Microsoft Phasing Out Oracle AI Compute Leases Over Older Nvidia Chips

The signal here is about hardware vintage, not demand: even a hyperscaler with vast internal capacity is treating older Nvidia generations as unfit for its own workloads, which pressures the economics of any leased fleet built on previous-generation silicon. Oracle's insistence that early-GPU demand remains strong suggests a two-tier market is forming, where legacy compute is absorbed by price-sensitive or less demanding customers. The open question is whether Microsoft's review of its cloud partnerships extends beyond Oracle to the other lease arrangements named in the report, and how quickly Oracle can refresh capacity to keep large tenants.