The AI market is facing significant challenges in cost modelling, creating uncertainty for both service users and providers. Although technology giants such as Microsoft, Google and Anthropic have invested hundreds of billions of dollars in developing large language models (LLMs), precise pricing for these advanced services remains surprisingly difficult. Providers are unsure what to charge, while users struggle to control usage costs.
The LLM technology, which underpins popular services such as ChatGPT, Claude and Gemini, is based on processing data through mathematical units called tokens. When a user submits a query, it is broken down into these tokens, which the model processes. The response is returned as tokens, which are then converted back into text, code, or automated commands. Because the economics of tokens are fluid and the processing is not entirely predictable, it is difficult to calculate the exact costs for individual tasks.
Currently, free versions of leading AI tools are offered to users, while paid versions provide additional features tailored to specific tasks, such as programming or more complex analysis. Meanwhile, third-party companies are building and selling services based on AI agents, which are typically built on LLMs and trained to perform specific functions. This diversity of offerings further complicates the market dynamic.
Simon Gooch, from the company Saviynt, which manages identities and integrates agency AI into its services, states that tying costs to 12, 24, or 36-month periods is pointless. The reason lies in the uncertainty of what will happen to technology and prices within that period. Given that conditions are changing too rapidly, long-term budget planning for AI services has become highly risky for businesses.
