Oracle expands Nvidia AI chip deal, adds Google and other suppliers - The Times of India

Oracle's move is not incremental procurement—it is a structural dismantling of CUDA's monopoly. By integrating Google TPUs alongside Nvidia GPUs, OCI must now sustain dual training and inference stacks (CUDA plus XLA), forcing a fundamental rewrite of its compute scheduler. This directly erodes Nvidia's ecosystem lock-in, the single most valuable asset underpinning its $3 trillion valuation. On supply-chain risk, single-vendor dependency on Nvidia is a systemic vulnerability under tightening US export controls. Diversifying into TPU and potentially AMD MI300 is a compliance hedge, not merely a cost optimization play. The competitive geometry is shifting: Google simultaneously operates GCP and supplies Oracle—cloud rivals are becoming each other's vendors. Expect AWS and Azure to counter by onboarding AMD or accelerating in-house silicon (Trainium, Maia) to preserve bargaining leverage. Within 18 months, heterogeneous compute orchestration becomes the real moat for hyperscalers. Nvidia's 75%+ gross margin, long insulated by software gravity, will face its first genuine de-CUDA pressure. The chip war is no longer about who builds the best accelerator—it is about who controls the scheduling layer above it.