Axios Pro Exclusive Content

AI and energy is top of mind at SF Climate Week

Illustration of a robotic hand holding a glowing dollar sign.

Illustration: Allie Carl/Axios

Artificial intelligence, as both a climate tool and culprit, dominated many of the discussions at SF Climate Week last week.

Why it matters: The looming electricity demand of generative AI models is threatening grid stability. And yet, AI can also help solve that problem.

What they're saying: "The way this technology is moving is infinitely faster than the way that utilities move and plan for infrastructure," said Brian Janous, former VP of Energy for Microsoft, and co-founder of Cloverleaf, a new startup helping computing companies access energy.

  • The main reason the clean electricity surge might not materialize as quickly is because utilities might not be able to build capacity fast enough, said Janous.

Zoom in: At the same time, so-called predictive AI, which can make infrastructure run more efficiently, could help grid operators access more capacity and reduce the time it takes to simulate grid scenarios.

  • Page Crahan, general manager for Tapestry, Google X's moonshot project for the electric grid, said it's using AI to help utilities run grid simulations and make decisions much more quickly.
  • "We're wondering if AI can make the grid instead of breaking the grid," said Crahan during a session last Thursday.

Driving the news: The Department of Energy released a report Monday and a series of initiatives on how a clean-energy build-out could leverage AI.

  • Our colleagues on Axios Generate dug into the utility data center conundrum on Friday.

By the numbers: At a Salesforce summit on AI at the event last week, Juliet Rothenberg, product lead of climate AI for Google, said it's important to keep in mind the numbers around AI and energy.

  • Electricity use for data centers has been holding steady at about 1% (of global electricity) for quite some time, said Rothenberg.
  • Somewhere in the ballpark of 25% of global electricity use for data centers is AI. So that's 0.25% of global electricity use, said Rothenberg.
  • "If you look on a greenhouse gas basis, hyperscalers are responsible for about 0.1% to 0.2% of global greenhouse gases," said Rothenberg.

The big picture: Computing companies are asking utilities for an unprecedented amount of new clean electricity for training AI models.

  • But grid operators could also use AI tools to help meet that demand surge.
Go deeper