Microsoft Gradually Terminates AI Cloud Computing Leases from Oracle
nashnova research
Microsoft is phasing out its AI compute lease with Oracle, triggered by outdated Nvidia chips on Oracle's hardware and driven by a broader pullback on external compute spending.
Why is Microsoft dropping Oracle's compute?
Oracle's servers run an older generation of Nvidia chips that no longer meet Microsoft's current hardware specs.
This means → the deal didn't collapse over price — the hardware aged out. Chip generations move so fast that last-gen compute loses AI-training value rapidly.
What bigger shift is behind this?
Microsoft had committed over $80 billion to leasing compute from external cloud providers — a massive outlay.
In practice, these "neocloud" vendors delivered higher-than-expected costs and repeated delays.
This means → cutting Oracle is not a one-off. It is part of Microsoft's full-scale reassessment of its external compute procurement strategy.
What signal does this send?
In plain terms = Microsoft spent $80 billion-plus renting other people's compute, found it too expensive and too slow, and is now pulling back.
This reflects a broader market shift: AI compute is moving from a "grab everything available" expansion phase into a "spend only where it pays" selection phase.
For Oracle, losing a customer this large is a clear negative — falling a chip generation behind was enough to get cut.
市场有风险，内容仅供研究参考，不构成投资建议。