Microsoft winds down AI cloud capacity rented from Oracle
Hardware & InfraNews

Microsoft winds down AI cloud capacity rented from Oracle

Microsoft is turning off AI compute it rented from Oracle after declining to renew a contract on older Nvidia chips, Bloomberg reported, as the company reassesses billions in cloud partnerships.

Tall server racks with blue indicator lights lining a data center aisle
Photo by Brett Sayles on Pexels

Part of our AI chips and memory explained: GPUs, TPUs and HBM guide

Microsoft is winding down an arrangement to rent AI computing from Oracle after declining to renew the contract, Bloomberg reported on Friday, citing people familiar with the matter. In recent months the company has been turning that rented capacity off. The reason those sources gave is hardware, not a public cut to AI spend: over time the machines stopped fitting Microsoft’s specifications because Oracle’s gear used older-generation Nvidia chips. Bloomberg could not determine the full size of the deal.

Reuters, summarizing the same Bloomberg report, said Oracle shares fell nearly 3% in afternoon trading while Microsoft was little changed. That move landed in the same week as Thursday’s chip-stock selloff, when investors already had AI infrastructure spending under a microscope. A nearly 3% session move and a multi-year rental contract are different facts; neither is a forecast of Oracle’s future AI revenue.

The rental itself dates to 2023, when Oracle agreed to supply servers for a new AI-assisted version of Bing, according to Investing.com’s write-up of Bloomberg. Renting from a cloud competitor was unusual then, and it was an early sign of how tight GPU supply had become. Bloomberg’s sources framed Friday’s unwind as part of a larger reassessment of billions of dollars in cloud partnerships, not as a claim that AI demand has disappeared.

Oracle pushed back on the demand question. An Oracle spokesperson told Bloomberg that demand for its AI infrastructure remains strong, including for earlier-generation GPUs, Investing.com reported. The company pointed to fiscal first-quarter renewals in which GPU capacity that came up for renewal was renewed or resold at a 20% premium to prior long-term pricing, even though most of those GPUs were at least four years old. That 20% figure describes Oracle’s broader renewal book; it is not the value of the Microsoft contract Microsoft is leaving, which Bloomberg said it could not size.

Microsoft declined to comment to Bloomberg. Separately on Friday, CNBC reported that Microsoft is evaluating a smaller roster of neocloud partners while it builds more of its own footprint, and that two people said Microsoft opted not to renew with Oracle. A Microsoft spokesperson told CNBC its strategy mixes owned data centers, leased facilities, and third-party providers and is “built on flexibility and choice.”

The story fits a wider shift from renting whatever GPU capacity was available toward matching chip generations to the workload. Older Nvidia silicon can still run many jobs, which is why Oracle stresses residual demand, but newer training and serving stacks often want the memory bandwidth covered in our HBM explainer. Friday’s report is about one rental winding down on older hardware, not a scorecard of every Microsoft cloud commitment.

Sources

  1. Bloomberg Law, Microsoft Winding Down Oracle Deal to Rent AI Cloud Capacity (Nov. 6, 2026)news.bloomberglaw.com
  2. Reuters via Investing.com, Microsoft is winding down Oracle deal to rent AI cloud capacity, Bloomberg News reports (Nov. 6, 2026)investing.com
  3. Investing.com, Oracle falls 4% as Microsoft reportedly winds down AI cloud capacity deal (Nov. 6, 2026)investing.com
  4. CNBC, Microsoft plans to use fewer neoclouds as AI buildout shows signs of strain (Nov. 6, 2026)cnbc.com

Topics

Kelly B.Hardware & Infra desk

Kelly B. is a Codema.ai house byline: a pen name used by our editorial desk, not one individual journalist. It covers Hardware & Infra: GPUs and the memory inside them (HBM, GDDR7, DDR5, NAND), hardware for running AI locally, GPU rental prices, chipmaking, and the data centres and power behind AI. Pieces under this name are written with AI assistance and reviewed by a human editor before they're published.

All stories by Kelly B.

More in Hardware & Infra

See all
  1. Blacklisted Chinese university bought two Nvidia H200s

    Derrick K.

  2. SoftBank Grab and PETROS plan $1.85B Malaysia AI build

    Kelly B.

  3. Nvidia, Rebellions and SK Telecom agree NVLink Fusion chip deal

    Kelly B.