Corporate finance chiefs spent the past few years being told artificial intelligence would save their companies money. Many are now working out why it keeps costing more than planned.

Finance executives are scaling back AI spending and consolidating budget authority within a central function, according to Bloomberg. A wave of surveys published between September and October 2026 indicates the squeeze is widespread.

The Meter Is Running

The core problem is pricing. Many AI vendors have shifted to usage-based token pricing, which bills companies for how much processing their models actually perform. Tokens—the small units of text that models read and generate—mean a bill grows with every query, summary, or line of code a company runs. Unlike a fixed software license, that makes monthly costs difficult to forecast, and the unpredictability shows up clearly in the data.

In Pigment's Q3 2026 CFO Index, 83% of respondents said consumption-based AI costs came in above their expectations. A U.S. Bank survey released in September 2026 found that 51% of US finance leaders said their AI spending over the previous year exceeded budget.

Deloitte's September 2026 research adds a forward-looking warning: 60% of finance leaders expect AI costs and complexity to rise substantially through 2027. A separate 2026 report from Mavvrik found that 25% of firms delayed or canceled AI projects because of unexpected costs.

Big Names Reach for the Brakes

Some of the most aggressive AI adopters have already started rationing. Uber exhausted its entire 2026 AI budget by April 2026, roughly a third of the way through the year. Reuters reported in June 2026 that Amazon and other companies had put spending caps in place, while Cisco has moved to curb the use of non-essential AI tools.

In response, companies are adopting stage-gate funding, which releases money in tranches only after a project hits defined milestones. Finance teams are also insisting on measurable return on investment before approving further spend. Centralized budget management—the shift Bloomberg highlighted—means individual departments have less freedom to sign up for tools on their own.

Still Believers, Just More Skeptical Ones

Finance leaders in Pigment's index rated their confidence in AI's value at an average of 8.2 out of 10, and U.S. Bank found that 69% of finance leaders reported commercial benefits from their AI investments.

CloudZero's FY2027 survey captures the contradiction best. Some 87% of finance leaders plan to increase AI budgets for the coming fiscal year, and 45% plan increases of more than 20%. Yet in that same survey, 61% admitted their current AI spending exceeds what they can justify by outcomes.

What This Means for AI Vendors and Their Customers

For companies selling AI, customers who once experimented freely are now asking for proof, and finance is increasingly the department signing off. Usage-based pricing helps providers capture value from heavy users, but it also makes their product the line item a CFO is most likely to cap.

The CloudZero data suggests overall budgets are still growing. But if budgets are managed centrally and gated by milestones, spending may flow to a smaller set of proven tools while experimental projects get squeezed out—consistent with the 25% of firms Mavvrik found delaying or canceling work.

The key signal to watch is whether usage-based costs stabilize as companies get better at forecasting them. If the 83% surprise rate in Pigment's index stays high into next year, the spending caps at Uber, Amazon, and Cisco could start to look less like outliers and more like standard corporate policy.