Why Enterprises Are Prioritizing AI Speed Over Cost Efficiency

According to new research from VentureBeat, enterprises are buying AI compute with performance and GPU availability topping their decision criteria, while fewer than half can rigorously track what those workloads actually cost. The survey of 170 organizations paints a picture of an operational cohort running AI in production — but treating the economics as an afterthought.

The new ranking: speed over TCO

Across the cohort, 66% have AI workloads live and 29% run them at scale, with only 4% not yet running AI at all. The average enterprise operates three infrastructure platforms, with OpenAI (49%), Google Gemini (48%), Microsoft Azure (47%), and Google Cloud (42%) all present in roughly half of them. Azure leads as the single primary platform at 26%.

Here is the catch: the decision criteria have quietly rewritten. Integration with the existing cloud and data stack remains first at 40%, but performance — latency and throughput — has climbed to second at 35%, and GPU availability sits at 24%. Total cost of ownership trails at 22%. The same logic governs measurement: uptime and reliability at 51%, developer productivity at 39%, and cost per million tokens at 31%.

The blind spot underneath

In practice, the rationale makes sense when teams are under production pressure. The uncomfortable part is what sits underneath. Among the 155 enterprises that operate their own GPUs, 69% report utilization at 50% or less, only 23% clear the halfway mark, and 12% do not measure utilization at all. Fewer than half (47%) rigorously track what their AI compute costs and returns, and even among the organizations running AI in production at scale, that figure only reaches 56%. Satisfaction scores tell the rest: value for money sits at 3.87 against 4.14 for overall satisfaction — the softness landing precisely on the dimension hardest to judge without measurement.

Where the next dollar is heading

The next round of spending points somewhere telling. AI-specialized clouds are the top planned evaluation area at 44% and carry the strongest net momentum of any infrastructure approach (+36), yet CoreWeave and Lambda each register at 3.5% of current usage and the rest of the neocloud field sits below 3%. Non-Nvidia accelerators draw 39% of intent. And 62% of enterprises intend to switch or add a provider within 12 months — though the consideration set is dominated by the same incumbents they already run.

For IT leaders and department heads, the operational takeaway is straightforward: treat GPU utilization and cost-per-inference as a first-class metric, not a finance afterthought. The teams that can answer those two questions cleanly will be the ones who can actually evaluate whether the next provider switch delivers more than a slide deck.