AI Firms Struggle to Price Services as Token Use May Jump 24-Fold by 2030
Updated
Updated · bbc.co.uk · Aug 3
AI Firms Struggle to Price Services as Token Use May Jump 24-Fold by 2030
3 articles · Updated · bbc.co.uk · Aug 3
Summary
Businesses selling AI tools and agent-based services are finding it hard to lock in prices because token consumption—and therefore costs—can swing unpredictably from prompt to prompt and model to model.
Goldman Sachs forecasts external token use will rise 24 times between 2026 and 2030 to 120 quadrillion tokens a month, even as per-token prices fall, leaving total bills difficult to predict.
That volatility is already hitting users: Microsoft has reportedly curbed some engineers' use of third-party coding tools, while Uber apparently burned through a full-year AI coding token budget within months.
Vendors are testing workarounds such as flat-fee accounts, tighter prompt design, model selection, and pricing by results or bundled incidents, but those plans can be disrupted whenever LLM providers reset their own prices.
As companies embed AI into products for thousands of users, especially agentic systems, the industry still lacks a stable way to pass rising and variable token costs on to customers.