North American CSPs to Buy NVIDIA Rubin Systems, Boosting AI Inference Capacity

North American cloud service providers are expanding AI infrastructure. NVIDIA GB and Rubin rack-scale systems integrate GPUs, networking and cooling to accelerate data center deployment. Providers including Google and Amazon are also developing custom ASICs such as TPU and Trainium to lower costs and reduce reliance on a single supplier.

A new report forecasts that North America’s five largest CSPs will make major purchases of NVIDIA GB and Rubin rack-scale systems in 2026, driving a 122% increase in AI inference capacity from 2025—equivalent to a jump of more than 1.2 times. Providers are also expected to adopt liquid cooling across the board as power consumption and computing density rise. The report has not disclosed how much each company plans to spend.