Hidden AI token spending threatens corporate earnings, Palihapitiya warns
Untracked corporate spending on AI tokens threatens to trigger unexpected earnings shortfalls as executives lose visibility over decentralized usage.
Chamath Palihapitiya warned on Tuesday that unchecked AI token consumption inside large companies is setting the stage for sudden earnings misses. The venture capitalist and Social Capital founder told CNBC that corporate leadership is largely blind to the scale of these expenses. He expects the issue to surface as a surprise EPS shortfall.
The problem stems from a practice known as tokenmaxxing, where companies encourage maximum AI usage under the assumption that higher consumption drives productivity. Because AI vendors price their services per token—the discrete data chunks models use to generate responses—this unchecked volume translates directly into escalating enterprise costs. Palihapitiya argued that finance chiefs have failed to track this spending adequately.
"CEOs and the CFOs, in my opinion, probably have no idea how much tokenmaxxing is going on inside of their organizations," Palihapitiya said. "I suspect what'll happen is one day you're going to have a miss, and EPS will be off by a few pennies, and the CEO will say to the CFO, 'What happened?'"
His warning aligns with a growing industry backlash against unprofitable AI deployment. Uber exhausted its annual allocation for Claude Code ahead of schedule before capping individual developer spending at $1,500. Microsoft also restricted employee access to the tool, while Meta's CTO Andrew Bosworth told staff in April that token usage alone does not measure impact.
The financial strain is compounded by a shifting competitive landscape. Palihapitiya noted that cheaper alternatives from Meta and Google are now "80 to 95% as good" as premium models for most tasks. The performance gap between new model releases has narrowed to resemble incremental iPhone upgrades rather than transformative leaps. "You're like, 'Oh my God. We went from kerosene to jet fuel,'" he said of previous generational shifts.
This convergence makes it difficult for companies to justify premium vendor pricing. Palantir Technologies CEO Alex Karp recently argued that OpenAI and Anthropic have fundamentally mispriced their services, noting enterprises are extracting little value from their token expenditure. Palihapitiya's comments extend this critique, highlighting the tangible risk these costs now pose to corporate income statements.
Palihapitiya speaks from direct experience managing these expenses. As CEO of enterprise software company 8090, which secured $135 million in a Salesforce-led funding round in June, he noted in March that his own firm's AI costs were trending above $10 million annually.
