The rapid integration of artificial intelligence into the corporate world is hitting a significant financial roadblock. While initial enthusiasm for AI was boundless, Chief Financial Officers at major U.S. corporations are now facing a difficult decision: whether to allocate capital toward massive AI expenditures or toward expanding their human workforce. This ‘tokens vs. humans’ trade-off is becoming a central theme in boardroom discussions as the true cost of frontier models begins to surface.

The financial strain is more acute than many anticipated. Some enterprise leaders report that annual AI budgets are being depleted within just a few months of implementation. This rapid burn rate is driven by the fact that newer, more advanced AI models are often significantly more expensive per token than their predecessors. Arvind Jain, CEO of Glean, notes that for the first time in business history, technology costs are approaching the level of human labor costs, forcing companies to make direct comparisons between software spend and headcount.

A major driver of these ballooning costs is the inefficient use of high-end models. Currently, an estimated 95% of enterprise AI tasks are being processed by the most expensive, high-tier models, even when simpler, more affordable alternatives would suffice. This lack of optimization means companies are paying a premium for intelligence that isn’t always necessary for every task.

To combat these costs, companies are shifting their strategies toward smarter model routing. Instead of using ‘frontier’ models for every query, businesses are looking to automate the process of sending tasks to the most cost-effective model suited for the job. Matan Grinberg, CEO of Factory AI, suggests that the difference between top-tier models is often negligible for standard tasks, and by implementing better routing, enterprises could potentially achieve up to ten times in savings.