# Enterprises are investing in AI infrastructure but can't measure costs AI infrastructure spending outpaces economic visibility. - **Source:** [VentureBeat (feeds.feedburner.com)](https://brifing.si/source/c51474dc-6b09-49dd-aaca-085f4ed00c84) - **Published:** 2026-07-23 - **Topic:** [Data & AI](https://brifing.si/c/data-ai) - **Original article:** - **On Brifing:** https://brifing.si/s/6bdc592d-6fba-4b00-bb21-090c22884407 ## AI infrastructure investments are accelerating Organizations are investing in AI infrastructure faster than they can measure costs. A survey of 107 enterprises shows spending is outpacing economic visibility. Most run their AI on hyperscalers and model APIs, highlighting a gap between investment and oversight. ## GPU utilization remains low 83% of enterprises report GPU utilization at 50% or less, indicating inefficient resource use. Fewer than half (44%) rigorously track AI compute costs, making financial control challenging. This gap between spending and accountability risks unchecked expenses. ## High intent to switch providers 64% of organizations plan to switch or add infrastructure providers within 12 months—38% within a quarter. Decisions prioritize integration (41%) and total cost of ownership (35%) over headline pricing. Token cost per million influences only 8%, reflecting long-term operational priorities. ## Specialized clouds barely used today Most enterprises rely on hyperscalers for AI workloads, with niche GPU providers near 0% adoption. Yet 45% plan to evaluate AI-specialized clouds next year—a stark shift from current stacks. This signals impending market disruption as compute needs evolve. ## Integration outweighs pricing When selecting providers, enterprises prioritize stack compatibility (41%) and TCO (35%) over per-token costs (8%). This underscores the importance of seamless operations in AI infrastructure, where long-term efficiency trumps short-term savings. ## Limited production-scale AI Only 21% run AI in production at scale, while 76% remain in experimentation or partial deployment. Infrastructure decisions are thus being made by early adopters whose costs will grow substantially, not by operators with mature deployments. ## Memory bandwidth awareness lags The coming shift from GPU compute to memory bandwidth as inference scales remains off-radar for 20% of enterprises. This looming bottleneck—critical for next-gen AI—highlights a knowledge gap in future-proofing infrastructure investments. ## Mid-market leads adoption Survey respondents skew toward mid-sized firms (101-1,000 employees), with Technology (26%) and Healthcare (15%) as top sectors. This reflects AI infrastructure build-out by active adopters rather than hyperscale operators, shaping early-market dynamics. ## Decision-makers drive purchases 45% of participants hold final purchasing authority for AI solutions, with 30% as influencers. Their focus on integration and TCO reveals how operational needs—not just cost—guide infrastructure choices in this nascent market. ## Cost tracking remains weak Fewer than half of enterprises rigorously monitor AI compute expenses, risking unchecked spending. Improving cost visibility is critical for sustainable investment—especially as GPU utilization hovers below 50%, signaling wasted capacity. ## Future challenges emerge Rapid AI infrastructure growth brings cost control and efficiency to the forefront. Organizations must develop better measurement tools and strategies to align investments with returns, ensuring scalable and economically viable AI deployments. ## AI infrastructure management grows critical The speed of AI investment makes governance indispensable. Enterprises need new approaches to optimize resource use and maximize ROI. Mastering this balance will define who sustains competitive advantage in the AI era. ## Sources covering this story - https://venturebeat.com/resources/the-ai-compute-gap-enterprises-are-buying-infrastructure-faster-than-they-can-measure-what-it-costs — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._