Artificial intelligence may be becoming cheaper to run, but businesses are discovering a new challenge: the more they use AI, the harder it can be to predict the bill.
The growing adoption of AI agents and systems capable of performing multiple tasks with limited human intervention are driving a sharp increase in the amount of data processed by large language models (LLMs). While the cost of individual AI tokens has fallen significantly, overall consumption is rising rapidly, creating new financial pressures for businesses.
Unlike traditional software, where customers often pay a predictable subscription fee, AI costs can vary depending on how complex a task is, how many requests are made and how much computing power is required.
AI agents can make the problem even more difficult. A single instruction may trigger several automated processes, each consuming additional tokens. As companies deploy more agents across areas such as coding, customer service, security and administration, their AI expenditure can quickly become difficult to forecast.
That uncertainty is forcing companies to rethink how AI services should be priced. Possible models include charging customers according to usage, the number of tasks completed, specific business outcomes or fixed bundles of services.
Each approach comes with challenges. Usage-based pricing can leave customers facing unpredictable bills, while flat-rate subscriptions could expose providers to unexpectedly high costs if customers consume large amounts of AI resources.
Businesses are therefore increasingly looking for ways to monitor token consumption, choose the most cost-effective models and improve the precision of prompts used by employees and AI systems.
The challenge is particularly important as AI becomes embedded in products used by thousands or even millions of customers. A service that appears inexpensive at launch could become significantly more expensive as usage scales.
For the AI industry, the next big question may not simply be how powerful AI can become, but how businesses can afford to use it at scale.
