Key takeaways
- Microsoft reportedly stops renewing Oracle AI computing capacity
- Older Nvidia GPUs no longer satisfy Microsoft's requirements
- Oracle insists demand remains strong despite contract changes
Microsoft is reportedly winding down its agreement to rent artificial intelligence computing capacity from Oracle, raising fresh questions about how quickly expensive Nvidia GPUs can become commercially outdated.
The companies originally announced their multiyear agreement in November 2023 to support AI-powered Bing searches. However, people familiar with the arrangement said on November 6 that Microsoft had been switching off rented capacity after declining to renew the contract because the underlying hardware no longer met its requirements.
Microsoft Retreats as Older Nvidia GPUs Lose Appeal
The decision exposes an emerging risk for AI infrastructure providers: computing hardware can remain operational while becoming less attractive to customers adopting newer AI models and increasingly demanding workloads.
Oracle has challenged suggestions that older GPUs are losing their commercial value. The company said capacity coming up for renewal in its fiscal first quarter had been renewed or resold at prices approximately 20% above previous long-term contracts.
That distinction matters. Microsoft's changing specifications do not necessarily mean older Nvidia chips have become obsolete across the entire
The reported withdrawal also follows Microsoft's decision to reject a potentially multibillion-dollar Nscale expansion, while leaving existing commitments intact.
Another infrastructure partner, IREN, recently delayed portions of its Microsoft deployment after customer-requested networking changes. These separate developments suggest Microsoft is becoming more selective about the computing capacity it purchases.
Oracle's $664 Billion Backlog Faces New Questions
The timing is particularly sensitive for Oracle, whose aggressive AI infrastructure expansion has attracted scrutiny over capital spending, debt and future cash generation.
Oracle's September earnings showed cloud infrastructure revenue jumping 121% year over year to $7.4 billion. Remaining performance obligations reached $664 billion, reflecting substantial contracted future business.
However, a large backlog does not guarantee that every server will maintain its economic value throughout its operating life.
The company reported negative free cash flow of $23.7 billion for fiscal 2026, illustrating the financial burden of building AI infrastructure ahead of revenue realization.
The issue extends beyond Oracle. Rising borrowing costs have already intensified concerns about AI infrastructure debt, particularly among companies financing enormous computing facilities through long-term contracts.
