AI is on track to be the US’s single biggest electricity user. We’ve talked a lot on the show about how we can manage this demand. The consensus is that AI is going to put huge strain on the grid, but what if it could instead keep it running?

Varun Sivaram is a founder & CEO of Emerald AI and a Senior Fellow at the Council on Foreign Relations. He says that “far from undermining the grid, AI could actually save it. If we can enable AI data centers to provide flexibility during peak stress, they can become a powerful ally for reliable, affordable, and clean electricity.” So what does flexibility mean in this context? Earlier in the year, Ph.D student and fellow at Duke university Tyler Norris authored a paper that talked about it; Rethinking Load Growth: Assessing the Potential for Integration of Large Flexible Loads in US Power Systems argued that if grid operators could ask data centers to dial back when the system is under strain, those new facilities could get online faster without waiting for long transmission and generation upgrades. In other words, flexibility is like a fast-track pass: by allowing short reductions in consumption during peak stress, data centers can connect sooner and the grid can handle more demand.

To unpack the big idea – that data centers can be a bit flexible – Varun joins host Ed Crooks, regular guest Amy Myers-Jaffe (Director of NYU’s Energy, Climate Justice and Sustainability Lab) and resident investment expert Shanu Mathew (Portfolio Manager and Research Analyst at Lazard Asset Management).

Their big idea: if data centers can dial down for a few peak hours each year, we could connect new projects faster, keep the grid more reliable, and help protect household bills.

What does “flexible” look like in practice? Spread computing tasks across multiple sites, pause the less time-critical ones during grid stress, and use smarter software and batteries to smooth short spikes. The gang discuss early real-world tests with utilities and tech companies, and why some regions are considering rules that let them temporarily reduce power to big users rather than risk neighborhood blackouts.

Is this all hype? Maybe not. Even as chips get more efficient, demand for AI is growing even faster. Varun wants to run more pilots, reward flexibility with quicker hookups, and build toward a “virtual power plant” made of data centers that can respond in milliseconds.

See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.

AI will need to embrace flexibility if the grid is going to cope. What do flexible loads look like, and how do we get there?

Energy Gang