AI usage cost is the ongoing price of using an AI model. Most providers charge per million tokens, with separate rates for the text you send in and the text the model sends back, and prices differ widely between small and large models. Some products bundle usage into a monthly seat or per-conversation fee.
Cost per request is the tokens used multiplied by the rate. A long system prompt, pasted documents, conversation history, retrieved passages and long replies all add up, and so do image or audio inputs. Techniques such as caching repeated content and routing simple questions to cheaper models can lower the bill considerably.
For a business, the question is cost per outcome: per customer enquiry resolved, per product description written or per order taken. Compare that with the staff time or sales it replaces, and include setup, integration, monitoring and subscription fees, not just model charges.
Costs can surprise you. Busy seasons, looping automated agents, abusive users or a change to a longer prompt can multiply spend. Set usage limits, alerts and per-user caps, and review the bill monthly. Prices have generally fallen over time but change often, so avoid fixed long-term assumptions.
Understanding what drives the cost lets you scale AI where it pays for itself and avoid paying for capability you do not need.