Enterprise AI Infrastructure Survey: Computing Power Investment Soars but Measurement Lags, a Significant 'Compute Gap'

VentureBeat Pulse Research surveyed 107 enterprises with over 100 employees in Q2 2026, revealing a significant 'compute gap'. Only 21% have AI workloads in production at scale, but 45% plan to evaluate AI-specific clouds within the next year, 32% non-Nvidia accelerators, and 28% next-gen Nvidia chips. Currently, almost none use dedicated GPU clouds. Meanwhile, 64% plan to switch or add infrastructure providers within 12 months, with 38% acting in the next quarter. Compute efficiency is concerning: 83% have GPU utilization below 50%, 49% below 25%. Only 44% strictly track AI compute costs; most track partially or not at all. Top purchasing criteria are integration with existing stack (41%) and total cost of ownership (35%), not per-token cost (only 8%). As inference scales, memory bandwidth will become the new bottleneck over GPU compute, but about one-fifth of enterprises have yet to address this.