Artificial intelligence has become a part of everyday life for millions of people. Free versions of popular tools such as ChatGPT, Gemini and Claude help users write emails, plan holidays, answer questions and even create computer code. While these services appear to be free for many users, the companies behind them have spent hundreds of billions of dollars developing the powerful large language models that make them possible. As a result, the technology industry is now facing a growing challenge of finding the right way to charge customers for AI services without creating confusion or unpredictable costs.
Technology companies including Microsoft, Google and Anthropic already offer paid subscriptions with additional features designed for professionals and businesses. At the same time, many software firms are building their own AI powered products using these large language models as the foundation. However, deciding how much customers should pay has become far more complicated than traditional software pricing.
One of the biggest reasons is the way AI systems consume computing resources. Every request made to an AI model is broken into small units known as tokens. These tokens represent pieces of words, numbers or symbols that the AI processes before generating a response. The more detailed the request and the longer the response, the more tokens are used. Unlike traditional software, however, AI does not always produce the same answer for the same question. Small changes in a prompt can produce different responses, leading to different levels of token consumption and different operating costs.
The challenge becomes even greater with the rise of AI agents. Instead of relying on a single model, businesses are increasingly using multiple AI agents that work together to complete tasks, make decisions and automate business processes. While this improves productivity, it also increases the number of tokens consumed, making future costs much harder to estimate.
Industry experts say this uncertainty is making it difficult for businesses to set long term pricing plans. Simon Gooch of identity management company Saviynt says predicting AI costs even a year ahead is almost impossible because the economics of token usage continue to change rapidly.
Although the cost of processing individual tokens has fallen significantly over the past few years, total usage has risen dramatically. According to analysis by Goldman Sachs, monthly token consumption is expected to increase twenty four fold between 2026 and 2030, reaching around 120 quadrillion tokens each month as more companies adopt AI agents across their operations.
Many businesses are already discovering how quickly AI expenses can grow. Reports suggest that even major technology companies have begun limiting the use of certain third party AI coding tools after seeing costs rise faster than expected. Some organisations have reportedly exhausted annual AI budgets within just a few months because of unexpectedly high token usage.
Will Venters, Associate Professor of Digital Innovation and Information Systems at the London School of Economics, says companies often struggle because AI produces non predictable results. Employees experimenting with different prompts or using AI for multiple business functions can rapidly increase costs without fully realising it.
Some smaller businesses have found temporary ways to reduce expenses by relying on personal subscription plans instead of more expensive enterprise accounts. However, experts believe this approach is unlikely to continue indefinitely. As major AI companies come under increasing pressure from investors to improve profits, many expect tighter controls and changes to pricing structures.
Businesses are also being encouraged to use AI more efficiently. Experts recommend selecting the most suitable AI model for each task instead of always using the most advanced option. They also advise writing more detailed and precise prompts, which can improve results while reducing unnecessary token usage.
Another concern is that organisations often underestimate the hidden costs of deploying AI widely. Beyond generating content or writing software, companies also need AI resources for testing, security checks, monitoring and implementing safeguards. These additional tasks can significantly increase token consumption as AI becomes integrated across entire organisations.
Despite the growing costs, many business leaders believe AI still delivers strong value by improving productivity and automating time consuming work. The challenge is finding a pricing model that fairly reflects those benefits while remaining predictable enough for customers to budget.
Software companies are currently exploring several options, including charging fixed subscription fees, pricing based on successful outcomes, or offering bundled services. However, each approach has drawbacks, particularly because the underlying AI providers frequently adjust their own pricing.
As AI adoption continues to accelerate, the search for a stable and transparent pricing model remains one of the technology industry's biggest unanswered questions. Until companies find a balance between affordability, profitability and predictability, both businesses and customers are likely to face ongoing uncertainty over the true cost of using artificial intelligence.