Companies are now spending more money on AI than at almost any point in history. Gartner expects worldwide AI spending to reach 2.59 trillion dollars in 2026, a jump of 47% from the year before.
Yet a separate Gartner survey of chief information officers found something uncomfortable sitting under that spending. Out of over 11,000 CIOs surveyed, most admitted they struggle to pick which AI projects are even worth doing, and 59% of AI projects never make it into actual use.
This is not a small hiccup. A study from MIT found that 95% of company AI pilots failed to produce any measurable financial return, despite tens of billions of dollars already invested. Put those two studies side by side and a pattern shows up: businesses are very good at starting AI projects and very bad at finishing them in a way that shows up on a balance sheet.
Part of the problem is that the tools available right now only measure spending, not results. Anthropic, the company behind the Claude chatbot, just added spend alerts and usage dashboards for its business customers, joining a wave of similar tools from other AI vendors. These tools are genuinely useful for stopping a runaway bill, but they answer the question "how much did we spend" and never touch the question "did it work."
This gap has a direct parallel in the recent history of business technology. When companies first moved their computer systems to cloud services over a decade ago, the same problem showed up: bills were unpredictable and nobody had a clean way to tie spending to results. That mess is what created an entire profession dedicated purely to matching cloud costs against business value, and AI spending is heading down the exact same road, just much faster.
The people who sit above CIOs have already noticed. A recent survey found that 92% of chief financial officers feel direct pressure to prove AI is paying off. Board members are growing impatient with technology leaders who can only report usage statistics instead of results.
That pressure is landing on IT departments at an odd moment. Token prices for AI keep falling, yet total bills keep climbing anyway, because companies are simply using more of it. Automated AI agents, for example, burn through far more usage per task than a simple chatbot question.
The practical fix that experienced technology leaders point to is not complicated, even if it takes discipline. Pick one real business number, such as how long it takes to approve a loan or resolve a customer complaint, and measure it honestly before AI touches that job. Then measure it again afterward, with the finance team involved from day one rather than brought in after the fact.
Businesses that build this habit now will keep growing their AI budgets without a fight. Businesses that keep reporting usage instead of results should expect their AI spending to face the same scrutiny that cloud budgets eventually did, and probably sooner than they think.