The cost for businesses to utilize AI has risen substantially, prompted by a shift from flat-fee structures to usage-based pricing by major AI providers gearing up for IPOs. This change, coupled with the increasing popularity and credit consumption of advanced AI agents, has led to numerous companies being surprised by their bills. Uber, for instance, reportedly spent its entire 2026 AI budget by April and has since capped employee AI coding tool usage at $1,500 per month.
A KPMG Global Q2 AI Pulse Survey revealed that nearly half of 2,145 global business leaders scaled back their use of AI agents because the associated costs outweighed the benefits. OpenAI chief Sam Altman noted this as a "huge issue" that "never came up" before 2026, with companies spending their entire annual budgets in the first quarter. Despite these concerns, Goldman Sachs projects a 24-fold increase in global token consumption by 2030, driven by AI agents.
To manage these escalating costs, some companies are implementing stricter controls. Atlassian, for example, has set caps on "tokens" (data units processed by models) for its employees, requiring managerial approval for additional usage. Its CEO, Mike Cannon-Brookes, criticized the "you only live once" approach where companies indiscriminately use the most expensive models. Businesses are also wary of vendor lock-in, where over-reliance on a single provider makes switching difficult if prices rise or performance declines.
Open-source AI models are gaining traction as a cost-cutting solution, offering free usage and local hosting capabilities, which also addresses data security concerns for sectors like banking and telecommunications. A study by a Mozilla Foundation academic estimated that open-source models achieve about 90 percent of the performance of their closed counterparts, potentially cutting AI costs for businesses by up to 70 percent. Companies like Hostinger, a web hosting provider, and Notion are also deploying routing tools to select the most cost-effective model for specific tasks, recognizing that different models excel in various applications (e.g., Anthropic's Claude for coding, OpenAI models for creative writing).
Despite the budget challenges and increased scrutiny of spending, overall AI usage continues to grow, driven by the rapid improvement of model capabilities and the release of new, more advanced models by top AI labs. Companies are striving to balance the innovative potential of agentic models with financial prudence, integrating new technologies while maintaining cost discipline and tying AI expenditure to tangible work outcomes.
