Just a year ago, integrating artificial intelligence into the corporate environment was encouraged like an Olympic achievement: tokens were handed out, and employee ratings grew in proportion to their "voraciousness." Today, the situation has changed dramatically. Managers are frantically cutting limits, disabling access, and feverishly revising budgets. Tokens have transformed from free fuel into the main line item of unforeseen expenses, and corporations are only beginning to grasp the scale of the disaster.

What is a token and why did it become money?

In the context of large language models (LLMs), a token is the smallest unit of processed information. On average, 100 words equal 130–140 tokens. They are divided into input tokens (user prompt) and output tokens (model response). For a long time, this was just a technical specification. But in March 2023, OpenAI released GPT-4 and set public pricing: $30 per 1 million input tokens and $60 per 1 million output tokens. From that moment, everything changed.

The period from 2023 to 2025 became the "era of subsidization": companies sold subscriptions to users, but corporate developers were already paying for tokens. Since spring 2026, all major providers (Anthropic, GitHub Copilot) have switched to full monetization of queries. An era has begun that experts call the "tokenapocalypse."

The Jevons paradox in action

The paradox is that over the past three years, the average price of an input token has dropped by about 98%. However, corporate AI spending has skyrocketed by 320% — from $1.2 million in 2024 to $7 million in 2026. The reason is a multiple increase in usage volumes, especially due to agentic technologies. A simple query in 2023 cost $0.04, while a query to an agentic system already costs $1.2. For complex workflows, costs increase by 50–500 times. Economists call this the Jevons paradox: a decrease in the price of a resource leads to an increase in its overall consumption. In the case of AI, the effect turned out to be hypertrophied.

Corporate "victims" of the tokenapocalypse

Uber. The company actively implemented Claude Code and Cursor. In May, CEO Dara Khosrowshahi boasted that agents were writing 10% of the code. Within a month, Uber introduced 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. In December 2025, the corporation granted thousands of developers access to Anthropic's Claude Code. Six months later, the experiment was shut down, with orders to migrate to its own Copilot CLI. The reason — excessive token consumption hit expenses. Engineers preferred the competitor's tool, but the economics didn't add up.

Amazon. In June 2026, the company restructured its partnership with Anthropic, switching to paying per token instead of compute hours. This could increase costs for Claude in Alexa, Kiro, and other services.

Minor incidents. One anonymous company forgot to set limits and received a $500 million bill for AI computing. A developer from Faros AI spent $40,000 in a month. At Priceline, the situation was compared to a "crack cocaine epidemic": they let you try it, get you hooked, and then you're dependent.

The future: discipline and standards

"Tokenapocalypse" is a loud name, but the reality is this: large corporations will pay the bills, while smaller ones will adapt through reasonable consumption. The Linux Foundation has already launched the Tokenomics Foundation project — an analog of FinOps for cloud spending. Tokens have become a new budget line item, and enterprises will have to learn to forecast these expenses just like costs for basic subscriptions.

My comment: The AI tools market is experiencing classic "growing pains." Tokens are the new currency, and companies that fail to learn how to manage their flows risk facing financial turmoil. The adoption of standards like Tokenomics Foundation is an inevitable step toward industry maturity.