OpenAI's coding assistant Codex is now on Amazon Web Services, and for the first time, enterprise buyers can apply those charges to existing AWS cloud spending commitments. That sounds like a partnership announcement. It is also a quiet admission that the company which positioned itself as the intelligence layer could not bypass the cloud's procurement layer to get there.
The mechanism is new. OpenAI has sold through Azure for years, but Azure customers could not use existing cloud commitments to cover Codex charges. That was a licensing friction that effectively locked AWS and Google Cloud enterprise buyers out. That restriction vanished Sunday when Microsoft and OpenAI published amended partnership terms ending the exclusive license that had given Azure sole access to OpenAI's latest technology. Monday's Bedrock launch became legally possible because of that contract change. More than 4 million people already use Codex every week, OpenAI said. The procurement path just opened.
The financial architecture did not budge. OpenAI must spend at least $250 billion on Azure services by 2032, according to Reuters. That commitment does not shrink because OpenAI can now sell elsewhere. Azure wins the infrastructure layer regardless of where OpenAI's models actually run. OpenAI gains a broader sales floor, but the load-bearing structure of the relationship is unchanged.
For enterprise buyers, the question is whether this changes procurement math. Should they evaluate OpenAI on Bedrock against Anthropic, which has been there longer and without Microsoft's stake? D.A. Davidson analyst Gil Luria told Reuters that AWS and Google Cloud enterprise customers had been limited in their ability to integrate OpenAI products because of the exclusive relationship, and will now consider OpenAI alongside competitors. The path is open. Whether it shifts buying decisions is a different question.
OpenAI still pays Microsoft a revenue share through 2030, capped but not eliminated. Microsoft retains first shipping rights when it can support a capability. The Azure commitment is not a historical artifact. It is the load-bearing structure of the entire restructured relationship. OpenAI gained distribution freedom. It did not escape infrastructure gravity.
What the deal restructuring revealed is the shape of a trap that every AI lab eventually hits. The hyperscalers own the procurement relationship with the enterprises that buy AI services. OpenAI spent years trying to build its own distribution moat, and now it rents shelf space from AWS. The real question, who controls distribution when the models themselves become commoditized, has been answered by the structure of OpenAI's own contracts. OpenAI just answered it, at least for now.