Nvidia thinks AI can solve electrical grid problems caused by AI

Nvidia announced Thursday it’s partnering with EPRI, a power industry R&D organization, to use AI to solve problems facing the electrical grid. Perhaps ironically, the issues are largely caused by rising power demand from AI itself.

The Open Power AI Consortium, which includes a number of electrical utilities and tech companies, says it will use what are known as domain-specific AI models to devise new ways to tackle problems that the power industry is predicted to face in the coming years. The models will be open sourced and available to researchers across academia and industry.

The power industry is facing surging demand from data centers in the United States and elsewhere as AI ramps up the need for computing power. Electricity demand is expected to grow by 4% annually in the coming years, according to the International Energy Agency, nearly double over 2023 figures. 

In addition to Nvidia and EPRI, the consortium includes PG&E, Con Edison, Constellation Energy, Duke Energy, the Tennessee Valley Authority, and ENOWA, NEOM’s energy and water company. On the tech side, Microsoft and Oracle are both members.

In an attempt to stay ahead of the trend, tech companies have been racing to secure generating capacity as power has transformed from a simple line item to a competitive advantage.

Over the last year or so, tech companies have been consistently inking new contracts. They’ve largely been spread across renewable energy projects, spurred mostly by solar’s low cost, modularity, and the speed at which it can be deployed.

Microsoft, for example, recently added 475 megawatts of solar power to its sizable renewable portfolio. Last year, it became an anchor investor in a $9 billion renewable development project run by Acadia, and earlier in the year it said it was working with Brookfield asset management to deploy 10.5 gigawatts of renewable power in the U.S. and Europe, all of which is expected to come online by 2030.

But even though new power sources may be the most obvious answer to losing power shortages, they aren’t the only one. 

One recent study found that by curtailing use when demand on the grid peaks, including shifting tasks that aren’t time sensitive to periods when demand is low, an additional 76 GB of capacity could be unlocked in the U.S. It’s a not insignificant amount, making up approximately 10% of peak demand in the U.S. 

It’s likely those are the sorts of solutions, among others, that this new consortium will be exploring.