Epoch AI: Chip Stocks Could Support Over a Billion AI Agents by 2028

Research from Epoch AI suggests that existing chip stocks could support over a billion AI agents. The estimates are based on approximately 27.6 million H100-equivalent chips sold as of June 2026, which is a standardized unit for counting AI computing power.

The next step in calculating the potential number of AI workers is to determine how much compute each agent needs. This is measured in FLOP/s, or floating-point operations per second. A September 2026 assessment published on LessWrong took Epoch's chip figures and ran the math across different efficiency assumptions, resulting in a wide range of possible outcomes.

The analysis forecasts that by the end of 2028, the installed stock will swell to roughly 140 million H100e, potentially enabling around 1.2 billion AI agents. However, this depends on various factors, including energy costs and infrastructure growth. The same analysis indicates that the projected cost of running these agents could come in significantly lower than human labor.

Epoch's GATE model points to fast growth in infrastructure capable of running AI inference. As of September 2026, Epoch's data center tracking covered 86 AI facilities, which house about 13.9 million H100e GPUs with a combined power capacity of 13.3 GW. This represents approximately 44-46% of global AI compute.