Major technology firms, including Microsoft, Google, and Anthropic, have invested heavily in Large Language Models (LLMs). While these companies offer free versions of their tools to consumers, they are increasingly focused on monetization through paid subscriptions and enterprise services. However, establishing sustainable pricing models for these AI-driven services has proven difficult due to the unpredictable nature of token consumption.

Tokens, which serve as the fundamental units of data processed by LLMs, are consumed whenever a user or an automated system prompts an AI. Because these models are non-deterministic, identical prompts can yield different outputs, making it nearly impossible for companies to forecast their usage expenses accurately. Simon Gooch of Saviynt noted that traditional long-term cost modeling is ineffective in this environment, as the underlying economics remain in flux.

Advertisement

The rise of "agentic" AI—where multiple AI agents collaborate to perform complex tasks—further complicates the issue. Goldman Sachs projects that token consumption will grow 24-fold between 2026 and 2030, yet many organizations struggle to monitor their usage until they receive unexpectedly high bills. High-profile instances of budget overruns have already been reported, with some firms reportedly restricting internal access to AI coding tools to manage costs.

Experts suggest that businesses must adapt by refining their prompting strategies and carefully selecting which models to employ. Will Venters, an associate professor at the London School of Economics, emphasized that while AI costs are difficult to control, they may provide greater value than traditional software, even as they scale rapidly. For now, companies like Sumo Logic are still exploring how to pass these variable costs to customers, considering options such as flat-rate bundles or performance-based pricing, even as they remain wary of shifting pricing structures from major AI providers.

Source: BBC News