Microsoft gradually winds down AI computing rental agreement with Oracle because the hardware uses older-generation Nvidia chips
According to Bloomberg, citing people familiar with the matter, Microsoft is gradually ending its agreement to rent AI computing resources from Oracle. In recent months, after refusing to renew the contract, Microsoft has been progressively shutting down the related computing capacity because Oracle’s related hardware uses older-generation Nvidia chips and no longer meets Microsoft’s specifications.
This move is part of Microsoft’s broader reassessment of its multibillion-dollar cloud partnerships. Microsoft had previously committed more than $80 billion to rent computing capacity from cloud companies such as CoreWeave, Nscale and Nebius. The agreement was announced in 2023, and Oracle said at the time that it would provide servers to Microsoft to support the AI-assisted Bing search engine.
An Oracle spokesperson said that demand for its AI infrastructure, including early GPUs, remains strong. Microsoft declined to comment.
Original source:
Title: Microsoft Winding Down Oracle Deal to Rent AI Cloud Capacity Publisher: Bloomberg Publication date: 2026-11-06 URL: https://www.bloomberg.com/news/articles/2026-11-06/microsoft-is-winding-down-oracle-deal-to-rent-ai-cloud-capacity Nature of source: media report (citing people familiar with the matter) Verification status: This read included Bloomberg’s report and Reuters cross-check confirmation, verification date 2026-11-06 Supporting content: Supports all core facts in the body, the chip-related reason, the scale of the partnership, and both sides’ responses
ABAB AI Insight
Microsoft is gradually ending the AI computing rental agreement it reached with Oracle in 2023. People familiar with the matter said Microsoft has been progressively shutting down the related computing capacity in recent months after refusing to renew the contract because Oracle’s hardware uses older-generation Nvidia chips and no longer meets Microsoft’s specifications. The agreement was originally intended to support the AI-assisted Bing search engine.
On the capital side, Microsoft had already committed more than $80 billion to rent computing capacity from cloud companies such as CoreWeave, Nscale and Nebius. Oracle said its AI infrastructure, including early GPUs, still faces strong demand, and it said in September that renewal contracts for four-year-old GPUs carried a 20% premium. Microsoft declined to comment.
Compared with Microsoft’s simultaneous reassessment of other neocloud partnerships, such as tighter oversight of data center projects with IREN, Nebius and Nscale, this is a contract termination targeting Oracle. The similarity is that both reflect Microsoft’s reevaluation of external computing specifications and costs; the key difference is that the Oracle case explicitly points to a mismatch in chip generation, while the other projects more often involve delays and cost issues.
For Microsoft, ending computing leases that do not meet specifications can optimize its AI infrastructure mix. For Oracle, demand for early GPUs still exists, but losing Microsoft as a customer will reduce some leasing revenue. For the Nvidia chip market, the reduced availability of older-generation chips among some large customers may accelerate the shift to newer-generation hardware.
ABAB News · Cognitive Law
The agreement is being terminated because Oracle’s hardware uses older-generation Nvidia chips and no longer meets Microsoft’s specifications.
Microsoft has committed more than $80 billion to computing leases with CoreWeave, Nscale and Nebius, and this move is part of that broader reassessment.
Oracle said demand for early GPUs remains strong, Microsoft declined to comment, and neither side disclosed the agreement’s specific scale.