A survey of 107 enterprises reveals a 'compute gap' where infrastructure spending is accelerating while most companies lack the ability to track costs or optimize GPU utilization.

A new VentureBeat Pulse Research report finds that enterprises are aggressively investing in AI infrastructure despite significant gaps in economic visibility. While spending plans are rising, most organizations struggle to track their actual compute costs or maintain efficient hardware utilization. The study, which surveyed 107 enterprises in June 2026, highlights a disconnect between rapid infrastructure acquisition and the ability to manage its underlying economics.

The AI Compute Gap

The research identifies a 'compute gap' characterized by heavy investment running ahead of operational visibility. Only 21% of surveyed enterprises currently run AI in production at scale, yet 45% plan to evaluate AI-specialized clouds in the next year—a category currently used by almost none of the respondents. Meanwhile, existing infrastructure remains largely underutilized, with 83% of organizations reporting GPU utilization rates of 50% or less.

Infrastructure Switching and Selection

A majority of enterprises (64%) intend to switch or add an infrastructure provider within the next 12 months, with 38% planning such changes within the next quarter. When selecting providers, companies prioritize integration with their existing technology stacks (41%) and total cost of ownership (35%). Notably, the headline cost per million tokens is the primary deciding factor for only 8% of respondents.

Measurement Challenges

Despite the focus on total cost of ownership, fewer than half (44%) of the surveyed enterprises can rigorously track their AI compute costs. Furthermore, 8% of organizations do not measure their GPU utilization at all. This lack of visibility persists even as companies prepare to re-platform their AI operations toward specialized cloud providers and non-Nvidia accelerators.