CEOs Now Expect AI Costs, Not Wages, to Reshape Their Margins

For most of the last decade, the line that kept CFOs awake was payroll. Wages were the biggest controllable cost, and automation was pitched as the way to bend it. The newest read on chief executive thinking suggests that framing is flipping. The cost leaders now worry will move their margins is not the people doing the work. It is the AI they bought to help them do it.

KPMG’s 2026 CEO Outlook, released October 15 and drawing on more than 1,300 CEOs of large companies worldwide, including 400 in the U.S., found that 81% of U.S. CEOs expect rising AI costs to reshape margins more than labor costs over the next five years. About half say managing AI implementation costs is mostly or very challenging. Tim Walsh, KPMG U.S. Chair and CEO, described leaders as acting “with conviction and a heightened sense of urgency.”

Spending Keeps Climbing While Payoff Stays Modest

The conviction shows up in budgets. In the same survey, 84% of U.S. CEOs call AI a top investment priority for the next three years, and 65% plan to spend at least 10% of their capital budget on AI in the next year. Most still expect the payback to take time: 68% expect returns in one to three years, 14% within six to twelve months and 13% in three to five years.

The results so far are more measured than the spending. Only 9% of U.S. CEOs say AI agents are transformational for their organization. Another 39% report significant efficiency or growth improvements from agents, and 43% describe moderate impacts from targeted use cases. That is real progress, but it is a long way from the step change that a double-digit share of the capital budget implies. Notably, the same CEOs are not planning to shrink: 89% expect headcount to grow over the next three years, so AI costs are arriving on top of the wage bill rather than in place of it.

The wider market explains why the bill keeps rising even for companies that never sign an AI contract. As CIO reported, Gartner expects worldwide AI spending to reach $2.7 trillion in 2026, up 49.5%, with AI software spending growing 60.2% and AI agents and assistants rising 77.3%. Gartner Distinguished VP Analyst John-David Lovelock said that “every dollar is going to be an AI dollar in one way or another,” because AI is now embedded in the laptops and enterprise software companies already buy.

Why AI Is Harder to Budget Than Payroll

Payroll is predictable. Headcount, salary bands and benefits rates are known months ahead. AI costs behave more like a utility bill that moves with usage, and usage grows every time a team finds a new prompt, a new agent or a new feature switched on inside an existing tool. That is why the finance function is scrambling to catch up.

The State of FinOps 2026 report, based on 1,192 respondents, found that 98% of practitioners now manage AI spend, up from 63% in 2025 and 31% in 2024. AI cost management is the single most desired skillset across organizations of all sizes. The challenges respondents name are telling: pricing models differ widely between providers, so costs are hard to see; assigning AI spend to business units is harder than allocating traditional infrastructure; and because many investments are exploratory, returns are difficult to define. One practitioner summed up the state of play: “Is your AI providing value? No one can answer that question yet.”

The report also notes that many organizations are being asked to self-fund AI through optimization savings elsewhere. That puts operations leaders in an awkward spot. They are expected to find the savings that pay for AI, while the cost of AI itself is still poorly measured.

Value Over Token Budgets Only Works if You Can Measure Value

The most revealing number in the KPMG data may be this one: 88% of U.S. CEOs say they prioritize realized AI value over strict token budgets. In plain terms, leaders are willing to let the AI bill run as long as the business gets something measurable back. That is a reasonable position, but it shifts the burden onto measurement. If value is the control, someone has to be able to show what the value is.

Most organizations struggle here for a simple reason. The work AI is meant to speed up was never costed in the first place. Hours lost to rebuilding the same presentation, reconciling data across disconnected tools or chasing status updates rarely appear in a budget. Without a baseline for what that work cost before, there is no credible way to prove what AI saved after, and the token bill becomes the only number on the page.

The AI Cost Control Playbook

• Put AI on the P&L as its own line: Separate AI spend from general software and cloud budgets so leadership can see the trend the way 81% of CEOs now expect to feel it, in the margins.
• Baseline the work before automating it: Measure the hours and cost of the workflow today, so any AI deployment is judged against a real number rather than a vendor estimate.
• Allocate costs to the teams that use them: FinOps practitioners cite allocation as a top challenge. Showback by business unit turns AI from a shared overhead into a decision each team owns.
• Audit the AI you already pay for: With AI embedded in existing software, some of the bill arrives through renewals. Review what features are switched on and whether anyone uses them.
• Consolidate before you add: Every extra tool adds licensing, integration and context switching. Fewer, better-connected systems make both the spend and the payoff easier to track.
• Tie each agent to a business metric: Only 9% of CEOs call agents transformational. Require every agent in production to report against a named outcome, and retire the ones that cannot show one.

The CEOs in KPMG’s survey are not pulling back from AI, and there is little sign they should. But their own forecast is a warning: if AI costs are about to matter more to margins than wages, AI deserves the same discipline as payroll. The companies that build that discipline now will be the ones that can prove their AI is paying for itself.