Uber burned through its entire 2026 AI coding budget by April. Microsoft, months after enabling them, revoked its developers’ Claude Code licenses across several key product divisions, per The Verge. Meta quietly took down the internal tokenmaxxing leaderboard its own employees had built. The FOMO phase of enterprise AI procurement is ending, and the FinOps phase is starting.

J.R. Storment, executive director of the FinOps Foundation, says companies began calling him in April and May to report they were “3x over their entire 2026 token budget” four months in. The conversation he used to have was “tokenmaxxing and go fast.” The one he has now is “we need guardrails, how do we control this?” One unnamed company, per Storment, woke up to a $500 million Claude bill after forgetting to set employee usage limits. Priceline’s routine Cursor contract came back at renewal four to five times more expensive.

The underlying math is Jevons paradox dressed in an Nvidia keynote. Bain analysts say token costs halved between December 2024 and December 2025 while tokens consumed grew 450%. The Silicon Data Token Expenditure Index pegs LLM spending as having doubled since late last year, even as the price of a single token has fallen more than 90% since 2023. Cheaper inputs, bigger bill.

The cultural artifacts are the giveaway. Amazon employees, the Financial Times reported, were spinning up agents to perform meaningless tasks because managers used token-usage stats to assess performance, prompting an internal note asking staff to “please don’t use AI just for the sake of using AI.” That’s the dot-com “eyeballs” KPI ported to 2026, with the same downstream consequences.

Now the messaging is pivoting. Salesforce CEO Marc Benioff told investors his company’s Anthropic bill will run about $300 million this year and that Salesforce is building a “smart router” to decide which queries actually needed the most expensive models. Sam Altman, at a recent OpenAI event, called token usage among customers promised big productivity gains “a huge issue.” GitHub Copilot tied pricing to token use at the start of June. The Wall Street Journal reports OpenAI is weighing token-price cuts to peel customers off Anthropic.

In March, Nvidia’s Jensen Huang said he’d be “deeply alarmed” if a software engineer making $500,000 wasn’t spending $250,000 a year on tokens. The CFOs reading their April invoices appear to have reached a different conclusion.

Sources