Microsoft Winds Down Oracle AI Server Deal Over Older Nvidia Hardware

Microsoft is winding down an artificial intelligence server rental agreement with Oracle as it phases out older hardware in favor of newer accelerator architectures, according to reporting from [Yahoo Finance](https://247wallst.com/investing/2026/11/07/older-nvidia-chips-just-cost-oracle-its-microsoft-ai-deal/?.tsrc=rss). People familiar with the arrangement said Microsoft chose not to renew the contract and spent recent months turning off leased clusters because Oracle's servers relied on older Nvidia graphics processing units that no longer meet Microsoft's current workload specifications.

The exit marks a distinct turning point in cloud capacity dynamics. During the initial compute shortages that accompanied the generative AI boom, large technology companies leased capacity wherever clusters could be found. Now that next-generation hardware is reaching operational scale, hyperscale operators are auditing those stopgap arrangements and winnowing third-party suppliers to control running costs and operational friction.

The rental contract originated in 2023, when Microsoft sought supplemental infrastructure to power the conversational AI rebuild of its Bing search engine. The two companies never publicly disclosed the financial terms or compute volume of the deployment. The specific generation of Nvidia silicon inside the clusters was not detailed, and Microsoft declined to comment on the phase-out.

Oracle maintained that secondary demand for legacy silicon remains resilient. A company spokesperson told the publication that demand for its AI infrastructure continues to be robust across multiple hardware generations. In September, Oracle reported that GPUs reaching renewal were either extended or re-leased to other customers at an average 20% premium, noting that most of those processors were four years old or older. The company reported overall GPU fleet utilization of 97.9% for that quarter.

The Bing cluster accounts for a modest slice of Oracle's expanding cloud division, which generated $18 billion in revenue for the fiscal year ended May 2026 on the back of large-scale commitments from clients such as OpenAI. Oracle has maintained a longer-term financial target of at least $90 billion in revenue and non-GAAP earnings of $8.10 per share for fiscal 2027.

Microsoft has simultaneously expanded its balance-sheet investments in direct infrastructure ownership, reporting $45.2 billion in capital expenditures during its September quarter. Nvidia GB300 systems now comprise the majority of Microsoft's internal GPU fleet. Although Microsoft has committed more than $80 billion to external compute providers such as CoreWeave and Nebius, it plans to narrow its external roster to a consolidated group of key vendors.

Oracle shares slid to an intraday 52-week low of $109.21 on Nov. 6 following word of the contract wind-down before recovering to finish up 0.74% at $116.01, valuing the company at $361.69 billion. Oracle's stock has surrendered more than half its value since reaching a record $248.15 on June 1, complicated by negative free cash flow of $5.396 billion in its most recent quarter. Upcoming earnings disclosures from CoreWeave on Nov. 10, Nvidia on Nov. 17, and Oracle's own quarterly release around Dec. 9 will test whether secondary market premiums for aging GPU fleets can endure.