For two years, the pricing philosophy of the AI industry could be summarized in one word: more. Then the bills came due. AI companies are now raising API prices and restricting internal token usage, according to a Business Insider report, ending the period of effectively unlimited consumption and pushing the sector into what the report calls a “calorie counting” phase.
The shift has two engines. On the cost side, the consumption boom of early 2026 stretched providers’ capacity: inference is expensive, and the biggest models consume compute at a rate that makes generosity unsustainable. On the demand side, enterprises that adopted AI at scale have begun budgeting like buyers of any other input, asking what a token costs and whether the spend is justified. Both forces point in the same direction: toward metering.
The “calorie counting” label captures the new discipline. Model providers are tracking every token the way a dieter tracks calories, because every token is a cost. The report describes internal dashboards at the major labs showing token consumption per employee and per team, with budgets set the way marketing budgets are set, and exceeded the way marketing budgets are exceeded.
OpenAI and Anthropic, the two companies that defined the generous era, are both tightening. API prices are rising, internal usage at the model developers is being capped, and employees at the companies themselves are being asked to ration their access, according to the report. The details are the kind that never made it into the marketing: the era of treating tokens as free is over.
The change is visible to users as well as customers. Anthropic’s highest-tier consumer plans now carry weekly caps, and a federal lawsuit filed this week alleges those caps are stricter than the marketing implies. OpenAI has signaled similar adjustments on its own consumer tiers, according to the report. The limits that used to be footnotes in the terms of service have become the product.
Then comes the contradiction. The Wall Street Journal reported the same day that OpenAI is weighing significant price cuts to win back users. Price increases on the API side, price cuts on the consumer side: the two moves target different markets, but together they reveal an industry that has not found its equilibrium price.
The contradiction is not new to technology markets. Companies routinely price to acquire users in one segment while extracting margin in another; the unusual part is the speed with which AI pricing is swinging between the two, a sign that the industry is still measuring its cost curves in real time.
The tension is structural. Consumer AI is a scale game with thin margins and fierce competition; enterprise AI is a value game where customers pay for outcomes. OpenAI and Anthropic are being pulled in both directions at once: raising prices where demand is inelastic, cutting them where rivals circle. The result is a market sending signals in opposite directions.
The shift has historical echoes. Every subscription business eventually meets the meter: cloud computing, mobile data and streaming all moved from generous early pricing to metered or tiered models once the customer base was locked in, analysts said. The AI industry is running the same playbook, compressed into a shorter cycle and a more visible one, because every price change is now debated in public.
The underlying economics have not improved as fast as usage has grown. While inference costs have fallen steadily, the models have grown more capable and more expensive to run, and agentic workloads can consume a hundred times the tokens of a simple chat. The free tiers that fueled the boom were a subsidy; the industry is now deciding who pays for the party.
For consumers, the end of the unlimited era means reading plan terms more carefully. For developers, API price increases change the unit economics of products built on top of the big models. For the model companies, the tightening is the difference between growth and viability. The industry’s leaders have said inference costs will keep falling; the question is whether the price cuts arrive before the usage caps drive customers away.
Analysts who follow the industry said the current combination of caps, hikes and discounts is the visible churn of a market discovering its cost curve in real time. The early phase of AI, financed by venture capital and strategic ambition, could afford to underprice compute. The mature phase cannot.
For users, the era of unlimited everything ends with a counter on a screen. For the industry, it ends with a pricing question that has no obvious answer yet. The companies that set the right price, high enough to cover inference costs and low enough to keep users, will define the next phase of the market. So far, the signals point in both directions at once.
