AI Agents Could Reach 1.2 Billion by 2028
Epoch AI data suggests existing and projected AI computing capacity could support hundreds of millions of AI agents, with the potential to reach about 1.2 billion by 2028. The nonprofit estimates that roughly 27.6 million H100-equivalent chips had been sold by June 2026. A LessWrong analysis estimates that this hardware could support between 24 million and 24 billion human-equivalent AI workers, depending on each agent’s computing needs and efficiency.
The projected H100-equivalent stock could rise to 46 million by the end of 2026, potentially supporting about 410 million AI agents under one efficiency scenario. By the end of 2028, the installed base could reach 140 million H100-equivalents and support around 1.2 billion agents.
Epoch’s September 2026 data covered 86 AI facilities, representing about 13.9 million H100-equivalent GPUs and 13.3 gigawatts of power capacity. The tracked sites account for an estimated 44% to 46% of global AI compute. Energy availability and inference efficiency remain the main constraints.
For crypto traders, the figures reinforce the long-term investment case for AI infrastructure, semiconductors, data centres and energy. However, the estimates are highly sensitive to assumptions and do not directly guarantee higher cryptocurrency prices. Traders should monitor chip deployments, power capacity, AI infrastructure spending and adoption of AI agents. IDC separately forecasts 1 billion active AI agents by 2029.
Neutral
The market impact is neutral because the report presents long-term, model-dependent projections rather than a confirmed earnings event, capital flow, regulatory decision or cryptocurrency adoption milestone. The estimate that AI agents could reach 1.2 billion by 2028 may support bullish sentiment toward AI-related equities, chipmakers, data-centre operators and power providers. It could also benefit crypto projects linked to AI computing, decentralised infrastructure and data markets if traders interpret the figures as evidence of sustained demand.
However, the wide range of 24 million to 24 billion potential AI workers shows how sensitive the conclusion is to efficiency assumptions. The 13.3 GW power requirement at tracked facilities also highlights a major scaling constraint. In the short term, traders may react to chip shipment data, AI capital expenditure and semiconductor guidance, but the report alone is unlikely to create a durable move across the broader crypto market. Similar AI infrastructure narratives have previously generated sharp rallies in AI-linked tokens and technology stocks, followed by volatility when valuations outran actual revenue or deployment.
Longer term, continued growth in H100-equivalent capacity, falling inference costs and real-world agent adoption could improve the outlook for AI-related crypto sectors. Conversely, slower chip deliveries, energy shortages, weaker demand or regulatory concerns could undermine the narrative. Traders should therefore treat the report as a thematic signal, not a standalone buy indicator.