AI Agent Usage Jumps 14-Fold, Driving Token and HBM Demand, a16z Says

AI agents are evolving from chat-based assistants into systems that can break down tasks, repeatedly call models and execute multi-step workflows. That shift makes them far more compute-intensive than ordinary human interactions and turns token consumption into a key measure of enterprise AI adoption. Andreessen Horowitz, known as a16z, said the trend has implications beyond software spending, extending to computing capacity, memory and data-center infrastructure.

AI agent usage increased roughly 14-fold over six months, according to a16z’s latest report, while agents consumed nearly five times as many tokens as human users. Adoption is also becoming increasingly concentrated: the top 10% of enterprises by usage are generating eight times the token output of a typical company. Because agent workloads make extensive use of cached tokens, continued growth is expected to increase demand for AI infrastructure, including high-bandwidth memory, or HBM.