A new study by researchers at the Massachusetts Institute of Technology (MIT) reveals how shifting energy consumption in data centers to non-peak hours can lower overall energy costs but may have complex effects on carbon emissions depending on the region. Published today in the journal iScience, the research was led by economist Christopher Knittel from the MIT Sloan School of Management together with postdoctoral researchers Juan Ramon L. Senga and Shen Wang from MIT’s Center for Energy and Environmental Policy Research.
What Happened
With the rapid growth of U.S. data centers—driven largely by demand for artificial intelligence applications—their increasing energy consumption has raised concerns about environmental impact and strain on power grids. To address these issues, MIT researchers conducted detailed simulations using the “Gen X” model of the U.S. power system. They analyzed energy use patterns in three major regions hosting most U.S. data centers by 2030: Texas, the Mid-Atlantic, and the Western Interconnect, which encompasses 11 western states.
The study modeled scenarios where data centers shift a significant portion of their energy use—sometimes over 20 percent and up to 50 percent—from peak demand periods to off-peak hours. This flexible energy use schedules could lead to cost savings on grid operation and influence emissions of carbon dioxide differently across regions.
Key Facts
The study, titled “Flexible Data Centers Reduce Power System Costs But Can Increase Emissions,” appears in iScience in 2024. The research was supported by MIT’s Future Energy Systems Center under the MIT Energy Initiative. It employed a year-long simulation of power grid operations across three U.S. regions that collectively will host approximately 82 percent of data centers nationally by 2030. The model combined data on grid fixed and variable costs, renewable energy penetration, and flexibility in data center electricity consumption.
What This Means
This study highlights that the environmental and economic impacts of data center energy use depend critically on how consumption is timed within the grid’s demand cycles. By reallocating energy consumption to off-peak times—particularly when renewable energy generation like midday solar or overnight wind is abundant—data centers can help flatten demand spikes, which may reduce grid operating costs by up to 5 percent in Texas, 4 percent in the Mid-Atlantic, and 2 percent out west. These savings translate into significant monetary value considering the scale of U.S. electricity markets.
However, the effects on carbon emissions are more nuanced. In regions like Texas, where wind power is a major source, flexible consumption by data centers could decrease emissions by increasing demand for renewables. Conversely, in the Mid-Atlantic, flexibility might cause coal plants to continue operating during times when renewable output drops, inadvertently raising emissions slightly. This underscores that energy strategies must be tailored to regional generation mixes and that demand flexibility alone is not a silver bullet.
For communities and policymakers, these findings suggest that incentivizing or regulating data centers to adopt flexible energy use could improve grid efficiency and environmental outcomes if aligned carefully with local power sources. It also points to the growing importance of managing the expanding data infrastructure’s energy footprint amid rising AI-driven computing demands.
Background
Prior to this work, concerns have mounted over the impact of rapidly growing data centers on U.S. electrical grids and greenhouse gas emissions, especially as AI workloads surge. Data centers consume power continuously but have some inherent flexibility as most operate below full capacity, allowing for some load shifting. Earlier research has examined grid stress from new large power users but often treated data center loads as static. This study provides an integrated, region-specific analysis of flexible consumption’s system-wide consequences.
What Remains Unclear
The researchers note uncertainties in how much actual flexibility data centers could or would implement in real-world operations, given technological and business constraints. The balance between AI training workloads, which run consistently, and inference workloads, which fluctuate with user demand, affects flexibility potential but is not fully predictable. Furthermore, the study does not conclusively determine long-term effects on grid investment decisions or detailed local environmental impacts beyond carbon emissions.
What Comes Next
The MIT team plans further modeling to refine understanding of data center impacts under varying market and policy scenarios. They underscore the potential value of policies like expedited grid connections paired with contractual obligations for flexible energy use, which could encourage widespread adoption of load-shifting without delaying data center deployment. Such regulation could help align industry practices with grid stability and emissions reduction goals as data center capacity continues to grow.
Sources
This article is based on reporting and publicly available information from the following sources:
Read more Science & Technology stories on Goka World News.