From TechCrunch, March 20:
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....
....MUCH MORE