ToolNavs Find Useful AI Tools
Submit Sign in
Back to AI information
Microsoft Won't Renew Oracle AI Capacity: Older Chips Can't Keep Up as Its $80B Cloud Bill Gets Reworked

Microsoft Won't Renew Oracle AI Capacity: Older Chips Can't Keep Up as Its $80B Cloud Bill Gets Reworked

AI information • Admin • • 18 views

On November 6, Bloomberg reported that Microsoft is gradually switching off AI computing capacity it rents from Oracle, because the servers run older-generation Nvidia chips that no longer meet Microsoft's needs. This is not a sudden cutoff: Microsoft decided months ago not to renew the contract and has been decommissioning the capacity in stages since. For Oracle, it is a warning shot — the AI compute rental business is shifting from "any GPU will do" to "only new hardware counts".

Why walk away from three years of rented capacity?

The deal was signed in 2023, when Microsoft turned to Oracle Cloud Infrastructure (OCI) for extra AI compute to power its AI-assisted Bing search, signing a multi-year agreement whose value was never disclosed. Three years on, Microsoft's needs have moved on: models are bigger and the hardware bar for training and inference is higher, while Oracle's servers still run previous-generation Nvidia chips. People familiar with the matter told Bloomberg that Microsoft judged the capacity "no longer meets the company's evolving needs" and chose not to renew. Microsoft declined to comment.

Oracle is not the only one

The backdrop is bigger than one contract. Bloomberg said Microsoft is reassessing more than $80 billion in commitments to rent capacity from outside providers, including CoreWeave, Nscale and Nebius. Oracle shares fell about 3% on the news. Oracle responded that demand for its AI infrastructure "remains strong", and that capacity coming up for renewal was renewed or resold at a 20% premium to prior pricing. The two accounts are not contradictory: there are buyers for older chips — Microsoft just wants newer ones.

For ordinary users, the change has almost no direct effect — Bing's AI features do not depend on this rented capacity. For the industry, the signal matters more: older GPUs in cloud providers' hands are losing value fast, while the biggest buyers would rather build their own capacity than keep paying for outdated hardware. Also in November, Google announced a $17 billion data center in Wyoming. Building new capacity while Microsoft cuts old — the filter in the compute race is getting stricter. What to watch next is how much more of Microsoft's $80 billion-plus in outside commitments gets cut, and how many new buyers Oracle can find for its older chips.

Recommended Tools

More