Inside FinRAG
Why AI needs power plants
One AI prompt uses less energy than nine seconds of TV. So why are tech giants restarting nuclear reactors? A simple guide to AI's electricity bill: chips, racks, heat, gigawatt campuses and what comes next.
In this article
Every time you ask an AI a question, a power plant somewhere works a little harder.
1. One prompt is tiny
Google measured it. In an August 2025 technical report, the median Gemini text prompt used about 0.24 watt-hours, roughly the energy of watching TV for less than nine seconds. Sam Altman has put an average ChatGPT query at about 0.34 Wh, without sharing a method. Independent estimates range wider, and these figures leave out training.
2. Tiny times billions is enormous
AI is used billions of times a day. That is why the US Energy Information Administration expects record US electricity use in 2026 and 2027, with data centers a major driver, rising to about 4.37 trillion kWh in 2027.
3. The chips are hungry
| Unit | Approx. power |
|---|---|
| One modern AI accelerator (GPU) | ~1,000 W |
| One AI rack (e.g. 72 GPUs) | 100+ kW |
| A typical traditional server rack | ~10 kW |
So an AI rack can draw roughly ten times a normal one. Figures are rounded and vary by model and configuration.
4. Power becomes heat
Almost all the electricity a chip uses ends up as heat. Air cooling struggles at these densities, so new AI data centers increasingly pump liquid straight onto the chips to carry heat away.
5. Gigawatt campuses
The biggest AI campuses are now planned in gigawatts. One gigawatt is about the output of one large nuclear reactor.
The International Energy Agency's Energy and AI report projects data-center electricity use roughly doubling by 2030, to about 945 TWh, slightly more than Japan's total electricity use today. Its 2026 update keeps the same picture.
6. Why tech giants are going nuclear
Nuclear offers steady, low-carbon power around the clock, which is what data centers need:
- Microsoft and Constellation: a deal to restart a reactor at Three Mile Island.
- Google and Kairos Power: small modular reactors.
- Amazon and X-energy: small modular reactors.
What it means for markets
Our reading: the AI race is becoming a race for electricity. That puts utilities, grid equipment, cooling and energy supply alongside chips in the AI value chain, and makes power availability a real constraint on how fast AI can grow.
So next time you hit enter, remember: intelligence has a power bill.
Narration in the video is an AI voice (ElevenLabs). The data-center and cooling-tower image is an AI-generated illustration. Figures are rounded. Not investment advice.
Worth sharing?
Pass this article along to someone who would find it useful.
The work continues in the studio.
More build notes, walkthroughs, and perspectives from the FinRAG team.
Explore all stories
