The Future of AI: Jensen Huang's Vision of Nuclear Power
Nvidia CEO Jensen Huang told podcast host Joe Rogan on December 3 that electricity supply, not semiconductor availability, has become artificial intelligence's primary constraint, predicting tech companies will operate their own small nuclear reactors within six to seven years.
During the wide-ranging Joe Rogan Experience interview, Huang called energy "the bottleneck" for AI development. When Rogan asked if energy had become the main obstacle, Huang quickly affirmed. He envisioned a future where companies deploy compact nuclear reactors producing "hundreds of megawatts" near data centers, transforming tech firms into power generators.
"We'll all be power generators, just like somebody's farm," Huang said, arguing that localized nuclear energy could reduce grid strain while providing dependable power and allowing excess electricity to return to communities. Rogan called this approach "the smartest way to do it".
Nuclear Deals Already Underway
Huang's prediction aligns with industry moves already in motion. In October 2024, Google announced plans to purchase 500 megawatts from small reactor developer Kairos Power, targeting its first advanced reactor by 2030. In August 2025, Kairos and the Tennessee Valley Authority signed the first U.S. utility commitment to purchase electricity from next-generation reactors, with the 50-megawatt Hermes 2 plant in Oak Ridge, Tennessee, set to power Google data centers.
Goldman Sachs projects data center electricity consumption will surge 175% by 2030 compared to 2023 levels—equivalent to adding another top-10 energy-consuming nation to the global grid. The International Energy Agency forecasts global data center consumption will more than double to 945 terawatt-hours by 2030, up from 415 terawatt-hours in 2024.
Deep Learning's Humble Origins
Huang also revealed historical details about Nvidia's AI journey. He disclosed that in 2012, researchers at the University of Toronto invented deep learning using just two GTX 580 graphics cards to train AlexNet, the breakthrough image recognition model. "We were basically the FSD computer version 1," Huang said, describing how he helped Tesla CEO Elon Musk build the Model S computer and early autonomous vehicle hardware. When Huang later announced Nvidia's first DGX-1 supercomputer in 2016, Musk became his first customer, using it for OpenAI.
