KPMG’s Q2 2026 Global AI Pulse survey found that 29% of senior business leaders cannot understand or control their AI operating costs as deployments scale. The survey, which polled 2,145 executives across 20 countries, highlights significant challenges faced by companies as they transition to large-scale AI use.
Many companies have begun capping spending on AI technologies after costs exceeded their expectations. Firms including Uber, Meta, and Amazon have scaled back their AI rollouts due to rising expenditures. One unnamed enterprise reportedly incurred a $500 million charge in a single month for using Anthropic’s Claude without setting usage limits.
The shift from fixed subscription models to usage-based billing by AI providers such as Anthropic, OpenAI, and GitHub has contributed to this cost struggle. Respondents indicated that their limited understanding of AI economics is a barrier to effectively deploying AI solutions.
Steve Chase, KPMG’s global head of AI, stated that AI technology is “a new resource that needs to be managed.” He noted that some clients have depleted their entire annual token and cloud budgets within months due to unanticipated costs.
Only 26% of companies reported having a comprehensive real-time view of AI spending. Half of the surveyed executives had partial oversight, while 22% lacked any visibility into their AI expenditures until they received bills.
KPMG also found that nearly half of organizations have delayed or slowed the rollout of their AI products as their costs have outpaced the anticipated value. Cheaper high-fidelity models are gaining traction, increasing by seven percentage points since the first quarter.
In an effort to manage rising costs, Uber exhausted its AI coding budget for 2026 by April, limiting engineers to $1,500 per tool per month. Meta has informed approximately 6,000 employees that it is developing an internal system to monitor budgets and spending. Amazon dismantled a leaderboard designed to encourage AI use after staff exploited it for trivial activities.
Coinbase CEO Brian Armstrong suggested using open-weight models to mitigate AI expenses, though this proposal has faced criticism due to potential security risks.
