Enterprise AI compute spending outpaces cost visibility
Companies' AI infrastructure investments are accelerating faster than their ability to track costs. Compute utilization is low, and few can accurately monitor expenses.
New VentureBeat Pulse research indicates a significant "compute gap" in enterprise AI infrastructure. Investments are accelerating well ahead of organizations' ability to understand or steer the associated economics.
The survey of 107 enterprises found that while most AI operations run on familiar hyperscalers and model provider APIs, the next wave of investment is targeting specialized compute that few of these companies currently utilize. A majority plan to switch or add providers within a year, many within the next quarter.
A key issue is inefficient compute utilization, with 83% of enterprises reporting GPU utilization of 50% or less. Furthermore, fewer than half (44%) can rigorously track the actual costs of their AI compute resources. Companies are acquiring new infrastructure faster than they can account for existing owned resources.
Decisions are driven by integration and total cost of ownership rather than headline token prices, which is fortunate given that most enterprises cannot yet see their unit economics clearly. While only about one in five (21%) run AI in production at scale, spending intentions are outpacing this maturity. The primary planned focus for the upcoming year is specialized AI clouds (45%), a layer almost none of the surveyed enterprises currently use.
Despite enterprises being unsettled with current infrastructure providers, with 64% planning to switch or add a provider, decisions remain driven by integration with existing stacks (41%) and total cost of ownership (35%), rather than immediate interface pricing.