Goldman Sachs: AI Agent Token Usage Could Surge 24x by 2030
Goldman Sachs projects that AI agent token consumption could increase 24-fold by 2030, warning that agents represent the first real stress test on AI economics. Unlike a single-response chatbot, an agent plans, calls tools, verifies results, and iterates—consuming 10x to 50x or more tokens per user request. In a bull case scenario, monthly inference demand could reach 120 quadrillion tokens by 2030, though Goldman expects inference cost per token to decline by 60–70% annually. The report comes as Uber and Microsoft are reportedly re-evaluating expensive agent use cases.