A shortage contract, three years on

In November 2023, Microsoft announced that it would use Oracle Cloud Infrastructure alongside Azure to run inference for Bing conversational search. The multi-year agreement was notable: Microsoft was using a direct cloud competitor to secure additional AI computing capacity. The financial terms were not disclosed.

Bloomberg reported on November 6 that the arrangement is now being wound down. Microsoft has been turning off the rented capacity in recent months after declining to renew the contract, according to people familiar with the matter. The capacity became less useful over time because it no longer fits the company's specifications. The Oracle hardware used an older generation of Nvidia chips. Bloomberg did not name the chip model, and it could not determine the full size of the contract.

Read the two gaps in that report before reading anything into it. Without a chip model, a reader cannot tell how many generations behind the hardware is. Without a contract size, a reader cannot tell whether this is a rounding error in Microsoft's capacity book or a material loss for Oracle. What the report does establish is narrower and still useful: the reason given is specification, not demand.

Oracle's answer is a price, not a denial

Oracle did not dispute that Microsoft is leaving. Its spokesperson told Reuters that demand for OCI's AI infrastructure remains strong, including for earlier-generation GPUs. In the fuller statement reported by Bloomberg, the company pointed to its latest earnings call: GPU capacity that came up for renewal in its fiscal first quarter was renewed or resold to other customers under long-term contracts at prices 20% above the previous agreements, and most of those GPUs were at least four years old.

Those two accounts are not in conflict. They describe a market splitting by workload. A buyer training or serving frontier models at Microsoft's scale can reject a four-year-old GPU on performance per watt alone. A buyer running inference for a narrower product can take the same hardware, at a premium to its last contract, and still be ahead of waiting for new machines. The hardware did not change. The buyer's specification did.

Part of a wider re-sorting, not a stated cut

Reuters framed the move as part of a larger reassessment of billions of dollars in cloud partnerships. A book that size is not one decision. It is a portfolio that gets re-sorted whenever chip generations, delivery dates or network designs move.

A separate development involving Microsoft and IREN illustrates another form of supplier risk. This time, the issue was network design and delivery timing, not GPU generation. IREN, one of those outside providers, said on Thursday that some of its infrastructure deliveries for Microsoft were delayed after Microsoft requested changes to its network design, pushing its target for bringing more than $4 billion in annualized revenue capacity online from year-end to January, with a possible further delay until late February. Bloomberg also reported on Thursday that Microsoft is tightening oversight of some data center projects run by neocloud providers including IREN, Nebius and Nscale, amid delays and rising costs.

Put together, the week's reporting does not show Microsoft spending less on AI compute. No source in it states a cut, and Microsoft declined to comment. It shows a large renter applying its current specifications more strictly: non-renewal of Oracle capacity running older-generation chips, a changed network design at IREN, and closer oversight of neocloud projects. Suppliers holding older hardware or late deliveries are the ones exposed to that discipline. Suppliers with current-generation machines to rent are, on Oracle's own account, charging more for them than they did four years ago.

Sources

  1. [1] Reuters, "Microsoft is winding down Oracle deal to rent AI cloud capacity, Bloomberg News reports", November 6, 2026, updated 19:24 UTC (read in full via syndication: the wind-down and non-renewal, the older-generation Nvidia chips, the Oracle spokesperson statement, the IREN delivery delay and the more-than-$4 billion annualized revenue capacity target moving from year-end to January with a possible further delay to late February, Bloomberg's Thursday report on tightened neocloud oversight, the 2023 Bing deal and Microsoft's non-response)Read source
  2. [2] Bloomberg Law, "Microsoft Winding Down Oracle Deal to Rent AI Cloud Capacity (2)", November 6, 2026 (Bloomberg's own text of the report: Microsoft turning off rented Oracle capacity in recent months after declining to renew, the capacity no longer fitting the company's specifications, and the Oracle hardware using older-generation Nvidia chips, per people familiar with the matter)Read source
  3. [3] Oracle, announcement of a multi-year agreement with Microsoft, November 7, 2023 (Microsoft using Oracle Cloud Infrastructure AI infrastructure, along with Microsoft Azure AI infrastructure, for inferencing of AI models optimized to power Microsoft Bing conversational searches; carried by BigDATAwire)Read source
  4. [4] Oracle, first-quarter fiscal 2027 earnings call, September 10, 2026 (co-CEO Clay Magouyrk: of all the GPUs that came up for renewal in the quarter, the capacity was renewed or resold at a 20% premium to prior contracts, and the majority of those GPUs were four years or older; full transcript)Read source
  5. [5] Oracle, statement of November 6, 2026, as carried by crypto.news, "Microsoft winds down Oracle AI cloud capacity agreement" (the fuller spokesperson statement pointing to the latest earnings call: capacity that came up for renewal in the fiscal first quarter was renewed or resold at a 20% premium to prior pricing under long-term contracts, most of the GPUs at least four years old)Read source