AI’s power appetite is getting ridiculous

The AI boom just found its biggest weakness

Turns out the AI gold rush doesn’t just need more chips, more servers, and more engineers. It needs a mountain of electricity. Goldman Sachs now sees global data center power demand surging 220% from 2023 levels by 2030, a sharp step up from its earlier 175% forecast. In plain English: the cloud is getting thirstier, and the power bill is about to look like it belongs to a small country.

Uncle Sam gets the biggest bite

The U.S. is expected to shoulder about 60% of that new demand, up from roughly 50% in prior estimates. Goldman’s revised math points to U.S. data center power use hitting around 750 TWh by 2030, while global demand climbs to 1,350 TWh. That means utilities, grid operators, and anyone selling the stuff that keeps the lights on could be staring at a very long runway.

Who wins when the grid gets crowded?

This is where the investor angle gets interesting. If data centers keep scaling like this, the winners may be less about the flashy chatbot and more about the boring but essential plumbing:

• Utilities with nuclear or always-on generation, like Constellation Energy and Duke Energy
• Grid and electrical equipment suppliers such as Eaton and Quanta Services
• Behind-the-meter power solutions like Bloom Energy
• Big Tech hyperscalers like Amazon, Microsoft, and Alphabet, which may pay more for power but also have the scale to absorb it

The catch: power is the new bottleneck

Data centers already account for about 6% of U.S. electricity demand, and that could rise to 11% by 2030. So while everyone is obsessing over AI model quality, the real constraint may be whether the grid can keep up without tripping over its own extension cords.

Big picture: if Goldman is right, AI isn’t just creating a computing supercycle — it’s creating an energy supercycle too.