Token economics challenge sustainability of AI service pricing models

Third-party AI service providers face mounting difficulties establishing predictable pricing for their offerings as token consumption patterns remain inherently unpredictable. Tokens—the mathematical units that language models process—vary in quantity depending on prompt variations, model selection, and especially in multi-agent systems where several agents operate simultaneously. Although individual token costs have declined significantly, Goldman Sachs forecasts token consumption will surge 24-fold between 2026 and 2030, reaching 120 quadrillion tokens monthly as enterprises scale AI agent deployments. Major companies including Microsoft and Uber have struggled to manage these unpredictable expenses, with Uber depleting an entire year's AI coding budget within months. Smaller organisations are experimenting with alternative pricing approaches, but the fundamental challenge persists: AI outputs and consumption remain non-deterministic, making long-term cost commitments difficult for both providers and customers.