POWER GRIDS 2050
The grid collapses under exponential demand. The fix: $16 trillion.
Key figures
The bill: $16 trillion, 800 extra-large reactors or 3,000 SMRs, 7,500 TWh of new electricity, 28 years to build.
The demand shock: AI data center electricity +50% in 2025, $400 billion capex in 2025, +75% in 2026 (IEA).
The tell: data centers already run on old jet engines because the grid cannot deliver.
Big Tech buys power plants: Meta multi-gigawatt nuclear, Google and Microsoft fusion offtakes, Google training 130,000 electricians.
The future runs on electrons, and we are plugging exponential demand into a 20th-century grid. EVs, AI, crypto, robotics, all fighting over the same socket. I forecast the grid collapses under that load. The bottleneck is no longer code. It is megawatts, water, and permits.
The demand shock is on the tape: AI data center electricity jumped 50% in 2025, on $400 billion of capex, with 75% more coming in 2026. Data centers are already bolting old jet engines to the wall because the grid cannot deliver. That is not innovation. That is a power failure in a suit.
Do the honest math the politicians won't. The world needs 800 extra-large reactors, or 3,000 small modular reactors, plus transmission, storage, and efficiency: 7,500 terawatt-hours of new electricity. The bill: $16 trillion. The build: 28 years. Not the fantasy timelines CEOs sell at earnings calls.
Watch their hands, not their mouths. Meta signs multi-gigawatt nuclear deals. Google, Microsoft, and Eni lock fusion offtakes for plants that do not exist yet. Commonwealth Fusion builds with Siemens and Nvidia. NuScale ships small reactors. Form Energy and Crusoe bury 12 gigawatt-hours of iron-air batteries next to AI data centers. Google is training 130,000 electricians because the grid ran out of hands. When the hyperscalers buy power plants, believe the electrons, not the press releases.