AI Providers Shift to Usage Pricing as Corporate Token Costs Soar

From unlimited buffet to metered consumption

The economics of enterprise AI shifted dramatically in the first half of 2026. Between February and June, OpenAI, Anthropic, and GitHub each replaced flat-rate subscriptions with usage-based pricing tied to token consumption—the fundamental units measuring AI input and output. The change caught many organizations off guard as bills multiplied.

Coinbase infrastructure executive Rob Witoff watched usage "go parabolic" after Anthropic released its improved Claude Opus 4.6 coding model in February. The crypto exchange now enforces weekly spending caps ranging from $500 to $5,000 per employee based on role and seniority. "Once people understand what's possible, usage takes off on its own," Witoff explained. "Then the focus shifts from 'Are people using AI?' to 'Are they using it well?'"

The pattern repeated across industries. Walmart imposed limits on internal programming tools. Amazon shut down an internal "tokenmaxxing" leaderboard that had encouraged maximum AI usage. By April, Uber had exhausted its entire annual AI budget without corresponding product releases, according to details first reported by Business Insider.

Why it matters

This pricing shift arrives as OpenAI and Anthropic prepare for potentially market-moving IPOs. The companies risk losing enterprise customers to cheaper alternatives—including Chinese models from Deepseek and MiniMax—at precisely the moment they need to demonstrate sustainable unit economics. For enterprises, the change forces long-overdue discipline around AI ROI measurement and strategic deployment rather than experimental proliferation.

Enterprise response: caps, metrics, and cheaper models

A March-April survey of 200 executives by Wakefield Research found 79% were concerned about AI budget cuts due to unclear return on investment. Companies responded with new governance frameworks. Accenture, IBM, Oracle, and JPMorgan Chase backed a "Tokenomics Foundation" to standardize AI budgeting metrics.

A senior Deloitte software engineer told Business Insider that GitHub's June pricing changes are "already wreaking havoc" on developer workflows. Under the new model, a single detailed prompt requiring hours of AI processing could cost over $100. "The cheap 'AI buffet' days are over," the engineer said.

Salesforce CTO Parker Harris acknowledged his company is spending "far more" than planned on Anthropic tokens but said they're avoiding premature constraints. Still, he noted the tension: "We gotta run a business, we're a public company. We can't tell our investors like, 'Yeah, sorry, we gave half of our upside this year to Anthropic so they can go public.'"

The economics behind the shift

OpenAI CEO Sam Altman acknowledged at a recent event that AI budgeting "went from at the beginning of this year, an issue that never came up—people were totally happy with the amount they were spending—to all of a sudden, a huge issue."

GitHub's chief product officer Mario Rodriguez explained that under flat-rate billing, "a quick chat question and a multi-hour autonomous coding session can cost the user the same amount." The company had been absorbing the difference, but that approach became "no longer sustainable."

An Anthropic spokesperson told Business Insider the usage-based model benefits customers through customization—heavy users aren't capped, while light users avoid paying for unused capacity.

Companies are adapting by routing routine tasks to less expensive models. Harness senior vice president Trevor Stuart compared using cutting-edge models for basic work to "taking the Ferrari to the grocery store." Ahmad Awais, founder of coding agent startup Command Code, reported gaining 10,000 customers in 30 days driven largely by demand for cheaper alternatives.

The details were first reported by Business Insider reporters Stephen Council, Charles Rollet, and Polly Thompson.