Corporate finance chiefs are scrambling to track their companies’ artificial intelligence (AI) consumption to avoid severe budget shocks, as major tech vendors aggressively pivot toward usage-based pricing measured by tokens.
The shift to token-based metrics — the foundational unit of measurement for AI computing — is upending traditional corporate budgeting. Chief financial officers (CFOs) accustomed to predictable, flat-rate software subscriptions are discovering that building AI agents and funding ambitious tech projects yield highly volatile costs that defy standard financial modeling.
A paltry 26% of companies maintain a comprehensive view of their AI expenses, according to an upcoming survey from professional services firm KPMG. Meanwhile, 50% possess only partial visibility, and 22% report having no insight into costs until after billing occurs.
Executives and market analysts draw stark parallels between the current AI investment boom and the pandemic-era cloud computing surge, which was ultimately followed by aggressive corporate cost-cutting. As token expenses mount, CFOs are signal boosting a clear message: the era of unmonitored AI experimentation is ending.
“It’s a new resource that needs to be managed that didn’t exist quite that way, and we’re seeing exponential growth,” said Steve Chase, KPMG’s global head of AI. Chase noted that some clients have entirely exhausted their annual token and cloud computing budgets within months, with one organization witnessing a sixfold explosion in token consumption.
To mitigate these risks, AI providers like OpenAI and Anthropic, alongside tech giants Microsoft Corp. and Salesforce Inc., are increasingly leveraging metered pricing. Analysts note this model aligns vendor revenue with operational costs while protecting tech companies against seat-reduction strategies by enterprise clients. However, it shifts the financial risk entirely to the buyers.
The financial exposure has triggered varying corporate responses. At fintech firm Affirm, token spending took center stage during its recent annual budgeting process following an overnight, step-function surge in consumption driven by automated coding agents. Affirm has since instituted near-real-time monitoring and weekly executive cost reviews.
Conversely, some corporate cultures have embraced “tokenmaxxing,” a term describing the practice of exhausting computing resources purely to appear AI-forward. Gil Luria, head of technology research at D.A. Davidson, warned that less vigilant executives “are going to see their Anthropic bill and freak out this quarter.”
Other firms are tightening controls to curb waste. Life360 Inc. is redesigning its AI agents and implementing specialized tools specifically to reduce token consumption. Corning Inc. has restricted the number of AI tools available to staff, focusing its capital on a select group of high-priority projects. Reckitt slowed its marketing AI rollout after discovering that employee usage plunged after a few weeks, adjusting its return-on-investment timelines accordingly. And Amer Sports Inc. is intentionally pacing its back-office AI integration to avoid spiraling costs.
“It depends on the level of adoption in building skills and elevating work,” IBM Fellow Bala Rajaraman said in an interview. “Token-maxxing is a definite concern but a bigger obstacle is adoption.”
