As agentic AI now dominates the conversation when it comes to AI computing, investment bank Goldman Sachs is out with a report that agentic computing will drive a jump in token consumption. The bank believes that token consumption will grow 24 times by 2030 compared to 2026’s figures as AI queries jump from five billion to 23 billion, driven particularly by non-human agents.
Goldman Sachs Out With Optimistic Report About Agentic AI
The central theme in Goldman Sachs’ report concerns a rapid growth in the use of agentic AI. Agentic AI involves the use of software bots that compute through problems, and the growth in its usage has also led to fresh optimism for the fortunes of CPU companies such as Intel, AMD and Arm. According to Goldman Sachs, the rise in agentic AI will spur token usage, which in turn will make the massive capital expenditure into AI computing capacity profitable.
The bank believes that agentic AI will drive token consumption since agents operate 24/7 to monitor their environment, verify their data and use external tools. Consequently, it predicts that by 2030, global token usage will surge by 24 times compared to 2026’s figures.
Goldman Sachs’ Optimism About Agentic AI Is Accompanied By Warning About Data Quality
Goldman Sachs adds that by 2030, AI queries will jump from 5 billion in 2025 to 23 billion. It adds that 30% of the queries in 2030 will be represented by agentic use. The higher usage and agentic proportion will be driven by use cases ranging from supply chain management, programming and law, with agents running queries without the risk of fatigue and machine-to-machine interaction.
The bank is even more optimistic about 2040 as it believes that by then, agentic AI use will have grown token consumption by 55 times. It believes that enterprise use cases for agentic AI are currently in their nascent stage, as less than a quarter of enterprises are using it. Additionally, according to Goldman Sachs, those enterprises that are currently using the technology have yet to make the transformation to fully autonomous mode.
As for the costs, the bank believes that the latest chips from NVIDIA and AMD, as well as those such as Trainium, the costs per token computation are dropping by as much as 60% to 70% annually. Using this, Goldman argues that by the first half of this year, the gross margins for AI providers and cloud providers can turn positive. However, it warns that risks such as poor data could lead to large amounts of resource consumption without adequate returns.
About the author: Ramish is a seasoned technology writer and editor with more than a decade of experience. He specializes in semiconductor fabrication and market analysis.
With a background in finance and supply chain management – via his bachelors in Finance and a micromasters in supply chain management from MIT – Ramish combines financial rigor with deep industry insight to deliver accurate and authoritative coverage.
Follow Wccftech on Google to get more of our news coverage in your feeds.