Unknowable energy footprint

For MIT Technology Review, James O’Donnell and Casey Crownhart ran numbers and interviewed experts go piece together a projection for how much energy AI will use. The takeaway is that it’s impossible to know with any certainty, because companies don’t disclose what they’re building.

The Lawrence Berkeley researchers offered a blunt critique of where things stand, saying that the information disclosed by tech companies, data center operators, utility companies, and hardware manufacturers is simply not enough to make reasonable projections about the unprecedented energy demands of this future or estimate the emissions it will create. They offered ways that companies could disclose more information without violating trade secrets, such as anonymized data-sharing arrangements, but their report acknowledged that the architects of this massive surge in AI data centers have thus far not been transparent, leaving them without the tools to make a plan.

“Along with limiting the scope of this report, this lack of transparency highlights that data center growth is occurring with little consideration for how best to integrate these emergent loads with the expansion of electricity generation/transmission or for broader community development,” they wrote. The authors also noted that only two other reports of this kind have been released in the last 20 years.