AWS launches OpenAI models after Microsoft drops Azure exclusivity

OpenAI has sharply widened the channels through which it sells its models: on 28 April 2026, AWS announced that it will offer them to customers through Amazon Bedrock. The news came a day after Microsoft and OpenAI officially rewrote their partnership terms and removed Azure's previous exclusivity on reselling OpenAI technology.

What changed

Until this week, enterprise customers who wanted OpenAI models inside the cloud infrastructure they already used effectively had to go to Azure. On 27 April, Microsoft said its license to OpenAI's intellectual property remains in force until 2032, but is now non-exclusive. Microsoft keeps its status as primary cloud partner, and OpenAI products must still launch first on Azure, unless Microsoft cannot or chooses not to support the required capabilities.

In practice this is an important shift: OpenAI can sell its models not only through its own services and Azure, but also through other clouds. For AWS, this fills one of the main gaps in the Bedrock lineup. For OpenAI, it opens the door to companies that have already built their processes, security, procurement and compliance around Amazon Web Services and do not want to move everything for the sake of a single model provider.

The old AGI clause is also gone from the deal. It tied Microsoft's commercial rights to the hypothetical achievement of artificial general intelligence. In real business it looked too vague and got in the way of long, predictable agreements. The new contract makes the relationship between the two companies less ideological and far more market-driven: Microsoft remains a major shareholder and partner, while OpenAI gets freedom in distribution and room for new alliances in the future.

What arrived on AWS

AWS launched access to OpenAI products in limited preview inside Amazon Bedrock. According to the company, some of the current models became available for testing right on 28 April, while the most powerful GPT versions are due in the coming weeks. It is not only about the models themselves, but also a broader set of services meant to make OpenAI workflows convenient for large teams already living inside the AWS ecosystem.

• The latest OpenAI models in Amazon Bedrock
• Codex for use in the CLI, the desktop app and VS Code
• Bedrock Managed Agents powered by OpenAI
• Access through AWS accounts, IAM and PrivateLink
• Cost tracking within existing AWS cloud contracts

AWS says OpenAI models in Bedrock inherit the mechanisms enterprise customers are used to: encryption, logging through CloudTrail, guardrails and network isolation. There is a particular emphasis on agents: back in February, OpenAI and Amazon announced joint development of the Stateful Runtime Environment, a setting where an agent can keep context, tool state and memory between steps instead of working in an "one request, one response" mode.

"Our customers have been asking for this for a very long time," said AWS chief

Matt Garman, commenting on the launch.

For the market, this is not just one more way to call an API. AWS is trying to show that it can sell not a set of disconnected models but a ready platform for production scenarios, with security, orchestration, agents and billing in a familiar perimeter. That is exactly what large companies need if they already use Bedrock and want to add OpenAI without migrating to Azure, changing procurement processes and rebuilding compliance requirements.

Where the risk lies for OpenAI

The AWS expansion looks like a strong commercial move, but it coincided with an awkward moment for OpenAI itself. According to The Wall Street Journal, in 2026 the company missed some internal targets for revenue and user growth, and ChatGPT did not reach its internal goal of 1 billion weekly active users by the end of 2025. OpenAI and its leadership publicly dispute that reading, but the fact that the debate exists matters: the market has started to look not only at model quality but also at the economics of scaling them.

The problem is that OpenAI is simultaneously taking on gigantic infrastructure commitments. The company earlier announced a multi-year partnership with Amazon, an Amazon investment of $50 billion, and a further expansion of compute purchases from AWS by $100 billion over eight years. Microsoft, meanwhile, remains a key Azure partner. The wider the distribution, the more potential sales channels OpenAI has, but the sharper the question of whether demand can pay back that volume of compute.

That is why the main point of the news is not that OpenAI models are "now everywhere." What matters more is that multi-cloud is no longer theory but a working OpenAI strategy. The next test is to turn this availability into steady enterprise revenue and to prove that the new sales channels actually speed up the business rather than merely widening reach on paper, while Anthropic, Google and Microsoft itself strengthen their positions at the same time.

What this means

For the enterprise AI market, this is the end of an era in which access to OpenAI was a strategic advantage for Azure. Competition now shifts from exclusivity to integration quality, security and ease of deployment: the winner will not be whoever simply has an OpenAI model, but whoever embeds it better into real workflows and turns models into a working business product faster in the coming years.