Firms spent billions on AI they can't measure — and 83% of GPUs are half idle
venturbeat surveyed 107 enterprises and found 83% report gpu utilization under 50% while only 44% track compute costs effectively, signaling a dangerous disconnect between aggressive ai infrastructure spending and economic oversight.
venturbeat research covering 107 enterprises identifies a compute gap where organizations aggressively purchase ai infrastructure despite lacking visibility into its economics. key data shows that 83% of organizations run gpus at 50% utilization or less, only 21% have scaled ai into production, and fewer than half (44%) rigorously track compute costs. spending is shifting towards ai-specialized clouds (45% plan to evaluate) and switching providers (64% within a year, 38% within a quarter), driven by integration (41%) and total cost of ownership (35%) rather than token price (8%). this indicates that while enterprises prepare for inference scaling constraints like memory bandwidth, they currently risk over-provisioning hardware they cannot efficiently measure or manage, signaling a necessary correction towards better unit economics before further capital expenditure.