The surge in demand for AI-driven computing has fueled investments in large datacenters that require a lot of power. Amazon, Microsoft and Google invested $42 billion more into datacenters from 2022 to 2023. This spike in datacenters will place significant pressure on power grids, causing grid operators to halt or slow new datacenter connections to ensure grid reliability. University of Chicago computer scientist Andrew A Chien and his colleagues assess how relaxing new datacenter power reliability guarantees could increase the available grid power for new datacenters. Specifically, they study two datacenter hotspots: Ireland served by EirGrid (datacenters=16% grid load in 2022) and Northern Virginia served by Dominion Energy (datacenters=24% grid load in 2022).
If EirGrid were to provide weaker guarantees to data centers for reliability, it would be able to meet AI demand for the next five years and still provide datacenters with 99.6% power availability. Note that data centers would receive the same amount of power, but with slightly weaker guarantees. Studying Dominion, however, tells a different story. Even if Dominion were to reduce new data centers’ power reliability, it cannot meet the growing demand. Starting from an existing datacenter capacity of 2,767 megawatts (MW) in 2022, demand is expected to skyrocket to nearly 9,300 MW by 2028 due to the rapid expansion of AI and cloud-based technologies. However, Dominion’s infrastructure can only reliably support about 5,400 MW—about half of the projected demand.
Relaxing how much power Dominion guarantees to datacenters to 80% still does not allow the grid to meet the need. Even if it guarantees zero reliability to the datacenters, the grid is projected to meet only 74% of projected demand. This shows that drastic reductions in reliability alone are insufficient to meet projected AI demand supported by data centers. The contrasting results of the same approach in Dominion and EirGrid highlight the importance of region-specific solutions in addressing these capacity challenges, as well as the need for innovative grid planning, investment, and load management to support sustainable data center growth amid the AI boom.