AMD secured 2.5 gigawatts of data center capacity from Core Scientific under 15-year leases worth $14B, directly addressing the infrastructure bottleneck constraining AMD deployment. This is the first major power-locked capacity deal for AMD, signaling it can now match Nvidia's hyperscaler relationships and reduce inference cost variance across customer deployments.

Google Cloud's A5X bare-metal instances built on Nvidia Vera Rubin NVL72 deliver 10x lower inference cost per token and 10x higher token throughput per megawatt. This represents a material shift in inference economics and locks Nvidia's lead on per-watt efficiency, directly reducing customer burn on long-running serving workloads.

Nvidia is discussing $250B in financing guarantees to OpenAI for a massive data center project, expanding Nvidia's role from chip supplier into infrastructure financier. This consolidates Nvidia's control over end-to-end capacity and signals confidence in sustained inference scaling, but also creates single-vendor risk for frontier labs.