The expansion of U.S. data centers is primarily driven by the need to support artificial intelligence programs, raising concerns about their environmental impact and the strain on the energy grid. A recent MIT study suggests that the effects of data centers could vary significantly based on their energy consumption patterns. Shifting significant energy use to non-peak hours could potentially lower average energy costs. However, the environmental impact would depend on the local energy mix, with some areas increasing renewable energy use and others relying more on fossil fuels.<\/p>\n
Christopher Knittel from MIT Sloan School of Management, who co-authored the study, emphasizes that the challenge is integrating data centers into the grid without significantly increasing peak usage. By offering grid flexibility during expensive high-demand periods, data centers could help manage energy costs. The study indicates that flexible energy consumption could lead to savings of up to 5% in Texas, 4% in the Mid-Atlantic, and 2% in Western U.S. states if data centers adjust over 20% of their consumption to non-peak times.<\/p>\n
The paper, titled \u201cFlexible Data Centers Reduce Power System Costs But Can Increase Emissions,\u201d appears in the journal iScience. Authors include Juan Ramon L. Senga and Shen Wang, both postdocs at MIT\u2019s Center for Energy and Environmental Policy Research, and Knittel. The study uses extensive simulations to explore scenarios where data centers expand, applying the \u201cGen X\u201d model of the U.S. power grid to assess energy use over a year.<\/p>\n
Research focused on three grid systems: Texas, the Mid-Atlantic, and the Western Interconnect, which is expected to host 82% of U.S. data centers by 2030. Adding data centers could lower energy costs by allocating fixed grid costs over a larger energy volume. However, this cost reduction hinges on data centers increasing their average energy use more than their peak use, taking advantage of their existing capacity flexibility.<\/p>\n
The study modeled shifting energy use from peak times to midday when solar energy is more available. Knittel notes that data centers must decide how much of their energy load is flexible and how far they can shift their usage. This flexibility varies based on the type of AI-related computation, with AI training data centers having steadier energy use, allowing more flexibility than inference data centers.<\/p>\n
Significant cost savings, between 2% and 7%, could be achieved according to the study. Knittel points out that even a 3% saving across a $100 billion grid is substantial. However, the environmental impact could be considerable, with data center growth potentially increasing carbon emissions by 58% in Texas, 20% in the Mid-Atlantic, and 24% in the Western U.S. by 2030.<\/p>\n
The study also finds regional differences in clean-energy use implications. In Texas, flexible data centers could reduce emissions by boosting wind energy use. In contrast, in the Mid-Atlantic, increased data centers might raise both renewable and fossil fuel energy use, potentially increasing CO2 emissions by 3%. Knittel notes that flexible energy use can inadvertently support coal energy, depending on the timing of renewable production.<\/p>\n
For the benefits to materialize, data centers would need to adopt flexible energy-use schedules, which might require policy intervention. Knittel suggests that offering faster grid connections in exchange for time-of-use flexibility could incentivize tech firms. Such policies could encourage industry-wide flexibility, making it feasible for companies to agree to these terms.<\/p>\n