Microsoft (NASDAQ: MSFT) is gradually ending a multiyear arrangement under which it rented artificial intelligence computing capacity from Oracle (NYSE: ORCL), according to people familiar with the situation.
The move forms part of a wider review of the company’s extensive external cloud commitments, which total more than $80 billion across various providers.
In recent months, the software giant has been shutting down the Oracle-supplied capacity after choosing not to renew the underlying contract. Sources indicate that the rented resources grew progressively less suitable for Microsoft’s requirements.
The hardware in question relied on an earlier generation of Nvidia processors that no longer aligned with the company’s evolving technical needs for its AI workloads.
The original agreement, announced in late 2023, involved Oracle granting Microsoft access to servers on its Oracle Cloud Infrastructure platform.
Those servers were intended to support an AI-enhanced version of the Bing search engine, working in tandem with Microsoft’s own Azure infrastructure. At the time, the partnership was viewed as a notable example of a major technology company turning to a rival for additional processing power amid intense industry-wide demand.
The precise financial scale of the arrangement has not been disclosed.
Microsoft has declined to comment on the development.
An Oracle spokesperson stated that interest in the company’s AI infrastructure continues to be robust, including for graphics processing units from earlier generations.
The company has previously noted that certain older GPU contracts have been renewed or reallocated at a premium to prior pricing.
The decision arrives against a backdrop of Microsoft reassessing how it secures the massive computing resources required for its AI initiatives.
The company has signed substantial agreements with specialized “neocloud” operators such as CoreWeave, Nscale, and Nebius to supplement its internal capacity.
However, senior leaders have expressed frustration with elevated costs and project delays associated with some of these partnerships.
In response, Microsoft has increased oversight of data-center developments, withdrawn from initiatives it considers higher-risk, and is internally evaluating whether to pursue additional contracts with such providers.
The original willingness to rent capacity from Oracle underscored the acute shortage of advanced processing power that has characterized the rapid expansion of generative AI services.
What began as a practical response to constrained supply is now being recalibrated as Microsoft’s own infrastructure and requirements mature.For Oracle, the cloud business has grown significantly in recent years, driven in part by large AI-related agreements with major technology companies.
Its cloud revenue reached $18 billion in the fiscal year ending May 2026.
Nonetheless, news of the Microsoft wind-down contributed to a roughly 3 percent decline in Oracle’s shares.
The episode illustrates how quickly AI infrastructure arrangements can shift as chip generations advance and large buyers refine their technical specifications and procurement strategies. While external capacity remains important for many organizations, Microsoft’s actions suggest a growing preference for tighter control over the resources that underpin its AI products and services.
