Major technology firms investing in artificial intelligence are encountering significant challenges in establishing sustainable pricing models for their AI-based services and products. The difficulty stems from the inherent unpredictability of how AI systems consume computational resources, measured in units called tokens, which form the foundation of large language models and AI agent systems.
Tokens are mathematical chunks that break down user prompts and convert AI responses back into usable formats such as text or software code. The process is not entirely deterministic—variations in prompts can produce different outputs, and different AI models yield different results. This unpredictability becomes more pronounced when companies deploy multiple AI agents simultaneously to make decisions and take actions, further complicating cost calculations.
While the per-token cost has declined significantly in recent years, overall token consumption has surged dramatically. Goldman Sachs forecasts that external token consumption will increase 24 times between 2026 and 2030, reaching 120 quadrillion tokens monthly as businesses transition to AI agent systems. Companies frequently struggle to track their actual token usage until budgets are exhausted or billing statements arrive, with even major firms like Microsoft and Uber reportedly experiencing unexpected overages.
Companies attempting to build AI capabilities face particular challenges when scaling services to thousands of users, as token costs can escalate unpredictably across development, testing, security, and implementation phases. The non-deterministic nature of AI outputs creates complications for cost management, as managers cannot reliably predict expenses the way they can with traditional software infrastructure.
Various pricing approaches are being explored by vendors, including flat fees, performance-based models, and bundled incident pricing. However, any pricing structure risks becoming obsolete if large language model providers adjust their own cost structures. Industry observers indicate that the market has not yet settled on standardized solutions, with vendors and enterprise customers continuing active discussions about appropriate pricing mechanisms.
Article Attribution | Read More at Article Source
Article summary produced by Claude AI