Something funny happened on the way to the automated future. Executives spent two years telling boardrooms that AI would slash headcount, boost margins, and practically pay for itself. Sam Altman called it “intelligence as a utility” — flip the switch, pay pennies. Turns out, the utility bills arrive with surprising force when every prompt, query, and background task runs through a meter.
Here’s how the mechanism works. AI vendors like OpenAI, Anthropic, Microsoft, and Salesforce have shifted from flat-rate contracts to usage-based pricing, charged per “token” — the chunks of text and data processed with every interaction. Early pilot programs were cheap, often subsidized. Scaled daily use across thousands of employees is not. GitHub Copilot’s move to full usage-based billing on June 1, 2026 made the math brutally clear: one developer’s projected monthly cost jumped from roughly €67 to €966, according to employer branding analysis. These escalating costs echo the ambitions behind the Stargate Project, where infrastructure investment is measured in the hundreds of billions.
The cost visibility picture is equally alarming. Consider what KPMG’s survey of 2,145 senior leaders across 20 countries actually found:
Cost visibility gaps:
- Only 26% of organizations have real-time, comprehensive visibility into AI spending.
- Another 22% discover costs only after the bills land — classic sticker shock.
- A further 29% of senior leaders admit they struggle to understand their own AI operating costs as deployments scale.
Spending blowouts:
- One unnamed enterprise reportedly ran up $500 million in a single month on Anthropic’s Claude after failing to set usage limits on employee licenses.
- Uber burned through its entire 2026 AI budget in just four months.
- Separately, 78% of IT leaders report experiencing unexpected charges tied to consumption-based AI pricing.
“Many CFOs are going to see their Anthropic bill and freak out this quarter,” warned Gil Luria, head of tech research at D.A. Davidson.
Meanwhile, the layoffs already happened. Meta cut 8,000 employees to fund AI infrastructure — infrastructure that now carries its own escalating, usage-driven tab. Even as ROI erodes, AI remains useful to management as a threat: a negotiating chip against workers who can be told their replacement is one deployment away.
The Hangover Begins
Nearly half of organizations have already scaled back AI rollouts as financial reality collides with the hype cycle.
KPMG found that close to half of organizations have rephased AI deployments after costs outweighed expected value. Steve Chase, KPMG’s global head of AI, describes the technology as “a new resource that needs to be managed,” noting that clients are exhausting token and cloud budgets within months. Ali Ansari, CEO of Micro1, told Axios the pullback represents a “healthy shift,” adding that “AI truly works reliably for only a narrower set of tasks than marketing suggests.”
Think of it like subscribing to every streaming service simultaneously — no single rewatch justifies the aggregate bill. Except in this scenario, you already let the cable guy go. The tools that were supposed to replace workers are now competing with those same workers for a share of a shrinking budget — much like consumers who are paying too much without realizing it — and the meter keeps running either way.
