OpenAI diversifies cloud compute partners

This reveals that compute has become too strategic for OpenAI to leave in any one partner's hands. The company is still deeply tied to Microsoft through a $250B Azure purchase commitment and a revenue share that runs through 2032, but the October 2025 reset removed Microsoft's gatekeeper role on new workloads and opened the door for OpenAI to split training and inference across Oracle, AWS, CoreWeave, and its own Stargate buildout.

• The practical change is not a breakup, but a shift from single landlord to portfolio buyer. Microsoft remains a massive supplier and economic partner, yet Oracle operates the Abilene Stargate site, AWS provides a separate large capacity lane, and OpenAI is adding more sites so one provider no longer controls model launch speed.
  1 sacra 2 sacra 3 sacra 4 sacra
• The stack is also getting unbundled. Crusoe builds and co owns infrastructure, Oracle runs and rents the compute, Nvidia fills racks, and OpenAI consumes the output. That is different from classic Azure dependence, where one hyperscaler handled most of the chain from data center to cloud contract.
  1 sacra 4 sacra 5 sacra
• This mirrors a broader market shift. CoreWeave grew into a major supplier on the back of a $10B Microsoft contract, and Crusoe is following a similar path through Stargate. Frontier labs are creating a new layer of AI specific infrastructure vendors that sit between chip makers and hyperscalers.
  4 sacra 5 sacra 6 sacra

Going forward, the winners in AI infrastructure will be the groups that can mix capital, power, chips, and cloud capacity fastest. OpenAI is moving toward a multi provider supply chain that should improve bargaining power, raise resilience, and make compute availability itself a core competitive advantage in model development and product margins.