India, Aug. 26 -- AM Intelligence's (AMI) order for nearly 9,000 Nvidia Rubin GPUs has all the ingredients of a big-ticket AI infrastructure announcement. It brings Nvidia's next-generation architecture to Hyderabad, promises about 450 exaFLOPS of NVFP4 inference compute, and forms part of a much larger plan for 1 GW of Compute-as-a-Service capacity.

Yet, one of the more interesting numbers in the announcement is easy to overlook: 30 megawatts (MW). That is the capacity AMI says will support the first deployment of its Vera Rubin NVL72 rack-scale systems, with deliveries beginning in the first quarter of 2027.

The pairing of 9,000 GPUs with 30 MW brings another dimension of the AI infrastructure race into focus. As accelerator deploymen...