Microsoft has begun winding down an agreement to rent AI computing capacity from Oracle after deciding not to renew a contract involving older Nvidia chips that no longer met its requirements.
Summary
- Microsoft has been switching off rented Oracle computing capacity in recent months, according to Bloomberg.
- Oracle says older GPU capacity was renewed or resold at a 20% premium.
- Microsoft committed more than $80 billion to computing capacity from outside cloud providers.
- The pullback follows growing concerns inside Microsoft over suppliers’ costs and delivery delays.
In a Nov. 6 report citing people familiar with the matter, Bloomberg said Microsoft had been turning off Oracle capacity after declining to renew the agreement. The rented computing resources had become less useful as Microsoft’s specifications changed, the people said.
The arrangement dates to 2023, when Oracle agreed to supply servers supporting Microsoft’s AI-powered Bing search engine. Bloomberg could not determine the full size of the contract, and Microsoft declined to comment.
Oracle defends demand for older AI chips
Oracle responded by saying demand for its AI infrastructure remained strong, including for earlier generations of graphics processing units.
In a Nov. 6 statement, the company pointed to comments from its latest earnings call about capacity that became available for renewal during its fiscal first quarter. Oracle said the capacity was either renewed or sold to other customers under long-term contracts at prices 20% above the previous agreements.
Most of the GPUs covered by those renewals were at least four years old, according to the company. Oracle presented the pricing as evidence that its existing hardware continued to attract customers even as newer chips entered the market.
“Demand for OCI’s AI infrastructure remains strong, including for earlier-generation GPUs,” Oracle said.
The company’s response addressed demand across its infrastructure business without identifying the value of Microsoft’s departing contract or saying which customers had taken the capacity Microsoft was leaving.
Bloomberg described the equipment involved as older-generation Nvidia chips. Its report did not identify a specific chip model.
Microsoft reviews more than $80 billion in cloud commitments
Microsoft’s retreat from the Oracle arrangement follows an expensive expansion into computing supplied by outside companies.
According to Bloomberg, Microsoft committed more than $80 billion to renting capacity from cloud businesses, including CoreWeave, Nscale and Nebius, to support its AI operations. Several executives have since grown frustrated with high costs and delays at suppliers.
The company has tightened oversight of data center construction, withdrawn from projects it considers risky and discussed whether to sign further agreements with these providers, the publication reported.
In separate reporting published Nov. 5, Bloomberg said Microsoft had sent employees to supplier sites to check construction standards and compliance. Projects receiving closer attention included facilities being developed by Nebius, IREN and Nscale.
Microsoft told Reuters its infrastructure strategy combined company-owned data centers, leased facilities and outside partnerships. A spokesperson said the approach provided flexibility in bringing computing capacity online reliably and efficiently.
The company has declined an option to expand its existing Texas agreement with Nscale. Reuters reported that the decision did not affect the current contract or Nscale’s disclosed backlog.
Bitcoin miner IREN faces Microsoft delivery changes
The review of outside capacity comes as Bitcoin miners spend heavily to build AI infrastructure for large technology customers.
As crypto.news previously reported on Nov. 6, IREN’s Microsoft deployment faced delays following customer-requested changes to network infrastructure. The company moved its target for more than $4 billion in annual recurring revenue from December into early 2027.
IREN’s AI cloud revenue reached $200 million in the September quarter, while Bitcoin mining revenue fell 58% from the previous quarter to $28 million. Its shares closed nearly 17% lower on Nov. 5.
Financing terms have varied across miners’ AI projects. An October report on AI infrastructure borrowing costs cited BlocksBridge Consulting’s assessment of separate IREN financing packages carrying rates of 6% and 9%.
The consultancy cautioned that the packages funded different projects and carried different exposures, making a direct comparison misleading.
Oracle and Microsoft built several cloud partnerships
Beyond the Bing arrangement, Oracle and Microsoft have announced other collaborations involving AI computing.
In June 2024, Oracle announced a partnership with Microsoft and OpenAI to provide additional capacity through Oracle Cloud Infrastructure. The companies said the arrangement would extend Microsoft’s Azure AI platform to support OpenAI’s growth.
Oracle’s announcement named OpenAI among customers using its infrastructure for AI workloads, alongside companies including Nvidia, Together AI, and xAI.
