Nobody Has Figured Out How to Price AI Yet. That Is a Problem for Everyone

Key points

• Goldman Sachs forecasts that external token consumption will grow 24 times between 2026 and 2030, reaching 120 quadrillion tokens a month.
• Microsoft has reportedly pulled back on engineers' use of some third-party AI coding tools over cost concerns.
• Uber reportedly burned through a full year's AI coding token budget in a matter of months in 2025.
• A token is a small mathematical chunk of text that an AI model reads and writes; every question you ask and every answer you receive costs tokens.
• Industry insiders say no standard pricing model for AI agent services has emerged yet.

Using ChatGPT, Claude or Google Gemini for free feels like getting something for nothing. In a sense, it is. Microsoft, Google, Anthropic and their peers have poured hundreds of billions of dollars into building the large language models, the technology behind those chatbots, that power those free tiers. They want that money back.

Paid plans exist, and premium features are locked behind them. But the harder pricing problem sits one layer deeper, and it is tripping up businesses of every size.

What is a token, and why does it cost money?

A token is the basic unit an AI model works with. Think of it as a small fragment of a word or phrase that the model turns into maths, processes, then converts back into readable text. Every prompt you type gets broken into tokens. Every reply the model sends back costs tokens too.

Token prices have fallen sharply in recent years. That sounds like good news, and it partly is. The catch, as BBC Technology reported, is that usage has exploded to match. Goldman Sachs forecasts token consumption will hit 120 quadrillion a month by 2030, up 24 times from 2026 levels, driven mainly by AI agents, software programs that carry out multi-step tasks automatically rather than just answering a single question.

More agents mean more tokens. More tokens mean bigger bills. And the bills arrive before most companies have any idea what caused them.

Why is this so hard to control?

The short answer: AI output is unpredictable. Ask the same question twice and you may get two different answers of two different lengths, burning two different amounts of tokens. Build that unpredictability into a product used by thousands of customers, and costs can spiral fast.

Will Venters, Associate Professor of Digital Innovation at the London School of Economics, puts it plainly: "It's a non-deterministic output, so it's a non-deterministic value."

Smaller companies sometimes sidestep this by quietly using flat-fee personal accounts from big AI platforms. Oliver King-Smith, founder of engineering firm smartR AI, says vendors almost certainly dislike this, and it cannot last. Once shareholders start pushing for profit, he predicts the platforms "will start clamping down."

Even large organisations get caught out. Microsoft reportedly reined in its engineers' use of certain third-party coding tools. Uber reportedly tore through its AI coding token budget for an entire year within months.

What happens next?

No one has a clean answer yet. Bill Peterson, senior director of product marketing at Sumo Logic, a software firm currently previewing AI security services, says pricing conversations with customers are ongoing. Options on the table include raising prices across the board, charging per outcome, or selling bundles of incidents. But any structure a company picks today can be upended the moment an AI provider changes its own rates.

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

For ordinary users, the immediate risk is limited: free and flat-rate plans act as a cap. For small businesses starting to build AI into their own products or workflows, the practical advice from the people living this problem is simple. Be precise with your prompts, the way you would write a detailed shopping list rather than just saying "get some food." Choose the smallest AI model that actually handles your task. And set a spending alert before you need one.

Common questions

Will my free ChatGPT or Gemini account suddenly cost more?

Free tiers are currently subsidised by the big platforms competing for users. They may tighten features over time, but no major provider has announced plans to end free access outright.

What can a small business do right now to control AI costs?

Start by tracking usage monthly, the way you would a cloud storage bill. Write detailed prompts so the AI does not waste tokens guessing what you want. Where possible, test with a cheaper or smaller model before committing to a more expensive one.