Copy article

Enterprise AI Procurement Faces New Questions After OpenAI Ends Seven-Year Microsoft Azure Exclusivity

ended 05. May 2026

OpenAI just proved that even the company with the deepest cloud partnership in tech history couldn't afford to stay locked in.

On 27 April, Microsoft and OpenAI gutted their exclusive arrangement. Within 24 hours, OpenAI's models were live on Amazon Bedrock. 

A leaked internal memo from OpenAI's revenue chief, Denise Dresser, said it plainly: the Microsoft partnership had "limited our ability to meet enterprises where they are." 

The trigger was Anthropic. Its run rate hit $30 billion in April, up from $9 billion at end of 2025, with over 1,000 enterprise clients spending more than $1 million a year. That growth happened largely through AWS Bedrock, where enterprise buyers could access Claude without leaving the cloud they already operated in. 

OpenAI was locked out of that market by its own exclusivity deal. 

Microsoft traded its exclusive licence for something quieter: it stops paying OpenAI revenue share, keeps collecting 20% of OpenAI's revenue through 2030, retains 27% equity, and holds non-exclusive IP rights to 2032. 

The practical effect of the AWS move is that enterprise customers can now access OpenAI's models, including GPT-5.5, Codex, and a new managed agents platform, through the same Bedrock APIs, security controls, and procurement processes they already use. Usage counts against existing AWS cloud commitments. 

For organisations already running production workloads on AWS, that removes the need to set up a separate Azure relationship, negotiate a second set of contracts, or move data between clouds just to use OpenAI. 

It also means OpenAI is now directly competing with Anthropic on the same platform, through the same buying mechanism, for the same enterprise budgets.

We'd like your views:

  • If OpenAI's own revenue chief says exclusivity with a $13 billion backer held them back, what does that tell smaller businesses negotiating single-provider AI contracts with far less leverage?
  • Anthropic grew from $9 billion to $30 billion run rate in four months, largely through multi-cloud availability. Should enterprise AI procurement now treat single-cloud deployment as a risk factor rather than a simplification?
  • Microsoft keeps 27% of OpenAI and a 20% revenue share through 2030, but lost exclusivity. Did they sell strategic control for accounting convenience, or is equity the smarter long-term position when cloud margins are compressing?
  • OpenAI committed to 2GW of Amazon's Trainium chips and extended its AWS contract by $100 billion over eight years. When the company that built its reputation on Nvidia GPUs bets this heavily on custom silicon it hasn't proven at scale, who carries the risk if performance doesn't hold?

2 responses from the Newspage community

Copy all

Star Quote
Copy

Colette Mason
AI Ethics Consultant at Clever Clogs AI
The real story here isn't a corporate breakup. It's a live demonstration that vendor lock-in eventually costs more than the subsidy that created it. Every business running its AI stack through a single cloud provider just watched the most compute-rich company on earth conclude it couldn't afford to.

If it took OpenAI seven years to break free of that deal. Businesses need to ask themselves how long is their cloud-based AI contract locking them in for.
Copy

Rohit Parmar-Mistry
Founder at Pattrn Data
Enterprise AI procurement is entering its awkward grown-up phase. The hard question is no longer which model looks most impressive in a demo, but who carries the operational risk when vendors, clouds and contracts shift underneath the business.

In AI Audits, this is where we often find the weakest evidence. Firms can name the supplier, but not the owner of the decision, the data trail, the escalation route, or the fallback plan if the system gives a confident but wrong answer.

Boards should stop treating AI procurement as a software shopping exercise. It is governance, accountability and resilience work. If a business cannot explain how the system is controlled after the contract is signed, it has bought dependency, not capability. The clever move is to buy less hype and demand more proof.