Grayscale: Cheaper AI Could Increase Demand for Computing Power

Grayscale's research head, Zach Pandl, explained that cheaper artificial intelligence and the development of autonomous AI agents may boost demand for computing power and physical infrastructure. AI agents consume up to 50 times more tokens than chatbots. Goldman Sachs predicts token consumption could increase 24 times by 2030 due to the rise of AI agents. Pandl noted that the competition from lower-cost models reduces token costs, with Anthropic's latest Sonnet model being up to 30% cheaper per task. Autonomous agents, which operate continuously on user goals, significantly increase computation volumes compared to chatbots. Grayscale estimates these workflows may consume 5 to 50 times more tokens than typical chatbot interactions. Data from OpenAI and others confirm that AI agents use nearly five times more tokens than humans, with consumption rising approximately 14 times since February 2026. The most active companies generate over eight times more tokens than average enterprises. Pandl referenced Goldman Sachs' forecast that token consumption could rise 24 times by 2030, driven by agent proliferation. Despite increasing model efficiency, overall demand for computing is expected to grow faster as AI usage expands. However, digital services rely on physical infrastructure, which takes time to supply. Pandl cited CoreWeave's new contracts at higher prices for computing power as an example. The trend of cheaper AI and autonomous agents could widen technology applications and intensify competition for limited computing resources, benefiting companies with the necessary infrastructure.