Analysis Scientific American · By Tam Hunt

AI’s power needs could undo the renewable energy revolution

Solar has finally won on price. In Scientific American, Tam Hunt argues that AI data centers could swallow those gains, and that there are only a few years to act.

By Tam Hunt

For decades, clean energy advocates fought to make solar cheaper than fossil fuels. That fight is largely won. As Tam Hunt writes in Scientific American, solar prices fall about 20% with every doubling of installed capacity, renewables made up 92.5% of new global generation in 2024, and solar is now the cheapest new power in most countries.

The threat. AI data centers could consume those gains. U.S. data centers used about 183 terawatt-hours in 2024, roughly the electricity demand of Pakistan, and projections run to 325–580 TWh by 2028, mostly driven by AI. Projects like the 10-gigawatt Stargate buildout, the restart of Three Mile Island Unit 1 for AI, and multi-gigawatt campuses show the scale. In the PJM grid region, data centers account for 30 of the 32 gigawatts of projected peak growth through 2030, and rates are rising.

Efficiency won’t save us by itself. When DeepSeek matched GPT-4 with far less computing, developers built bigger systems rather than using less energy. Official forecasts, Hunt argues, understate the risk because they miss concentrated megaprojects; if half of the announced U.S. projects are built, data centers could reach 15–20% of U.S. demand by 2030.

The bottleneck is the grid. Transmission and equipment lead times run past two years. The default response so far has been reactive: new gas plants and approved rate increases. The U.S. has no mandatory efficiency standards for AI and no coordination between AI growth and grid planning.

What to do. Hunt calls for slowing data center construction while policy catches up, and treating AI’s energy demand with the urgency we apply to climate. That is the thinking behind Grid Justice Law’s push for clean Bring Your Own Generation: new data centers should bring new clean power and pay their full grid costs, so that the renewable gains households paid for aren’t simply eaten by new load.

Read the full article: “AI’s Power Needs Will Destroy the Renewable Energy Revolution,” Scientific American