Reports say Microsoft is winding down its arrangement to rent artificial intelligence computing capacity from Oracle, letting a contract signed in 2023 lapse rather than renewing it. The rented servers, which once supplied extra power for Microsoft’s AI-assisted Bing search engine, run on older-generation Nvidia chips that no longer meet the company’s changing technical requirements. The move is part of a broader reassessment of billions of dollars in cloud partnerships, and Oracle shares fell nearly 3% in afternoon trading on Friday, after an earlier fall of about 4%.
The deal that is ending
Microsoft has been switching off the rented Oracle capacity in recent months, Bloomberg reported, after deciding not to renew the agreement. Over time, the capacity became less valuable as Microsoft’s specifications changed, and the older hardware could no longer keep up with its AI workloads. The full size of the contract has not been disclosed, and Microsoft declined to comment.
The arrangement dates to 2023, when Oracle agreed to supply servers to support Microsoft’s AI-powered Bing search engine. At the time, Microsoft needed extra computing power for its new AI-assisted search, and renting from Oracle was faster than building new data centers. Three years on, that shortcut has reached the end of its useful life.
Oracle pushes back
Oracle responded by arguing that demand for its AI infrastructure remains strong, including for earlier-generation graphics processors. In a statement on X, the company pointed to its latest earnings call, saying 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.
The response did not name Microsoft, identify which customers took the freed capacity, or say what the departing contract was worth. Oracle also noted that most of the graphics processors covered by those renewals were at least four years old, presenting the premium pricing as evidence that even ageing hardware still finds buyers as newer chips reach the market.
Part of a wider review
The Oracle exit is part of a wider reckoning. According to Bloomberg, Microsoft has committed more than $80 billion to renting computing capacity from cloud providers, including CoreWeave, Nscale and Nebius, for its AI operations. Executives have grown frustrated with high costs and delivery delays at suppliers, and the company has tightened oversight of data center construction while reconsidering agreements.
The pattern has been visible all week. Microsoft declined an option to expand its Texas agreement with Nscale, leaving its existing contract untouched, and supplier IREN said Microsoft-requested network changes had pushed back infrastructure deliveries. IREN moved its target for more than $4 billion in annualised revenue capacity from December into early 2027.
IREN’s shares closed nearly 17% lower on 5 November after the delay, a reminder that the suppliers caught in Microsoft’s review are feeling the consequences first. Its AI cloud revenue reached $200 million in the September quarter, while its Bitcoin mining revenue fell 58% from the previous quarter to $28 million.
Microsoft has said its infrastructure strategy combines company-owned data centers, leased facilities and outside partnerships, giving it flexibility in bringing computing capacity online reliably and efficiently. The Oracle decision fits that framing: a rented fleet that no longer matches the workload is dropped, while the spending itself continues elsewhere.
Our Read
This is not a pullback in AI spending. Microsoft has only trimmed its forecast as capacity delivery slips, and the Oracle exit is about where the money goes, not how much of it there is. A 2023 rental built for Bing-era workloads, running on ageing chips, is exactly the kind of capacity a disciplined buyer drops first.
For Oracle, the optics are awkward even if the economics are manageable. Its own defence — that the freed capacity was resold at a 20% premium — concedes the hardware was old enough to be four years into its life, and losing a tenant of Microsoft’s size still undercuts the story that its AI cloud can hold the biggest customers.
The June 2024 partnership that tied Oracle to Microsoft and OpenAI remains intact, but the relationship is now visibly a negotiation rather than a blank cheque. Other suppliers should read the Oracle exit the same way Microsoft’s other partners already have: as a warning that legacy capacity will not be renewed on loyalty alone.
Why It Matters
The episode matters because it shows how quickly rented AI capacity can become stranded. Chips bought for one generation of workloads lose value the moment a customer’s needs move on, and a supplier that built its AI narrative on hyperscaler demand now has to prove it can replace a departing Microsoft with new long-term customers.
For the wider market, it is another sign that the AI infrastructure boom is entering a sorting phase, where delivery and price matter as much as headline capacity. Investors punished Oracle’s shares on the news, while Microsoft’s barely moved — the market already treats this as Oracle’s problem, not Microsoft’s.
Background
Microsoft and Oracle have worked together on AI before: in June 2024 Oracle announced a partnership with Microsoft and OpenAI to supply additional capacity through Oracle Cloud Infrastructure, extending Azure’s AI platform to support OpenAI’s growth. The Bing rental that is now ending predates that deal, and the two companies’ wider relationship continues.
Source: Bloomberg’s report, via Reuters
0 Comments