Enterprise AI Adoption Moves From Usage Sprawl to Cost Control

AI spending is still rising, but the rules are changing

A commentary that ran in late August announced that "the AI spending spree is over." It offered companies five steps for the next phase of AI consumption: get visibility across the full technology stack, build governance, renegotiate contracts, align technology with finance and procurement, and use AI to keep optimizing spend on an ongoing basis.12 The author's main point is that leaders no longer judge success by how widely AI is used. They now ask what it costs, where it pays off, and whether the business can afford to scale it. The author's name for this shift is "valuemaxxing."1

The headline sounds like budgets are shrinking. The numbers say otherwise. Gartner's forecasts for worldwide AI spending in 2026 run from about $2.59 trillion, a 47% rise, to a September revision of $2.67 trillion, up 49.5% from 2025.68 BCG's AI Radar survey found that companies expect to spend about 1.7% of revenue on AI in 2026, up from 0.8% the year before.5 A Bloomberg Intelligence survey found that two-thirds of enterprises at least doubled their AI token use in the past year, and 92% expect token budgets to grow next year.7

So the spending hasn't stopped. What's ending is the period when companies spent without asking much about results. That is the right way to read the August piece, and it is a bigger change for enterprise AI adoption than a budget cut would be.

Wide use, thin returns

Most recent surveys tell the same story: nearly every company uses AI, and few can show it makes money. Stanford HAI reports that 88% of surveyed organizations used AI in 2025. McKinsey's latest State of AI survey finds that 37% of respondents credit AI with any EBIT impact at all, and only about 6% are "high performers" who credit it with 5% or more of EBIT.3 The share reporting any EBIT impact was basically flat year over year, even as the share saying AI is scaling across the enterprise rose from 38% to 44%.86

Other measures are similar. KPMG found that only 7% of leaders report established ROI, and only 12% of organizations consistently weigh AI value against cost.3 In the Bloomberg Intelligence survey, just 9% could link token consumption to ROI with high confidence.7 PwC's January 2026 survey found 56% of CEOs reporting no measurable ROI despite active deployments.5

The exact figures vary because definitions vary. One aggregator cites McKinsey data showing 63% of companies reporting measurable positive ROI, with a median payback of 14 months.4 Another analysis puts enterprise-level EBIT impact at 39% and cites Deloitte's finding that 73% of enterprises say their most advanced generative AI project meets or beats ROI expectations.10 These numbers don't really contradict the pessimistic ones. They measure different things: a single flagship project can pay off while the company as a whole sees no change in profit. Analysts at ArticSledge describe this directly: individual productivity gains run ahead of enterprise gains, and spending runs ahead of measurement.3

Across the data, the gap is real and it isn't closing on its own. Rising spend with flat EBIT attribution is exactly the situation that gets CFOs involved.

Why consumption pricing changes things

The August piece is right that the cost model has changed, not just the cost level. Seat-based software was predictable: one license per employee, one line item. AI spend now shows up as tokens, credits, model calls, GPU hours, data movement, and AI features built into existing applications.1 Bloomberg Intelligence found that 82% of enterprises reach models through APIs, either fully or in a hybrid setup. That means AI costs grow with the amount of work done, not with the number of licensed users.7

Vendor pricing is moving the same way. Hybrid pricing, a base subscription plus usage charges, is now the most common model among B2B software companies at 37%, up from 25% a year earlier. Only 5% of investors prefer pure seat-based pricing.6 One finance-focused analysis notes that usage pricing can swing a company's spend by 40% or more from month to month with the same headcount, and cites Zylo data showing 78% of IT leaders hit unexpected charges from consumption-based and AI pricing.9

The August commentary argues that companies can't manage what they can't see. It says only 31% of organizations have visibility into their AI software, while 59% report more wasted AI spend year over year.1 Ramp data suggests the median AI-spending business now puts about 15% of its software budget into AI tools, a category that barely existed two years ago.9 BCG estimates that roughly 80% of company AI spend sits outside enterprise IT budgets.3 Put together, AI has become a large, scattered cost that no single owner fully tracks.

The five steps, compared with the data

The prescriptions mostly hold up, and the stronger evidence elsewhere backs them.

Visibility comes first because the gap in visibility is the most direct cause of the ROI problem. Companies can't tie spending to value if they can't see a large share of the spending.13

Governance is moving fast. The commentary says 85% of large enterprises are naming dedicated teams or senior leaders for AI and technology governance. It frames the goal as moving from "use more" to "use better," with use measured against cycle time, customer experience, efficiency, or revenue.2 McKinsey's research on high performers supports this. Those companies are about twice as likely to have defined processes for measuring the impact of AI projects, and nearly three-quarters redesigned workflows around AI, compared with about a quarter of everyone else.6

Contract renegotiation depends on how easy it is to switch. The commentary warns that pricing will get more complicated, not less, and says contracts should match actual usage rather than lock in fixed terms.1 Bloomberg's data shows how much leverage buyers may have. If token prices rose 50%, 84% of respondents said they would switch models or tune their usage, and many already review their model choices every quarter.7 Weak loyalty to any one model gives procurement teams real bargaining power.

Aligning technology, finance, and procurement responds to a split in ownership. Engineering teams drive usage, finance carries the budget impact, and procurement owns the contracts.1 FinOps practice has already moved to cover this. According to FinOps Foundation surveys, 31% of practitioners managed AI spend in 2024, 63% in 2025, and 98% in 2026.12

AI-driven optimization, meaning AI tools that flag unusual usage and forecast demand, is the least proven of the five. The author notes it has to stay tied to goals that humans set.1 It makes sense, but no independent data yet shows it delivers.

What the piece leaves out

The commentary reads as a playbook for technology spend management, and it focuses on cost discipline. The wider data suggests cost discipline alone won't close the value gap. Deloitte respondents say 37% of them use AI at a surface level with little process change.3 Benchmark data from Eliassen found that enterprises with $1 billion or more in revenue outspend every smaller group on AI. Yet only about a third of their employees use AI weekly, and their leaders report 8.8 hours saved per worker per week, compared with nearly 12 at the smallest firms.11 More spending doesn't mean more adoption, and better tracking doesn't mean better results unless the work itself changes.

The broader industry setup also gets little attention. One analyst points out that model providers and hyperscalers benefit when consumption grows, while enterprises benefit only when consumption pays off.12 Steps like visibility and contract flexibility are partly a way to correct that imbalance.

Companies also aren't backing away. BCG found 94% of executives plan to keep investing even without immediate returns, and only 13% of Bloomberg's respondents would cut or pause AI if token prices jumped.67 Demand looks solid. What's changing is how much patience companies have for spending they can't explain.

Where this leaves enterprise AI

The "spree is over" framing gets the dollars wrong and the mood right. Budgets will keep growing through 2026 and likely beyond. But enterprise AI adoption is moving from a race to roll out tools to a test of accountability. The companies that do well will be the ones that can show a CFO what each token bought, not the ones using the most AI. The five steps are a reasonable starting point. They'll matter only if companies combine them with the workflow redesign that, according to McKinsey's research, separates the roughly 6% of high performers from everyone else.6