Enterprises Buy AI Infrastructure Faster Than They Track Costs

Across 107 enterprises, AI infrastructure spending is accelerating significantly faster than the ability to monitor and manage its associated economics. While most organizations currently utilize hyperscalers and model-provider APIs for their AI workloads, a substantial majority plan to adopt or expand their use of specialized compute providers within the next year, with many intending to make these shifts within a quarter. Purchasing decisions are increasingly driven by integration capabilities and total cost of ownership, rather than just the per-token price of models. This is a pragmatic approach, as most enterprises currently struggle to clearly ascertain their unit economics for AI compute.

This trend highlights a "compute gap," characterized by heavy and rapid investment in AI infrastructure that outpaces the necessary visibility for effective cost control. VentureBeat Pulse Research indicates that only about 21% of enterprises are running AI in production at scale. Despite this, spending intentions are outpacing maturity, with 45% of enterprises planning to evaluate AI-specialized clouds over the next year, a segment currently used by very few of these organizations. Concurrently, existing compute infrastructure is underutilized, with 83% of enterprises reporting GPU utilization at 50% or less.

Furthermore, fewer than half of the surveyed enterprises (44%) can rigorously track the actual costs associated with their AI compute. This indicates a critical disconnect where organizations are acquiring more infrastructure at an accelerated pace without a clear understanding of the expenses tied to their existing assets. The lack of vendor consolidation is also evident, with a clear majority (64%) of enterprises planning to switch or add new infrastructure vendors within the coming year, underscoring the dynamic and unsettled nature of the enterprise AI infrastructure landscape.