Companies developing AI-based products and services face significant challenges in establishing sustainable pricing models, according to industry executives and analysts. The difficulty stems from the unpredictable nature of token consumption, which represents the computational cost of processing requests through large language models like ChatGPT and Claude.
Tokens function as the fundamental units by which LLMs process information, converting user inputs into mathematical chunks that the models can interpret and respond to. However, the amount of tokens required varies based on subtle prompt variations, model selection, and the use of multiple AI agents in tandem. This unpredictability makes it difficult for businesses to forecast costs over extended contract periods. Goldman Sachs projects that external token consumption will increase significantly between 2026 and 2030, reaching 120 quadrillion tokens monthly as companies shift toward agentic AI systems that coordinate multiple AI agents for complex tasks.
Many organizations have discovered they lack visibility into their actual token usage until they exhaust budgets or receive billing statements. Notable instances include Microsoft reportedly restricting certain engineers’ access to third-party coding tools and Uber reportedly consuming an entire year’s AI coding token allocation within months earlier this year. According to academics and industry practitioners, the non-deterministic nature of AI outputs makes it challenging to establish clear value propositions tied to specific costs.
Companies are experimenting with various approaches to address this challenge. Some smaller organizations currently utilize personal accounts with flat-fee structures, though industry observers expect major AI platforms will eventually impose stricter controls once shareholder pressure for profitability increases. Other strategies include implementing more precise prompting practices, carefully selecting appropriate AI models, and establishing usage guidelines among employees.
When AI technology is embedded into products deployed to thousands of users, cost management becomes increasingly complex, as expenses can escalate across development, testing, security implementation, and guardrail mechanisms. Industry participants note that pricing structures could range from across-the-board price increases to results-based compensation or bundled incident charges. However, any pricing model could be destabilized by future changes to the underlying token pricing set by major language model providers, which have shifted multiple times in recent months, complicating customer budget planning.
Article Attribution | Read More at Article Source
Article summary produced by Claude AI