Major technology companies are imposing strict limits on the number of AI tokens employees can use to manage soaring operational costs [1, 2, 3].
This shift signals a transition from the unrestricted experimentation phase of generative AI to a period of fiscal discipline. As companies integrate AI agents into core workflows, the recurring cost of token consumption has become a significant threat to corporate profitability.
Meta has introduced hard caps on AI token consumption across the company to keep costs from spiraling out of control, a spokesperson said [2]. Before these caps were applied, Meta's AI token spending grew by roughly 250% year-over-year [2].
The trend is also affecting specialized AI firms. Winston Weinberg, CEO of Harvey, said token usage at the company jumped from roughly one trillion tokens a month to about 12 trillion, a 12-fold increase in just one year [1].
Industry reports from June 2026 indicate that average enterprise AI token budgets are being reduced by 30% to 40% [3]. This phenomenon, described by some as the "tokenpocalypse," has forced companies to quietly throttle AI capabilities as bills pile up [3].
Some organizations are targeting specific high-cost functions to save money. The Gizmodo editorial team said they are now throttling the number of tokens that can be generated by PDF-processing AI tools to stay within budget [4].
While some analysts suggest that the proliferation of AI agents will naturally drive total spend higher, many firms are prioritizing immediate cost-cutting to offset the inflation of model pricing [3, 5].
“Our token usage jumped from roughly 1 trillion tokens a month to about 12 trillion”
The move toward token capping indicates that the initial 'honeymoon phase' of AI adoption has ended. Companies are discovering that the scalability of AI is tethered to a linear cost increase in token consumption. This shift will likely drive a surge in demand for smaller, more efficient local models and a move away from expensive general-purpose APIs to maintain margins.
