Token Economics Create Pricing Chaos for AI Services

Companies building services on large language models face an unusual problem: they cannot reliably price their products. The culprit is tokens, the mathematical units that AI systems consume to process requests and generate responses.

When users interact with ChatGPT, Claude, or similar tools, their prompts get broken into tokens for processing. The AI's response also arrives as tokens, converted back into readable text or executable code. But this process resists prediction—the same prompt can yield different token counts, different models consume tokens at varying rates, and AI agents working together multiply both usage and uncertainty.

The budget explosion problem

While individual token costs have dropped sharply, total consumption is surging. Goldman Sachs projects token usage will increase 24-fold between 2026 and 2030, reaching 120 quadrillion tokens monthly as businesses deploy AI agents at scale.

This volatility has already caught major organizations off guard. Microsoft has reportedly restricted engineers' use of certain third-party coding tools, while Uber burned through an entire year's AI coding budget in just months, according to details first reported by BBC News.

"People are finding it really hard to manage that cost," said Will Venters, Associate Professor of Digital Innovation and Information Systems at the London School of Economics. "It's a non-deterministic output, so it's a non-deterministic value."

Why it matters

The token pricing dilemma threatens to slow AI adoption just as agentic systems promise to automate complex business processes. Unlike hiring human workers—which involves deliberate headcount discussions—deploying additional AI agents requires only a button click, making cost overruns dangerously easy. Companies that cannot predict their AI expenses struggle to build budgets, set customer prices, or justify investments to stakeholders.

No clear path to stable pricing

Simon Gooch at identity management firm Saviynt, which is incorporating agentic AI into its services, said locking customers into multi-year pricing agreements makes little sense. "We don't know" what costs will look like, he explained.

Software firm Sumo Logic is previewing security services based on agentic AI but remains in discussions with customers about pricing models. "Nobody's really figured it out," said Bill Peterson, senior director of product marketing. Options include flat price increases, pay-per-result models, or bundled incident packages—but any structure could collapse when LLM providers change their own rates.

"You get into variable pricing, and it's changing every couple of months," Peterson noted. "Customers don't like that. That's not how anybody builds a budget."

Rob Steele, CFO at UK accounting software firm iplicit, said companies must craft more precise prompts to control costs. Oliver King-Smith, founder of engineering software firm smartR AI, predicted that once AI platforms face shareholder pressure for profits, "they will start clamping down" on flat-fee personal accounts that smaller organizations currently exploit.

Venters acknowledged that despite unpredictable costs, companies may ultimately extract more value from AI than traditional tools. "The more you give it, the more expensive it is, but the better the result may be," he said.

These details were first reported by BBC News.