directory support. Google Cloud’s Vertex AI is the comparable play for GCP shops running Gemini alongside third-party models.

The cloud platform that hosts your models is becoming the new vendor decision

This shift is moving faster than most procurement cycles can track, and the tooling choices made in the next six months will set enterprise AI infrastructure for years. We cover tools like this every Friday — subscribe here and we’ll send the best ones straight to you.