Microsoft is switching off AI capacity it has been renting from Oracle since 2023, after declining to renew the contract, Bloomberg reported Friday.
The arrangement supplied servers behind Microsoft's AI-powered Bing search, and the hardware involved used older-generation Nvidia chips that no longer meet Microsoft's specifications, according to the report. Microsoft has been turning the capacity off over recent months rather than waiting for a hard end date. The company declined to comment; Bloomberg could not establish the contract's full value.
Oracle pushed back on the framing. In a statement on X, the company said demand for OCI's AI infrastructure remains strong, including for earlier-generation GPUs, and that GPU capacity up for renewal in its fiscal first quarter was either renewed or resold at a 20% premium to prior pricing on long-term contracts. Oracle shares fell about 4% on the report.
The detail that matters is not that a contract lapsed, but why: the GPU generation aged out of the buyer's requirements. Microsoft's inference and training stacks have moved on, and rented capacity on older silicon shifts from hedge to liability. That is a different story from a company deciding it has too much compute.
It also sits inside a larger reshuffle. Bloomberg described the move as part of a reassessment of billions of dollars in cloud partnerships; Microsoft has committed more than $80 billion to renting capacity from providers including CoreWeave, Nscale and Nebius, the report said. Microsoft told Reuters its infrastructure strategy combines its own data centers, leased facilities and outside partnerships.
For buyers of AI compute, the takeaway is about duration. Multi-year capacity contracts signed in 2023 and 2024 were priced against a hardware roadmap that has since been replaced twice. The winners in the next round of deals may be providers who can re-let aging capacity to customers whose workloads — batch inference, fine-tuning, simulation — do not need the newest accelerators. Oracle's claim of 20% premiums on renewed capacity is exactly that argument, made publicly and on the record.
What to watch: whether other Microsoft leases on older GPU generations are wound down too, and how much earlier-generation capacity returns to the spot and contract markets over the next two quarters. If it clears quickly at premium pricing, the compute shortage is real. If it sits, the industry has a generation-mismatch problem rather than a shortage.
