Enterprises lock GPUs in place through FOMO and it drives utilization to 5% as costs climb

GPU waste persists because releasing capacity risks losing it

A new report says many enterprises are running GPU fleets at only about 5% utilization, not due to poor software alone, but because procurement fear traps them. Shortages push companies onto costly reserved allocations, then makes releasing idle GPUs feel impossible. At the same time, AI workloads stay GPU-idle inside containers, compounding waste as frontier chip pricing rises.