Goldman Sachs Research said agentic AI adoption is poised to increase hyperscaler computing volumes and improve cash flow as costs fall.
The report modeled common use cases and forecast token consumption multiplying 24 times to 120 quadrillion tokens per month between 2026 and 2030. It said the increase is driven by consumer and enterprise agents.
The report also cited rapid declines in inference cost per token of 60% to 70% per year. It said this supports a near-term gross margin “inflection.”
Source: Goldman Sachs