Survey: 79% of enterprise GPUs are idle at least half the time, while 70% of enterprises still plan to expand their AI platforms.

Beating AI News Flash, a Q3 2026 survey compiled by VentureBeat Intelligence shows that 79% of enterprise respondents report their GPU computing power is idle at least half the time, yet 70% of surveyed firms plan to introduce, expand, or replace AI computing platforms over the next 12 months. Return on investment (ROI) assessments are also lagging. The survey found that among 64 respondents planning to adopt new AI platforms, 58 lack finance-audited ROI data, accounting for 91%. Additionally, while enterprises underutilize their own GPUs, they continue to access closed-source models from providers like OpenAI and Anthropic via API calls, creating a mismatch between idle hardware resources and model call expenditures. As AI Agents gain popularity, token consumption and inference costs from multi-step tasks are drawing growing attention. Gartner forecasts that by 2028, inference costs for a single Agent workflow will surge more than fivefold. The report notes that enterprise AI competition is shifting from competing for computing power to improving resource utilization and controlling per-task costs, with specific measures including migrating routine tasks such as classification and summarization to open models, optimizing context caching, and reducing unnecessary multi-Agent collaboration. However, GPU idleness may also stem from enterprises reserving computing power for peak demand or planning to replace outdated platforms; the survey data alone does not directly confirm the formation of an AI bubble.