There is a pattern playing out right now inside large companies, and it matters for any business operator thinking about AI spending.
First, leadership mandates AI adoption. Staff are told to use it, tracked on it, and in some cases threatened with career consequences if they do not. Then the bills arrive. Then the scramble begins.
Accenture is the clearest example this week. Leaked audio from an internal meeting reveals the firm is now trying to figure out how to stop non-technical staff from draining its AI budget on tasks like converting PDFs into presentation slides. This is the same company that, not long ago, told senior employees they would "risk losing out on promotions" if they did not use AI. The about-face happened in months.
Accenture's own agentic AI strategy lead put the problem plainly in that internal meeting: spend is becoming "very unpredictable," and executives at the CFO, COO, and CIO level are asking whether they are actually getting value.
This is not a one-company problem. Uber burned through its entire AI budget for coding by April of this year. One company reportedly spent half a billion dollars in a single month after failing to set any usage limits. Microsoft is winding down a major AI tool contract partly over costs. Salesforce's CEO said his company's bill from a single AI provider will hit around $300 million this year, and he openly wishes there were a smarter way to decide which tasks actually need the expensive tools.
The core issue is how enterprise AI is priced. Most tools now charge per token, which is roughly a unit of text processed. Every question an employee asks, every document uploaded, every automated task the system runs all add up. When prices per token fell sharply over the past year, companies assumed the cost problem was solved. It was not. Usage volume rose far faster than prices fell, so total bills kept climbing.
A research firm tracking developer AI usage found that the heaviest users were about twice as productive as light users but spent ten times as many tokens to get there. That math does not work at scale.
The broader numbers are sobering. According to Deloitte, only 28% of global finance leaders say they can point to clear, measurable value from their AI spending. Nearly half of business leaders expect it will take up to three years just to see returns from basic AI automation. Meanwhile, AI has become one of the largest and fastest-growing items in corporate technology budgets, with some firms reporting it now consumes up to half their IT spend.
What is happening right now is a correction, not a collapse. Companies are not abandoning AI. They are being forced to do what they should have done from the start: treat it like any other operational cost, with limits, oversight, and a clear reason for each use.
The lesson for any business operator is straightforward. If you are deploying AI tools across your team, decide now which tasks justify the cost and which do not. Using AI to write a summary of a three-page document costs real money. Using it to automate a process that used to take a team three days is a different calculation entirely. The companies that are in trouble right now are the ones that handed out AI access like office supplies and assumed productivity gains would follow automatically.
They did not. And the bill still came.