I did some research on the energy consumption of AI, as I wasn’t super informed on actual numbers of energy consumption as a whole and have only seen random article write-ups saying it will need a lot of energy.
Here’s the learnings if curious!
The US’ energy consumption is around 4.2T KW/H as of 2022, China 7.2T KW/H.
US consumption has been fairly flat YoY 3, hovering up and down around 4T KW/H. It decreased 1% in 2023.
Our energy consumption is split 35% commercial/public services, 37.4% residential and industry 19.9% and a few smaller slices. 1 It’s produced by 13% coal, 69% oil and gas, 8% renewable, and 10% nuclear.
Data centers today are using around 143B KW/H, ~3% of US’ total consumption.
For future projected energy consumption Most articles are citing this 2 article by Goldman Sachs. The article projects in 2030 using 400B KW/H, with 90B KW/H of that being specifically AI workloads. The article and many other sources cite this as “9% of total energy consumption by 2030” referencing today’s energy consumption, rather than a projected 2030 total energy consumption. From now till 2030, we’re expecting a 2.4% increase in energy consumption, 0.9% of that being from datacenters
- [1]https://en.wikipedia.org/wiki/Electric_energy_consumption#:~:text=About%20a%20fifth%20of%20global,in%201981%20(8%2C132%20TWh)
- [2]https://www.goldmansachs.com/insights/articles/AI-poised-to-drive-160-increase-in-power-demand
- [3]https://www.enerdata.net/estore/energy-market/united-states/#:~:text=United%20States%20Power%20Consumption,dipped%20by%201.5%25%20in%202020.&text=The%20main%20consuming%20sectors%20are,followed%20by%20industry%20(21%25).