Microsoft is winding down a multiyear arrangement to rent artificial intelligence computing capacity from Oracle, according to people familiar with the matter. The company has been turning off the rented servers in recent months after declining to renew the contract, as the hardware no longer aligned with its updated technical requirements.

The original agreement, announced in 2023, called for Oracle to supply servers that supported Microsoft’s AI-powered Bing search engine. The full value of the contract has not been disclosed. Microsoft declined to comment on the development.

Older Nvidia Chips No Longer Fit Evolving Needs

The Oracle equipment relied on older-generation Nvidia graphics processing units. Over time, those chips became less useful for Microsoft as the company’s specifications for AI workloads changed. The shift illustrates how rapidly customer requirements can evolve in the AI infrastructure market, even while the underlying hardware remains operational.

Oracle responded by stating that demand for its Oracle Cloud Infrastructure AI offerings continues to be robust, including for earlier-generation GPUs. In a statement, the company noted that capacity coming up for renewal in its fiscal first quarter was either renewed or resold to other customers under long-term contracts at prices about 20 percent above prior agreements. Most of those GPUs were at least four years old.

Oracle’s comments addressed overall market demand and did not specify which customers acquired the capacity Microsoft is leaving or the precise value of the departing arrangement.

Part of Broader Review of Outside Cloud Commitments

The decision forms part of a larger reassessment by Microsoft of more than $80 billion in commitments to rent computing capacity from external providers. Those partners include CoreWeave, Nscale and Nebius. Company executives have grown frustrated with elevated costs and delivery delays at some suppliers.

Microsoft has tightened oversight of data center construction projects, withdrawn from initiatives it views as too risky, and debated whether to enter additional agreements with these providers. Employees have been dispatched to supplier sites to inspect construction standards and compliance. Facilities under closer review include those linked to Nebius, IREN and Nscale.

In a separate step, Microsoft declined an option to expand its existing Texas agreement with Nscale. The company has stated that its overall infrastructure approach combines company-owned data centers, leased facilities and selective outside partnerships to maintain flexibility in bringing capacity online.

Related Developments with Other Partners

The review coincides with adjustments at other Microsoft infrastructure partners. Bitcoin miner IREN reported delays to portions of its Microsoft deployment after the customer requested changes to network infrastructure. IREN shifted its target for more than $4 billion in annual recurring revenue from December into early 2027. The company posted $200 million in AI cloud revenue for the September quarter while Bitcoin mining revenue fell sharply.

Oracle and Microsoft have pursued additional collaborations beyond the original Bing-focused arrangement. In 2024, the companies announced a partnership with OpenAI to supply further capacity through Oracle Cloud Infrastructure, extending Microsoft’s Azure AI platform.

The non-renewal of the older capacity contract does not appear to affect other ongoing relationships between the two firms. Microsoft has not publicly detailed how the specific AI workloads previously handled by the Oracle servers will be relocated.