KPMG: 29% of senior business leaders struggle to understand and control AI costs

KPMG’s Global AI Pulse for the second quarter of 2026 found that nearly a third of corporate leaders cannot understand or control what their AI systems cost to run. 

Companies that laid off employees and restructured their budget in favor of what they expected to be cheaper AI technology are now being charged by the token, leading to massive AI spending bills. 

Is AI too expensive for companies to use? 

KPMG polled 2,145 C-suite and senior business leaders across 20 countries for its quarterly Global AI Pulse survey. The results show that 29% of executives struggle with operating costs, especially as AI use grows past the testing stage to using it on a large scale. 

A separate third of respondents named their own thin grasp of AI economics as an obstacle to rolling out AI agents at all.

The main reason for this struggle is that AI companies are now charging based on usage. Providers, including Anthropic, OpenAI and GitHub, previously operated with fixed subscriptions, but now each prompt and response is charged per token. 

A token is a small piece of text or code that a model processes, and because it does not map cleanly to a single word, it is very hard to predict the cost of a project before the invoice lands.

In a related KPMG survey, only 26% of companies said they had a comprehensive real-time view of AI spending. Half had partial oversight. The remaining 22% either had none or learned what they had spent only when the bill showed up.

Steve Chase, KPMG’s global head of AI, stated that artificial intelligence technology is “a new resource that needs to be managed.” He also said his firm is currently advising clients who have drained a full year of their token and cloud budget in a matter of months. 

What are companies spending on AI? 

An unnamed enterprise reportedly ran up a bill of roughly $500 million on Anthropic’s Claude in a single month after failing to set any usage limits for employees. When GitHub Copilot moved fully to usage-based billing on June 1, 2026, one developer’s projected monthly cost climbed from about €67 ($72) to €966 ($1040). 

Companies are responding to these exorbitant costs by pulling back. KPMG found that almost half of organizations have slowed down or delayed the rollouts of their AI products after the costs outran the value they expected. 

Cheaper high-fidelity models are now the fastest-growing influence on AI strategy, up seven percentage points since the first quarter. 

Cryptopolitan reported that Uber (NYSE: UBER) exhausted its entire 2026 AI coding budget by April, four months into the year, and now restricts its engineers to $1,500 per tool each month. Meta (NASDAQ: META) has told roughly 6,000 employees that it is building an internal gateway with budgets and spending alerts to curb “tokenmaxxing.”

Amazon (NASDAQ: AMZN) dismantled the leaderboard it created to encourage AI use in late May after staff gamed it with pointless activity.

Cryptopolitan recently reported that Coinbase’s CEO, Brian Armstrong, proposed using cheaper open-weight models to control the rising AI costs, but that suggestion drew criticism and concerns due to security implications.