Microsoft winding down Oracle AI rental: check the older-Nvidia switch-off, not an exit from AI cloud

On Nov. 6, 2026, Reuters reported that Microsoft is winding down an arrangement to rent artificial-intelligence computing from Oracle. Reuters presented this as a report from Bloomberg News, which cited people familiar with the matter. The story places the move inside a larger Microsoft reassessment of billions of dollars in cloud partnerships.

Shares of Oracle and Microsoft were flat in afternoon trading. Both stocks, Reuters said, had come under pressure this week after a Democratic midterm election sweep in the United States raised worries about tighter scrutiny of the AI industry's rapid expansion. What a reader can check is not "Microsoft is quitting AI cloud." It is that a 2023 overflow rental of older Nvidia capacity is being switched off after the contract was not renewed.

The hard facts below come from the Reuters-bylined text reprinted by Channel NewsAsia and Investing.com, which I checked on Nov. 8, 2026. The original Reuters page returned 401 in this environment.

Microsoft had been turning off rented computing capacity from Oracle in recent months after declining to renew the contract. That rented capacity became less valuable over time because it no longer met Microsoft's evolving needs. The report says the Oracle hardware used an older generation of Nvidia chips.

An Oracle spokesperson told Reuters: "Demand for OCI's AI infrastructure remains strong, including for earlier-generation GPUs." OCI means Oracle Cloud Infrastructure. Microsoft did not immediately respond to a Reuters request for comment.

AI cloud company IREN said on Thursday it had delayed some infrastructure deliveries after Microsoft requested changes to its network design. That pushed back IREN's 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. Earlier on Thursday, Bloomberg News reported that Microsoft was tightening oversight of some data-center projects run by newer AI cloud firms, or neoclouds, including IREN, Nebius and Nscale, amid delays and rising costs.

Oracle announced the multi-year deal with Microsoft in 2023, saying Microsoft would use OCI AI infrastructure to power the AI-assisted version of the Bing search engine. The financial terms were not disclosed. A same-day note by The Verge matches those points: Microsoft has been turning off rented Oracle capacity since declining to renew a 2023 deal, and sources said the computing Oracle offered used older Nvidia chips and had become less useful.

This is not a statement issued by Microsoft. It is Reuters relaying Bloomberg, and Bloomberg relaying unnamed people familiar with the matter. The strength of that chain depends on whether later documents or company confirmations support it. What can already be checked is narrower. Oracle was willing to tell Reuters that demand remains strong, including for earlier-generation GPUs. Microsoft gave no matching comment at the time. IREN itself disclosed a delivery delay after Microsoft asked for network-design changes.

"Winding down" and "turning off rented capacity in recent months" should be read together. The cleaner picture is that the contract was not renewed and leftover rental capacity is being switched off. That is not the same as Microsoft announcing an exit from every cloud relationship with Oracle. The 2023 deal's public purpose was to power an AI-assisted Bing. Because the financial terms were never disclosed, this report cannot tell a reader how large the contract was.

The strongest steelman is that this can be ordinary hardware turnover. Not renewing a rental of older chips is a normal cloud-buying decision. It does not have to mean demand has collapsed. Oracle's on-the-record reply supports that reading: even for earlier-generation GPUs, the company says demand remains strong. The IREN, Nebius and Nscale line is also steadier if it is read as Microsoft rechecking how each watt of computing is bought, not as proof that the AI cloud business is shutting down.

The risks of that steelman are plain. Anonymous sources, Microsoft's silence, and the phrase "billions of dollars in cloud partnerships" cannot be unpacked from the Reuters reprint into an auditable list of contracts. "Became less valuable" and "no longer met evolving needs" are reported motives, not a technical explanation signed by Microsoft. IREN's delay of more than $4 billion in annualized revenue capacity is IREN's own timetable. It is not direct evidence that Microsoft's cloud revenue is falling.

Over the next few weeks, the checks are these: whether Oracle publishes capacity or customer details that match this report; whether Microsoft confirms or corrects the account that it declined to renew and is switching off older rented capacity; whether IREN's more than $4 billion in annualized revenue capacity comes online in January or slips toward late February; and whether Nebius and Nscale projects show similar tighter oversight or delays.

Readers should test whether the contract was renewed, which generation of chips is being switched off, and what Oracle said in public. They should not test a shortcut headline that Microsoft is giving up on AI cloud.

What this report makes testable is that Microsoft is switching off older-Nvidia OCI capacity rented in 2023 after choosing not to renew, while separately tightening oversight of projects run by newer AI cloud firms such as IREN, Nebius and Nscale. Not that Microsoft is leaving AI cloud computing. Hardware turnover can explain the first half. The second half has to be checked against those firms' own delivery calendars.

【No.2】