The Compute Gap: Spending Ahead of Sight
A VentureBeat Pulse Research survey of 107 enterprises paints a stark picture of the current AI infrastructure landscape. Organizations are pouring money into specialized compute at a pace that far outstrips their ability to measure or control its economics. The central tension is simple: only 21% of enterprises run AI in production at scale, yet their spending intentions are already racing toward infrastructure categories almost none of them use today.
The numbers are revealing. 83% of enterprises report GPU utilization at 50% or less, with nearly half (49%) running at 25% or below. Meanwhile, fewer than half (44%) can rigorously track what their AI compute actually costs. The accelerators are running cold, and the ledger is running blind.
A Stack in Flux: From Hyperscalers to Neoclouds
Today's AI runs on familiar ground. Google Cloud leads current deployment at 48%, followed by Microsoft Azure (29%), AWS (22%), and Oracle Cloud (22%). The major model APIs, Gemini, OpenAI, and Anthropic, account for essentially all current usage. The specialized "neocloud" GPU providers that dominate headlines, CoreWeave, Lambda, Crusoe, Nebius, and peers, register at or near zero.
But the next dollar tells a different story. 45% of enterprises plan to evaluate AI-specialized clouds over the next 12 months, the single largest planned evaluation area. Nearly a third (32%) intend to assess non-Nvidia accelerators like AWS Trainium, Google TPU, and AMD Instinct. This is not incremental expansion. It is the leading edge of a re-platforming.
Unusually High Churn for a Foundational Category
For infrastructure as foundational as compute, the churn intent is remarkable. 64% of enterprises plan to switch or add a provider within twelve months, and 38% within the next quarter alone. Only 36% intend to stand still. Much of the near-term movement is reshuffling among incumbents, Microsoft Azure and Google Cloud each draw 33% of switching consideration, while OpenAI draws 30%. The neocloud interest is a 12-month evaluation thesis; the quarter-to-quarter switching is mostly the majors trading share.
Integration and TCO Win; Headline Price Finishes Last
When enterprises choose, they choose on fit and true cost, not sticker price. Integration with the existing stack is the top criterion at 41%, followed by total cost of ownership at 35%. Cost per million tokens, the metric vendors compete on hardest, is the deciding factor for just 8%, dead last. The irony is that most buyers cannot yet measure the TCO they claim to care about most.
The Memory Frontier Arrives Unnoticed
The next constraint in large-scale inference, the shift from GPU compute to memory bandwidth, specifically KV-cache capacity, is barely on the radar. Roughly one in five enterprises are either unaware of this constraint or have not begun to address it. Dell leads planned reliance at 31%, Nvidia follows at 16%, and the rest fragments across storage vendors and open-source tooling. For a shift that will reshape inference architecture, the market is early and unsettled.
What This Means
The compute gap is not a capacity problem that more hardware will solve on its own. It is, first, a visibility problem. Enterprises are planning to buy specialized clouds and alternative accelerators they barely use today, while the GPUs they already own sit substantially idle. The open question is whether they build the instrumentation to spend well before the re-platforming arrives, or buy the next layer of infrastructure as blind to its economics as the last.
Based on survey responses from 107 qualified enterprise respondents (100+ employees), drawn from a single Q2 2026 (June) wave. The sample is self-selected, skews mid-market, and leans toward earlier-stage adopters.