CFOs Are Coming for AI Budgets: What Survives the Cut

The free-spending era of enterprise AI is over. CFOs across the Global 2000 are no longer asking "How do we adopt AI?" They're asking something far more uncomfortable: "What exactly did we spend, and what did we get for it?" The companies that survive the coming budget reckoning will be the ones that built measurement into the foundation — not bolted it on after the fact. A new KPMG Global AI Pulse survey released June 24, 2026, covering more than 2,145 C-suite and senior business leaders across 20 countries at organizations with annual revenues exceeding $50 million, laid out the numbers plainly. Only 7 percent of leaders report establishing AI ROI. Nearly one in four — 24 percent — face active investor pressure to prove value. And 42 percent have only partial visibility into how their AI spending accumulates. This is not a technology problem. It is a governance and accountability problem, and CFOs are now the ones being asked to solve it. ## The Signal Everyone Missed For most of 2024 and into 2025, enterprise AI spending followed a recognizable pattern: centralize the budget, lower the friction, encourage experimentation. The logic was sound. If you make AI tools easy to access and absorb the cost centrally, adoption accelerates and you learn faster. The problem is that it created a textbook spending dynamic — teams consumed the tool while another budget paid the bill. Usage expanded faster than outcomes could be documented. And unlike traditional software licenses with predictable monthly fees, AI runs on token economics: small charges per query, per output, per agent action, per tool call, per retry. One employee using AI to draft a document is trivial to price. Ten thousand employees using AI agents that touch customer data, contract repositories, CRM systems, and internal knowledge bases — each agent potentially calling dozens of tools per task — is not. A Forbes analysis published June 22, 2026, captured the moment precisely. Uber, a company with deep technical sophistication, set a monthly token cap per user after AI spending ran ahead of plan. Amazon, Walmart, Cisco, and Meta all moved to rein in AI tool use as costs strained budgets. These are not companies that lack engineering capability. They are companies that discovered token economics at scale behaves more like cloud compute than software licenses — and treated it accordingly. ## Gartner's Warning Proved Right (Twice) Gartner flagged the coming rationalization in stages. The initial warning: at least 30 percent of generative AI projects would be abandoned after proof of concept by the end of 2025. The sharpened update: at least 50 percent of generative AI projects had been abandoned after proof of concept by the end of 2025, citing poor data quality, weak risk controls, rising costs, and unclear business value. That 50 percent figure deserves careful reading. These were not failed experiments from naive companies. Many were well-resourced pilots at sophisticated enterprises that simply could not answer the question: What is this worth? Not in vision. Not in potential. In dollars this quarter. The KPMG data confirms where the gaps remain. Only 22 percent of organizations have reached a stage where AI is "part of everyday work" — that number is actually up dramatically from 13 percent in Q1 2026, the largest single shift on the AI maturity curve. The momentum is real. But turning momentum into documented financial returns is where most organizations stall. 33 percent of leaders cite limited understanding of usage costs as a key deployment challenge specifically for AI agents. ## What Vendors Are Reading Enterprise AI vendors are not waiting for their customers to figure this out. The vendors that thrive through the budget reckoning will be the ones that give procurement and finance teams a reason to defend the spend. OpenAI added ChatGPT Enterprise tools that give administrators credit usage analytics, consumption tracking by team, adoption pattern visibility, and cost exposure reporting. Microsoft built a comparable management layer around Copilot — admin dashboards for prompt activity, agent engagement, and business impact analysis. AWS added cost allocation tools for Amazon Bedrock, letting engineering teams tag and track model usage by application or business unit. Databricks is adding AI spend limits, safeguards against runaway agent costs, and cross-provider recommendations. The pattern is not accidental. These companies are betting that the next enterprise sale will be won not only on model quality, but on controllability. "Production systems need receipts" is the clearest framing I have heard from a vendor in this space. The era of performance benchmarks as the primary sales motion is giving way to an era of cost dashboards and chargeback reports. ## The CFO Conversation Is Now Technical This shift puts CIOs and CTOs in an unfamiliar position. The CFO conversation used to be strategic and relatively abstract: AI will unlock efficiency, transform operations, drive competitive advantage. That conversation is being replaced by a much more specific one. How much did the AI workflow cost last month? Which team used it most? Did it reduce headcount pressure, accelerate revenue, or improve customer outcomes? Did we pay for the same capability through two different vendors? Which model handled the task — and could a smaller, cheaper model have done it just as well? These are not questions that can be answered with a vision deck. They require infrastructure: cost allocation by business unit, usage tracking at the team and workflow level, model-level attribution, and some framework for calculating cost per outcome — cost per resolved support ticket, cost per reviewed contract, cost per qualified sales lead, cost per shipped feature. The good news from the KPMG data: leaders who built that infrastructure are seeing dramatically different results. Organizations with strong cost visibility are five times more likely to report established ROI — 15 percent versus 3 percent. That is not a marginal difference. That is a structural advantage. ## Accountability Changes Everything The KPMG finding that may matter most sits at the leadership accountability layer. In most organizations, AI ownership is murky. Only 24 percent of leaders say the CEO is accountable for AI-driven business outcomes. Another 29 percent point to the "broader C-suite" — which, in practice, often means the accountability diffuses and no single leader owns the outcome. The data on what happens when that changes is striking. Organizations where the CEO is explicitly accountable for decisions based on AI outputs report: * 60 percent confidence in their AI strategy (versus 22 percent) * 57 percent meaningful business value realization (versus 21 percent) * 14 percent established ROI (versus 4 percent) Clear accountability is not a bureaucratic exercise. It is a forcing function. When a CEO owns the outcome, the questions get sharper, the governance gets real, and the measurement gets built in from the start rather than requested after the fact. ## The Budget Center Migration Coming in H2 Finance leaders will push AI costs back to business units. This migration is already underway at early-moving companies, and the KPMG data s