Tokenomics: Why making AI pay is tricky

The challenge of pricing AI services is a complex one, as firms like Microsoft, Google, and Anthropic have invested billions in developing Large Language Models (LLMs) that power services like ChatGPT, Claude, and Gemini. These firms want to recoup their investment, offering paid versions with extra features for tasks such as coding or billing. However, third-party companies are also building and selling AI-based services, but setting a price for these offerings proves difficult.

The core issue lies in the economics of "tokens," the building blocks of LLMs and agentic AI. These tokens are used to process prompts, generate responses, and perform automated tasks. The number of tokens consumed can vary significantly based on the prompt and the model used, making it hard to predict costs over the long term. Goldman Sachs predicts that external token consumption will increase 24-fold between 2026 and 2030, posing a significant challenge for consumers and businesses alike.

Even Microsoft, a major player in the AI space, has had to curtail its engineers' use of third-party coding tools due to token costs. Similarly, Uber reportedly burned through its AI coding token budget within months. Associate Professor of Digital Innovation and Information Systems at the London School of Economics, Will Venters, highlights the non-deterministic nature of AI token costs, which can lead to unpredictable and costly billings.

Companies are finding various ways to manage this issue. Some opt for flat fee personal accounts, while others seek to be more precise with their prompts and choose AI models that offer better value. However, the costs can become even more unpredictable when AI is integrated into a product for a wide user base, as additional costs may arise for testing, security, and other tasks.

Despite the challenges, some companies believe the value they receive from AI use may justify the costs, likening it to the benefits of using a calculator. However, companies still need to pass these costs onto customers, a decision that is made increasingly difficult as token pricing strategies remain fluid. The potential for variable pricing and frequent changes could create further uncertainty for customers trying to budget for AI services.