Jensen Huang walked into a Stanford University lecture hall and casually explained that the AI industry is about to need roughly a thousand times more computing power than it uses today. Not ten times. Not a hundred. A thousand.
The Nvidia CEO, speaking during the CS153 Frontier Systems course in May 2026, outlined a roadmap where agentic AI systems, the next evolution beyond the chatbots and image generators we’ve grown accustomed to, will demand computational resources that make current generative AI look like a calculator app.
From generative to agentic: why the jump matters
Agentic AI systems can understand context, reason through problems, plan multi-step actions, and use external tools autonomously. According to Huang’s lecture, agentic AI requires approximately 1,000x more compute than today’s generative models. The infrastructure that powers ChatGPT and its peers would need to be scaled by three orders of magnitude to support the next generation of AI.
Huang also pointed to an expected 100x increase in the number of users interacting with these systems. The energy implications are equally staggering. Huang indicated that energy requirements for advanced AI computing could climb to roughly 1,000x current levels.
