Nvidia's CPU launch is a strategic response to the shift to agentic workloads that change the GPU:CPU ratio. CPU racks cost $300-500K vs $4M+ for GPU racks. Power draw ratio is 7:1. If the workload mix shifts even 20% from GPU-heavy to CPU-heavy, the total dollar value per rack drops dramatically. But on the CPU front there's more competition given every hyperscaler has their own design (TPU, axion, ). In that environment Nvidia is choosing an anti-canibalization strategy to sell chips whereever the compute supply grows. But the Nvidia strategy only works if NVLink lock-in holds. If customers demand interoperability with open source, the bundling strategy unravels.