Microsoft Reportedly Ends an AI Cloud Deal With Oracle
By Adam, AI Engineer · 9 November 2026 · 3 min read
Microsoft is ending a deal to rent AI computing power from Oracle, Bloomberg reported on 6 November, citing people familiar with the matter. Microsoft chose not to renew the contract and has been switching off the rented servers.
It is a small sign of a bigger change, as Wall Street questions the AI spending boom. The companies that spent the most on AI computers are starting to look harder at where that money goes.
What the Microsoft Oracle deal was
- The deal started in 2023, when Oracle agreed to supply servers for Microsoft's AI-powered Bing search, according to the report.
- The servers used older Nvidia chips. The people told Bloomberg those chips no longer matched what Microsoft needs.
- Bloomberg could not find out how big the contract was. Microsoft declined to comment.
Oracle says demand is still strong
Oracle pushed back in a public statement. It said demand for its AI computers is still strong, even for older chips. Oracle also said capacity that came up for renewal in its last quarter was renewed or resold at prices 20% higher than before, on long-term contracts.
A wider review of AI data center deals
Bloomberg says this is part of a larger review. According to its reporting:
- Microsoft has committed more than $80 billion to renting computing power from outside providers such as CoreWeave, Nscale and Nebius.
- Some executives were unhappy with high costs and delays from suppliers.
- Microsoft has tightened its checks on data centers being built for it, pulled out of projects it sees as risky, and sent staff to supplier sites.
Microsoft told Reuters that it mixes its own data centers, leased buildings and outside partners, which it says gives it flexibility.
Why it matters
For years, big tech firms rented almost any AI computing power they could find. This report suggests the buyers are now becoming choosier: they want newer chips and suppliers who deliver on time. For business owners, this is analysis, not a forecast, but it is a reminder that AI hardware ages fast. If you rent cloud AI services, prefer flexible contracts over long lock-ins, and ask your provider what hardware your plan runs on.