With the productive use of ChatGPT, Claude or Microsoft Copilot, the economic picture changes as well. What still looks manageable in pilot projects can become a considerable ongoing expense once usage is widespread. For more and more companies, AI services are therefore becoming a fixed cost category that has to be managed as carefully as spending on personnel, energy or traditional IT.<\/p>\n
More usage usually means a higher bill<\/strong><\/p>\n With the large, established language models, the price depends on the volume of tokens processed and generated. These small units of text form the basis of billing. As soon as more employees use AI or additional processes are connected, consumption \u2013 and with it, almost automatically, the bill \u2013 grows. In numerous organisations it is now becoming clear that actual spending is well above the assumptions made during planning. According to Forbes, Uber, for example, had already used up its AI budget planned for 2026 within the first four months. Tagesschau also reported in July that AI token consumption at the IT service provider Adesso had increased a hundredfold since December 2025.<\/p>\n However, higher costs do not arise solely from rising usage. Particularly powerful models are often used for tasks where a leaner system could serve the purpose just as well. New model generations such as Claude Fable 5 or comparable offerings are considerably more expensive to run. For most everyday use cases, though, their maximum performance is not needed at all.<\/p>\n