Enterprises buying AI compute faster than they can measure costs

VentureBeat Pulse Research examined AI infrastructure spending across 107 enterprises, uncovering a significant gap between the speed of investment and the visibility organizations have into its financial impact. The research captures a snapshot of enterprise AI deployment during a period of rapid change: organizations are currently running AI primarily on familiar infrastructure—hyperscalers (large cloud providers) and model-provider APIs—yet the next wave of spending is directed toward specialized compute almost none of them deploy today.

The deployment landscape is characterized by planned churn. A majority of the surveyed enterprises intend to switch or add compute providers within the year, with many of those changes planned for within a quarter. This high rate of provider switching suggests both competitive intensity in the specialized compute market and organizational uncertainty about the optimal configuration for their workloads. Rather than optimizing purely on token price—the metric often emphasized in public vendor comparisons—enterprises report that buying decisions are driven by integration capabilities and total cost of ownership. This indicates a maturing procurement function focused on long-term operational economics.

Yet this sophistication in decision-making does not extend to cost visibility. The survey found that GPUs sit at half utilization or less across many organizations, a sign of either overprovisioning, suboptimal workload matching, or poor scheduling. More strikingly, fewer than half of the 107 enterprises rigorously track what their compute actually costs. Without clear unit economics, organizations cannot determine whether they are optimizing effectively or burning money on idle capacity. The result is what the research terms a 'compute gap'—heavy, fast-moving investment running ahead of the visibility and steering mechanisms needed to control it. As AI infrastructure spending accelerates, this gap represents both a operational risk for enterprises and a potential opportunity for vendors and tools that can help close it.

Enterprise AI infrastructure spending is entering a critical inflection point where capital deployment is outpacing financial discipline. The survey reveals that while organizations have settled on familiar hyperscalers and model-provider APIs as their baseline, they are simultaneously moving toward specialized compute—infrastructure most do not yet use at scale. This suggests a market in flux: vendors are gaining traction not on token price alone, but on integration capabilities and demonstrable total cost of ownership.

The visibility gap is the core finding. With GPUs running at half utilization or lower and fewer than half of organizations able to articulate their unit economics, enterprises are effectively flying blind into what may be substantial cost overruns. The fact that buying decisions hinge on total cost of ownership—rather than the headline metrics that dominate vendor marketing—indicates that procurement teams are sophisticated enough to ask the right questions, but lack the internal instrumentation to answer them. This creates a paradox: spending is accelerating precisely where cost control is weakest.