Not long ago, corporations were handing out AI model tokens as corporate bonuses, encouraging employees to "burn" them without a second thought. Today, the same picture looks different: executives are panicking, imposing strict limits, cutting off access, and frantically revising budgets. What went wrong?

In the context of artificial intelligence, a token is the smallest unit of data processed by a language model. On average, 100 words are broken down into 130–140 tokens. They are divided into input tokens (user query) and output tokens (model response). For a long time, this unit was merely a technical detail, but everything changed in March 2023 when OpenAI released GPT-4 and established token-based pricing.

The "era of subsidies" lasted from 2023 to 2025: regular users paid via subscription, while corporate developers paid by volume. Prices were soft, encouraging mass adoption. However, starting in the spring of 2026, all major providers—Anthropic, GitHub Copilot, and others—fully switched to per-minute monetization of queries, citing rising computational costs. The era of the "tokenapocalypse" had arrived.

Jevons Paradox in Action

The most ironic part is that over three years, the average price of an input token dropped by nearly 98%. Yet total enterprise spending on AI skyrocketed by 320%. The average annual corporate budget grew from $1.2 million in 2024 to $7 million in 2026. The reason is an explosive increase in usage volumes, especially driven by agentic technologies. A simple query in 2023 cost $0.04, while a query to a full-fledged agentic system already costs $1.2—30 times more. For complex workflows, the difference can reach 500 times.

Economists call this Jevons paradox: a decrease in the price of a resource leads to a disproportionate increase in its consumption. In the case of AI, the effect has been hypertrophied.

Who Has Already Suffered

Uber is a striking example. The company actively deployed Claude Code and Cursor. In May, the CEO boasted that autonomous agents were writing 10% of the code. Within a month, news of restrictions emerged: a limit of $1,500 per month per employee. The CTO admitted that he personally spent $1,200 on tokens during a single two-hour demo, exhausting the annual programming budget.

Microsoft conducted a risky experiment by giving thousands of employees access to Anthropic's Claude Code. The tool proved so effective that even designers and managers started using it. Six months later, Microsoft shut down the program, requiring developers to migrate to its own Copilot CLI. Analysts note that abandoning Claude is a compelling signal that the economics of corporate AI coding at current token prices are not working. The tools are so good that they are used constantly, inflating budgets.

Amazon is also revising its strategy. In June, news emerged about a restructuring of its partnership with Anthropic: the corporation will now pay specifically for tokens rather than compute hours. This could significantly increase its costs.

Costly Mistakes and Dependency

Axios journalists cite an example of an anonymous company that forgot to set limits on Claude Opus usage. At the end of the month, the firm received a bill for $500 million. Representatives from Priceline compare tokens to a "crack cocaine epidemic": first, they give you a taste to get you hooked, and then they send the bill. One developer spent over $40,000 on tokens in a single month. Meanwhile, according to analysts, most companies still do not measure the return on such expenses.

A Look to the Future

The term "tokenapocalypse" clearly exaggerates the scale of the problem for giants like Microsoft and Amazon, which have direct investments in AI projects. But for more modest firms, it is already a reality. It is against this backdrop that the Linux Foundation introduced the Tokenomics Foundation project—a new standard designed to instill financial discipline in the context of AI costs, similar to how FinOps manages cloud spending.

My view as an analyst: Tokens have become a new line item in corporate budgets, and companies wishing to remain in the technological mainstream will have to account for this. As AI solutions evolve, such expenses will become as predictable as basic corporate subscriptions. But for now, we are witnessing a classic transition from "free lunch" to a harsh economy—and many companies have proven unprepared for it.