Chamath Palihapitiya warns AI tokenmaxxing will hurt earnings

Chamath Palihapitiya has a new favorite word, and CFOs everywhere should probably learn it. The Social Capital founder and CEO of AI startup 8090 warned on July 14 that “tokenmaxxing,” the corporate tendency to maximize AI token consumption without tracking returns, is quietly bleeding companies dry.

His prediction: earnings per share misses of “a few pennies” as AI usage costs pile up in places finance teams aren’t watching closely enough.

The hidden tab nobody’s checking

AI tokens are essentially consumption credits for large language models from companies like OpenAI and Anthropic. Every API call, every prompt, every automated workflow burns through them. And unlike a SaaS subscription with a fixed monthly price, token-based pricing scales with usage, often unpredictably.

Palihapitiya isn’t speaking purely as an outside observer here. His own startup, 8090, was on track as of March 2026 to spend over $10 million annually on AI usage credits.

He’s not the only one sounding alarms

Palihapitiya’s concerns echo those of Palantir CEO Alex Karp, who has also raised questions about the sustainability of token-based AI pricing models.

Palihapitiya’s 8090 just closed a $135 million Series A funding round led by Salesforce in June 2026. He’s simultaneously building an AI company, spending heavily on AI tokens, and warning everyone else that the spending model is broken.

What this means for investors

For anyone watching public equities, Palihapitiya’s warning translates to a straightforward risk: companies that report aggressive AI adoption without corresponding AI cost management frameworks are setting themselves up for negative earnings surprises. The companies most exposed are mid-cap tech firms and enterprises that have rushed to integrate AI across their operations without building the financial controls to match. Large-cap tech companies with their own foundation models, like Google, Meta, and Microsoft, are somewhat insulated because they control the underlying infrastructure.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.