What is powering AI?
A 15-minute video on how data centers are powered and what comes next. Covering the grid, Jevons paradox, gas turbines, nuclear, solar, and what happens in the gap between now and unlimited energy
Hey there! 👋
Skander here.
You need one year to build a data center. You need seven years to connect it to the grid. Something doesn’t add up.
I spent the last year looking into this and made a 15-minute video on what I found.
Data centers consume 1% of global electricity today. Household devices (TVs, phones, everything you own) use 2%. And there’s 12 gigawatts under construction with 60+ gigawatts planned. For context: a city of 1 million people uses 1 gigawatt.
In the video I try to answer: how are data centers actually powered, and what comes next?
I cover the grid problem, Jevons paradox, gas turbines, nuclear, solar, geothermal, fusion, data centers in space (yes, really), and what happens in the 5 to 25 year gap between now and when the moonshots come online.
Quick TLDR for the skimmers:
- Gas is the answer right now. Data center developers are shipping turbines in on trucks because they can’t wait seven years for a grid connection. Elon started it with xAI’s Memphis data center. Everyone followed. The US is a petrostate and gas is available today. There are problems.
- Nuclear is probably 10 years out. Google and Amazon are already buying offtake from reactors that don’t exist yet.
- Solar is exploding. Actual installations have been 3x higher than every five-year forecast. Mostly thanks to China, which installed more than twice as much solar as the rest of the world combined.
There’s a lot more in the video. Take the 15 minutes.
👉 If you want more videos like this, drop a comment: what did you learn, what do you agree with, and what was your favorite fun fact?https://www.linkedin.com/feed/update/urn:li:activity:7439989512795025408/
Chapters
00:00 Intro
01:15 The data center buildout wave
04:00 Demand, efficiency, and Jevons Paradox
06:20 Powering data centers: gas, grid, and bring your own
12:00 Solar, nuclear, and moonshots
For more Climate Drift: Community | Accelerator | Open Climate Firesides | Deep Dives
Excellent deep dive. One aspect that stands out is the role of AI-driven demand management and load flexibility mentioned. As hyperscalers sign these massive PPAs for solar and wind, the real "final boss" remains the 24/7 carbon-free energy (CFE) challenge.
I’m curious to see how the plummeting costs of 4-hour battery storage—driven by this same learning curve—will start to decouple data centers from the traditional grid constraints. If AI can prove the model for "local balancing" at scale, it provides a blueprint for every other heavy industry to follow.