The Cost of Compute:
AI Edition

Executive Summary

Across every part of this survey, one picture holds: AI infrastructure cost is large, rising fast, hard to forecast, and unevenly managed.

For a meaningful share of companies, infrastructure has become one of the largest lines on the P&L, reaching a third of revenue for the companies most dependent on AI, and it’s growing 25 to 50 percent or more a year. It is also hard to see coming. Nearly every company overran an AI cost forecast in the past year, and the overruns concentrate in compute and data storage rather than the model bill most teams watch. Forecasting that cost is now the top concern finance leaders name for the year ahead.

The companies that handle AI cost well share two capabilities: they attribute the cost precisely, down to the individual customer, and they price to recover it. Where those capabilities are present, AI is a healthy line of business. Where they are absent, it is a cost that grows faster than the tools to see it. The number of formal cost controls a company runs does little to separate the two groups; attribution and pricing power do.

The clearest fault line in the data runs between two kinds of company. AI-native companies, whose product is the AI, carry cost visibility as a structural feature of how they are built. AI-enabled companies, traditional software businesses that have added AI features to an existing product, more often absorb AI as a secondary cost they struggle to see or predict. That distinction explains who is managing AI economics well better than company size or company age does, and it holds even after accounting for scale. The report returns to it in every section, because the economics follow the architecture.

A note on scope. This survey measures the cost side of AI infrastructure (spend, growth, forecasting, attribution, pricing) and the margin outcome. It does not measure the drivers of margin change such as headcount, labor productivity, or overall pricing, so where margins are discussed, the analysis is bounded to what the data can support. Full methodology and respondent profile appear at the back.

1. The Cost Is Large and Growing Fast

The first thing the data establishes is scale. For a meaningful share of companies, AI infrastructure is one of the largest cost lines they carry, and it is growing faster than almost anything else on the P&L.

Across the full sample, infrastructure spend (cloud plus AI) runs at a median of 16 to 20 percent of revenue. That median understates the story, because the market is split. One cluster of companies runs infrastructure at 13 to 20 percent of revenue; a second, nearly as large, runs it at 31 to 40 percent. A third of all companies spend more than 30 percent of revenue on infrastructure. The sample holds two distinct populations rather than one, and the split between them follows a clear pattern.

The division runs along business model. For AI-native companies, infrastructure spend reaches roughly a third of revenue, and half of them spend more than 30 percent. AI-enabled companies sit materially lower, with a median in the 16 to 20 percent range and essentially none above 40 percent. When the AI is the product, the infrastructure to run it is a dominant cost of goods; when AI is a feature added to an existing product, it is a smaller, though still growing, line.

About This Research

Cloud Capital partnered with Sapio Research, an independent insights firm, to ensure this report reflects decision-grade, unbiased data. The study was designed to capture how senior finance leaders at growth-stage technology companies are navigating the cost, forecasting, and margin pressures of AI infrastructure spend.

Fieldwork was conducted online in June 2026 using a structured email invitation and a secure survey environment. This ensured consistent screening, validated respondent identities, and eliminated channel-based sampling bias.

We collected 100 fully qualified responses from the senior-most finance leaders at US- and UK-based companies employing 50 to 1,000 people, the scale at which cloud and AI infrastructure becomes one of the largest contributors to cost of goods sold and operating variability. By focusing exclusively on these leaders, the study captures strategic priorities and real-world decision-making rather than departmental sentiment.

These methodological choices give Cloud Capital and Sapio Research high confidence in the clarity, relevance, and reliability of the findings. Our intent is simple: to equip finance leaders with an objective, data-backed benchmark for navigating the economics of AI infrastructure.