Corporate America raced into artificial intelligence. Now the invoices have landed. Executives once treated the technology as a fast path to savings. They pictured quicker coding, smaller teams, and sharper products. Instead, many leaders now stare at growing software bills and pointed questions from their finance chiefs. AI costs have become a top worry inside boardrooms across the country.
For most of the post-ChatGPT surge, top AI vendors have priced down to grab users. Investors absorbed the difference. Customers got strong tools at friendly rates.
That stretch now looks shaky.
Kevin Simback of startup incubator Delphi Labs describes those early days as the era of “subsidized intelligence.” Investors shouldered the bills while companies tried the tools widely.
“But the tides are beginning to turn,” Simback warned.
So the debate has shifted. Businesses still crave the technology. Yet nobody wants a shock invoice. AI costs now sit at the heart of that tension.
Agents change the cost math
AI agents drive much of the pressure.
A basic chatbot replies to a prompt. An agent acts. It can write code, dig through files, and finish jobs across many apps.
That muscle carries a price. One agent task can spark dozens of steps. Each step burns tokens, the billing unit behind most AI products. So a single complex job can swallow far more tokens than one short message.
AI costs climb quickly once thousands of workers lean on these tools daily.
Goldman Sachs Research projects that agentic AI could push token use up 24-fold by 2030 as adoption spreads.
“Especially in developer circles, the cost to use AI for things like coding has grown exponentially,” said Mark Barton of tech consultancy Omniux. “All the costs are really starting to skyrocket.”
Uber’s warning hits home
Uber now stands as the clearest case.
The ride-hailing giant burned through its full 2026 AI budget in just four months, mostly on Anthropic’s Claude Code. According to Uber’s first-quarter report, 95% of its engineers used AI tools each month, and agents wrote more than one in 10 lines of code.
Uber President and Chief Operating Officer Andrew Macdonald called the disclosure a “head-exploding moment.” He then questioned whether the spending shipped real value to riders or drivers.
“That link is not there yet,” Macdonald said. “Maybe implicitly there’s more that is getting shipped, but it’s very hard to draw a line between one of those stats and ‘Okay now we’re actually producing like 25% more useful consumer features.'”
Still, Uber has not walked away from artificial intelligence. CEO Dara Khosrowshahi said the company slowed hiring to offset its rising spending. AI costs, in short, now demand proof.
Meta moved toward caution. The company once urged staff to use plenty of tokens. Then Chief Technology Officer Andrew Bosworth pushed back in an internal memo.
“Nobody should be using AI tools just for the sake of using them,” Bosworth wrote, according to The Wall Street Journal.
Tokenmaxxing triggers budget shock
A fresh habit has surfaced inside some firms: tokenmaxxing. Workers fire off huge volumes because the tools feel quick, simple, and always on.
The bills add up fast. AI costs can outrun the value before anyone notices.
“In some cases people are seeing the cost of tokens exceed the cost of the employee within a month or two of use, just because they’re using it too much,” said analyst Jack Gold of J. Gold Associates.
As a result, AI costs have dragged finance teams into software calls that once stayed with engineers. Budgets now carry caps, usage tracking, and return-on-investment reviews.
Companies hunt for cheaper models
The pushback does not signal a retreat from artificial intelligence. Instead, many buyers want leaner ways to run it.
Some firms test open-source models. Others lean toward smaller models tuned for one industry. So AI costs have nudged companies to split their work. They route simple jobs to cheap models and save premium tools for the hard ones.
Adrian Balfour of consultancy Enverso said the price gap can stun buyers.
“The big, large monolithic model, it’s US$15 per million tokens, but you can get that down to like five cents if you use the smaller mini model,” Balfour said.
The question has flipped. It no longer asks which model dazzles. It asks which model finishes the job at the right price.
Big labs keep their edge
The shift has not killed demand for premium systems. Power users still want the strongest models, and AI costs factor into every choice.
John Belton, a portfolio manager at Gabelli Funds, expects top users to keep spending for quality.
“The most advanced users” will always pay for the best, Belton said. “It’s a growing pie.”
That outlook helps explain why OpenAI, Anthropic, and other leaders draw investor interest. But customers now want discipline alongside speed.
The next corporate test
The next phase will reward careful operators. Executives may set token caps. Finance chiefs may track model spending like cloud bills.
The winners will not just adopt artificial intelligence. They will govern it.
So AI costs have climbed from a back-office line item to a boardroom flashpoint. The technology still promises big gains. Yet corporate America now wants the math to hold up.
What do you think? Should companies pump the brakes and challenge the spending, or keep investing hard despite rising costs? Please drop your comments below.