KPMG: Nearly Half of Executives Scaled Back AI Agents — The Bill Arrived

The AI Agent bill finally arrived. When executives opened the envelope, half of them asked for a refund.

Three Key Takeaways

KPMG's Q2 2026 "Global AI Pulse" survey covered 2,145 senior leaders across 20 countries at organizations with over $50 million in annual revenue. 49% of respondents said they had scaled back or delayed AI agent deployments because operating costs outweighed the benefits. Yet AI remains a top investment priority for 79% of leaders (up from 74% last quarter), with average AI spending holding steady at $188 million. This isn't an exit — it's a recalculation.

The cost explosion stems from a pricing model shift. Most vendors moved from flat subscriptions to usage-based token pricing — every question, every reasoning step, every tool call ticks the meter. When GitHub Copilot switched to usage-based billing on June 1, one user tracked a single day's usage and projected a monthly bill of $180 on what had been a flat $10 plan, driven by one long, tool-heavy session. Agents differ from chatbots precisely because they run multi-step workflows — and every step costs money.

The most damning number: only 26% of enterprises have real-time visibility into AI running costs. A third of global leaders admit limited understanding of AI cost structures, including how token pricing works. It's like driving a car with no fuel gauge at full throttle — you only know the tank is empty when the light comes on. KPMG recommends installing cost meters before scaling, making token economics a leadership literacy, and embedding cost reviews into AI approval processes.

WangDou's Take

Translate this report into plain English: enterprises got sticker-shocked by the AI Agent utility bill.

Last year everyone shouted "Agents change everything!" Pitch decks were gorgeous, demos ran smoothly, but nobody told you that each agent task burns through tokens at tens of times the rate of a normal conversation. An agent queries a database, calls an API, self-checks, runs another round — each step is money. The Copilot case is the most brutal: $10 became $180, a 1,700% increase. That's not a technical breakthrough — that's a telecom bill ambush.

But don't misread this — it's not the AI bubble bursting. 79% of executives are still increasing investment. 22% of organizations have AI in daily operations (up from just 13% last quarter, the largest single-quarter jump KPMG has ever tracked). What's happening is that companies are shifting from "run it first, think later" to "check the bill before you run". Gartner predicts over 40% of agent projects will be killed by 2027 — not because the technology fails, but because most projects never answered one fundamental question: how much does one agent run actually cost?

Source: Forbes