Company Boards Blindsided by Rising AI Bills, Uber Capped Its Budget in April

Companies that rolled out artificial intelligence expecting quick savings on headcount are starting to receive bills they didn't see coming. A new KPMG survey of more than two thousand senior executives shows that nearly a third of them cannot explain why the cost of maintaining and using AI in their organizations keeps rising. The reason is that vendors such as Anthropic, OpenAI, and GitHub have shifted from flat-rate subscription plans to billing based on actual token usage.

Where the Surprise Came From

In the early phase of the AI boom, many companies could count on preferential agreements under which model developers partially covered the enormous costs of running large language models. That arrangement, however, changed steadily as vendors shifted an increasing share of the real computing costs onto customers. The result is a situation where AI bills are growing faster and less predictably than companies budgeted for in 2026.

As usage-based pricing models become more common, many organizations are still only building the capabilities needed to forecast, monitor, and manage AI spending. - KPMG

Companies Respond With Budget Cuts

Some large organizations are responding to rising bills with hard spending caps. According to reports from June 2026, Uber burned through its entire AI budget allocated for all of 2026 by April, and has since limited employees to $1,500 a month in token spending on individual tools. Amazon, Walmart, Cisco, and Meta have taken similar steps, cutting internal budgets for AI tools and redirecting teams toward cheaper, less resource-intensive models.

At the same time, nearly half of the organizations surveyed by KPMG admitted they had to delay AI deployments when costs exceeded the expected business value. That's a sign that the wave of automation meant to deliver savings on hiring has, in many companies, collided with a computing infrastructure bill nobody had precisely estimated beforehand.

Cheaper Models Gain Ground

In response to cost pressure, cheaper but still sufficiently capable models are gaining importance. According to data cited in the report, the influence of cheaper, high-quality models on companies' AI strategy rose by 7 percentage points compared to the first quarter of 2026. Companies are increasingly checking where AI actually creates business value and where it remains a costly experiment without a measurable return.

What This Means for Polish Companies

For companies in Poland that are just starting to deploy AI agents or planning larger rollouts in customer service or software development, the story of the American giants is a practical warning. As vendors shift to usage-based billing, the cost of a pilot involving a dozen or so people may not scale linearly to the cost of a company-wide rollout, and without forecasting token usage in advance, an unpleasant surprise on the invoice is easy to come by. Finance and IT departments should jointly set spending limits on AI models before the vendor sets them instead, in the form of a bill.

Higher operating costs could also slow the pace of AI adoption among smaller Polish companies that, unlike corporations the size of Amazon or Uber, have not negotiated individual, preferential rates with model vendors.

Sources: AI bills are baffling the C-suite after shift to usage-based pricing (theregister.com), Artificial intelligence was supposed to replace workers for free (pcformat.pl)