The title may not win Christmas cheer awards, but letโs channel a bit of Scrooge to debunk a growing myth: nuclear as the savior for AIโs energy appetite.
In my year-end post, I dubbed small modular reactors the โfad of the year,โ but I missed the deeper storyโthe energy implications of the AI revolution. Then I came across Michael Liebreichโs article on AI-driven energy demand and its ripple effects. Itโs a must-read, packed with sharp insights and Liebreichโs signature wit. Here are the highlights (but trust me, the full article is worth your time):
๐ธ ๐ง๐ต๐ฒย $๐ฒ๐ฌ๐ฌ๐ย ๐ค๐๐ฒ๐๐๐ถ๐ผ๐ป
Thatโs the revenue AI needs to justify the billions poured into it. So far, only a fraction of this number has a clear roadmap.
โก ๐ฃ๐ผ๐๐ฒ๐ฟ-๐๐๐ป๐ด๐ฟ๐ย ๐ค๐๐ฒ๐ฟ๐ถ๐ฒ๐
A single ChatGPT prompt consumes 2.9 watt-hours. Compare that to 0.3 watt-hours for a Google search. Scale this across billions of users, and youโre looking at serious energy consumption.
โ๏ธ ๐๐ย ๐๐ป๐ฑย ๐๐ป๐ฒ๐ฟ๐ด๐ย - ๐ย ๐ง๐๐ผ-๐ช๐ฎ๐ย ๐ฆ๐๐ฟ๐ฒ๐ฒ๐ย
Deploying AI not only increases energy consumption but also helps to reduce it by optimizing power generation, transmission and consumption. Mr. Liebreich estimates 30 GW of new dispatchable capacity will be needed in the US by 2030, with 15 GW more globally.
โข๏ธ ๐ก๐๐ฐ๐น๐ฒ๐ฎ๐ฟย ๐๐๐ปโ๐ย ๐๐ต๐ฒย ๐๐ป๐๐๐ฒ๐ฟ
Despite the buzz around SMRs, nuclear wonโt power the AI boom. Itโs too slow to deploy and too expensive.
โ๏ธ ๐ฅ๐ฒ๐ป๐ฒ๐๐ฎ๐ฏ๐น๐ฒ๐ย ๐๐ฟ๐ฒ
Solar, wind, batteries, and some backup gas plants will shoulder the loadโdelivering the power AI needs at a fraction of the cost and time nuclear requires.
For your last weekend read of 2024, dive into Michael Liebreichโs articleโequal parts thought-provoking and entertaining. And as always, Iโm curious to hear your thoughts: Is AI a boon or bane for energy systems? Drop your take in the comments below, and follow me for more green tech scale-up insights!