Here is a number that should stop any CFO cold: 83% of enterprises running their own GPUs report utilization at 50% or less, and nearly half of those sit at 25% or below. The accelerators they bought to power AI are running cold while the invoices keep coming. And fewer than half of those same enterprises can rigorously track what their AI compute actually costs.
This is the compute gap, and it is the central finding of a new VentureBeat Pulse Research survey of 107 organizations with more than 100 employees, fielded in June 2026. The gap is not a shortage of hardware. It is a shortage of visibility. Enterprises are buying AI infrastructure faster than they can see what they already own, and faster than they can steer its economics.
The maturity mismatch
Only about one in five enterprises, 21%, say they run AI in production at scale. Three-quarters are still experimenting or running only some workloads live. That matters because the spending intentions in this report are not coming from mature operators who have found what works. They are coming from organizations still building out, whose compute footprints, and whose costs, are about to grow substantially.
The single largest planned evaluation area over the next twelve months is AI-specialized clouds, at 45%. That is the category almost none of these enterprises use today. CoreWeave, Lambda, Crusoe, Nebius, and their peers register at or near zero in current deployment. Google Cloud leads current usage at 48%, followed by Microsoft Azure at 29%, AWS at 22%, and the major model APIs. The stack is hyperscaler-and-API, but the next dollar is aimed somewhere else entirely.
This is not incremental expansion. It is the leading edge of a re-platforming. Specialized AI clouds carry the highest net momentum score in the survey, edging out even the hyperscalers. Two consecutive survey waves have produced the same pattern: the type of cloud enterprises are most eager to assess is the type they have barely begun to use.
Churn at the foundation
For a category as foundational as compute, the switching intent is unusually high. A clear majority, 64%, plan to switch or add an infrastructure provider within twelve months. Thirty-eight percent plan to do so within the next quarter alone. Only 36% intend to stand pat.
The near-term movement is mostly reshuffling among incumbents. Microsoft Azure and Google Cloud each draw 33% switching consideration, OpenAI 30%, and Gemini 22%. The neocloud interest is a twelve-month evaluation thesis. The switching in the next quarter is the majors trading share.
What buyers actually value
When enterprises choose a provider, they do not buy on headline price. Integration with the existing stack is the top criterion at 41%. Total cost of ownership follows at 35%. Cost per million tokens, the metric vendors compete on hardest, is the deciding factor for just 8%, dead last.
That pattern is coherent but also foreshadows the measurement problem. Enterprises say TCO matters most, yet most cannot yet measure it rigorously. The stated priority and the measured capability are out of step.
The cold fleet and the blind ledger
The compute already in place runs cold. Eighty-three percent of GPU-operating enterprises report utilization at or below 50%. Nearly half run at 25% or below. Only 12% clear the 50% mark, and a further 8% do not measure utilization at all. Idle accelerators are expensive accelerators, and the efficiency headroom in the current fleet is large and largely unmeasured.
Fewer than half of enterprises, 44%, rigorously track the cost and return of their AI compute. The majority track only partially, 39%, cannot quantify it yet, 20%, or have not prioritized it, 6%. Satisfaction with current infrastructure averages 4.0 out of 5, moderately positive but not enthusiastic. Ease of implementation scores 3.8, and value for money scores 3.9. The softness lands on cost.
The next frontier, barely seen
The shift from GPU compute to memory bandwidth, specifically KV-cache capacity, as the binding constraint in large-scale inference is arriving. Enterprises scatter in their response. Dell leads at 31%, Nvidia follows at 16%, and the rest fragments across storage vendors, open-source tooling, and model-level efficiency. Roughly one in five either do not recognize the constraint or have not begun to address it. It is the next chapter of the compute gap, arriving before most have closed the current one.
At 107 respondents in a single wave, this is a directional read, skewed toward the mid-market and earlier-stage adopters. But the direction is consistent. The appetite to spend is running well ahead of the instrumentation to spend well. The compute gap is not a capacity problem that more hardware will solve on its own. It is, first, a problem of seeing what the hardware already costs. The open question is whether enterprises build that visibility before the re-platforming arrives, or buy the next layer of infrastructure as blind to its economics as the last.