Untracked AI Token Spending Could Trigger Earnings Surprises

Hidden AI costs accumulating in corporate budgets

Venture capitalist Chamath Palihapitiya issued a stark warning this week that corporate executives have lost track of AI spending inside their organizations, setting up potential earnings surprises when token costs hit income statements.

In an interview with CNBC on Tuesday, the Social Capital founder and CEO of enterprise software company 8090 said most chief executives and CFOs remain unaware of the scale of "tokenmaxxing" happening within their companies. He predicted some will discover the problem only when quarterly earnings fall short and they ask finance teams what went wrong.

Tokenmaxxing refers to corporate policies encouraging maximum AI tool usage under the assumption that higher consumption drives productivity gains. Because AI vendors charge by the token—the discrete data units models process when generating responses—unchecked usage translates directly into enterprise costs that may not be properly tracked or budgeted.

The commodity shift in AI models

Palihapitiya also noted that competitive dynamics in the AI market are shifting rapidly. Lower-cost alternatives from Meta and Google now deliver performance within 80 to 95 percent of leading models for most business applications, he said. The performance improvements between successive model releases have become incremental rather than transformative, resembling iPhone generation updates more than breakthrough advances.

This convergence marks a departure from earlier periods when new model releases represented dramatic capability jumps. The venture capitalist described previous launches as moving "from kerosene to jet fuel," a contrast no longer applicable to today's more mature market.

Why it matters

This warning arrives as major enterprises pull back from aggressive AI adoption strategies. Uber exhausted its annual Claude Code allocation ahead of schedule and imposed a $1,500 per-developer spending cap. Microsoft restricted employee access to the same tool. Meta CTO Andrew Bosworth told staff in April that token volume alone does not measure business impact. Palantir CEO Alex Karp recently argued that OpenAI and Anthropic have mispriced their services and that customers are extracting minimal value from token spending.

The pattern suggests a reckoning is underway. Companies that embraced AI tools without establishing cost controls or ROI frameworks may face unexpected financial impacts. For CFOs, the challenge is that AI spending often disperses across departments through individual employee subscriptions and API calls rather than flowing through centralized procurement, making it difficult to monitor until costs aggregate into material amounts.

Palihapitiya disclosed in March that his own company's AI spending was trending toward more than $10 million annually, illustrating the scale these costs can reach even at mid-sized enterprises.

These details were first reported by Qz based on Palihapitiya's CNBC interview.