Only 6% of enterprises are AI "high performers." 20% say cost is already the reason they can't use it more.

McKinsey's 2026 State of AI survey of 1,719 executives found just 6% of enterprises qualify as AI "high performers" — attributing at least 5% of EBIT to AI with a self-described "significant" impact — unchanged from a year ago. Meanwhile, 1 in 5 respondents say AI-related operating costs are already constraining how much they can use it. Enterprises are scaling anyway.

The performance gap isn't closing

37% of respondents attribute at least some EBIT impact to AI use — flat compared to 2025. Only 6% qualify as "high performers," meeting both of McKinsey's criteria: at least 5% of organizational EBIT attributed to AI, and a self-described "significant" impact. That 6% figure is also unchanged year over year. Two full years into heavy enterprise AI investment, the share of companies actually converting it into measurable bottom-line impact hasn't moved.

Cost is now a stated constraint, not just a line item

20% of respondents report that AI-related operating costs have constrained their technology usage. That's a different and more specific claim than "AI is expensive" — it's a fifth of surveyed organizations explicitly saying cost is the reason they can't do more with AI, not a skills gap, not a data problem, not organizational resistance. In a survey this size, that's a meaningful share of enterprises where the budget itself, not the technology's capability, is the ceiling.

And they're scaling anyway

40% of respondents at organizations with $1 billion or more in annual revenue are now scaling AI agents, up from 27% the year before. Agentic workloads are exactly the category with the least predictable, most spiky token consumption — the pattern shows up across this year's reporting on agent cost overruns generally. So the same survey that finds a fifth of enterprises already cost-constrained also finds the fastest-scaling category of AI usage is the one with the least predictable cost profile. Spend is scaling ahead of the ability to forecast it.

Why "measure ROI better" isn't the whole fix

The high-performer number hasn't moved despite two years of industry-wide effort to build better AI ROI measurement frameworks. That flat 6% suggests the bottleneck for most organizations isn't measurement discipline — it's that cost visibility and control haven't kept pace with how fast agentic usage is scaling. It's difficult to attribute EBIT impact to a program whose actual, current cost you're still reconciling after the fact, weeks behind the spend that produced it.

What to check

• Whether your organization tracks "AI cost as a share of what it's constraining" — a project delayed, a use case not deployed — or only tracks total AI spend in isolation.
• Whether agent-driven workloads get the same budgeting rigor as the rest of the AI program, given they're the fastest-scaling piece and the least predictable to forecast.
• Whether cost visibility keeps pace with usage growth, or consistently trails it by a billing cycle.

The bottom line

A flat 6% high-performer rate across a year of continued investment is not a measurement problem waiting on a better framework — it's a signal that cost and usage are moving faster than the visibility built to manage them. Twenty percent of enterprises already know cost is the ceiling. The other eighty percent are about to find out, at the exact moment agentic adoption is accelerating fastest.