Data Centers, AI & Power Demand
Paper Session
Monday, Jan. 4, 2027 10:15 AM - 12:15 PM (EST)
-
Chairs:
Jamal Mamkhezri, New Mexico State University and Georgia Institute of Technology - Xiaochen Sun, New Mexico State University
Processing Power: The Effect of Data Centers on Wholesale Electricity Markets
Abstract
Artificial-intelligence-driven data centers are reversing two decades of flat U.S. electricity demand and have generated questions about how this growth will impact electricity prices. We quantify this effect using an hourly, unit-level least-cost dispatch model covering wholesale electricity markets in the continental United States. We find that existing data centers have already increased wholesale prices by 3 to 5% on average nationwide, with substantially larger effects in regions hosting major data center corridors. Extending the model through 2028, we show that if proposed construction proceeds under high-utilization scenarios, wholesale prices could rise dramatically (50%), while more moderate build-out yields smaller (20%) but still meaningful effects. Impacts vary due to utilization and build-out assumptions. Finally, we use the model to address several policy discussions including optimal data center siting decisions and renewable build-out uncertainty.An Econometric Model for Measuring System-level Impacts of AI on United States Power Grids
Abstract
Data centers now account for 4.4\% of U.S. electricity demand, up from 1.9\% just six years ago, yet systematic quantification of AI's effects on grid stability remains limited. We exploit the staggered release of Large Language Models as a natural experiment, using difference-in-differences on a novel dataset to estimate how model training and inference affect local power grids. We find that AI activity significantly degrades power quality and increases fossil fuel generation near data centers, with high-intensity periods associated with 0.5–1 additional outages per year. Impact magnitude scales with model size.We construct counterfactuals examining alternative AI development trajectories: transitioning from cloud to edge inference substantially mitigates grid impacts, as does relocating data centers to lower-density regions with greater energy availability. We further assess scenarios involving on-site generation colocated with the data center, local inference with smaller models, and plateauing training intensity. Using our regression estimates, we map efficiency improvements to grid impact reductions and quantify downtime costs to hyperscalers from spatial agglomeration. Together, these results demonstrate that AI's grid effects are tightly coupled to both the technical trajectory of model development and the spatial organization of data-center infrastructure.
Looming Large (Loads): The Generation and Emissions Impacts of Data Centers
Abstract
In this study, we explore how the expansion of data centers within electricity balancing authorities reshapes electricity generation and emissions profiles across the United States. Data centers tend to represent geographically concentrated electricity demand shocks. As such, public discourse has focused on the environmental externalities of data centers and increases in electricity prices. Our research presents empirical evidence for the former. Our sample consists of a monthly panel data set (2010-2024) of balancing-authority-level electricity generation, demand, and power plant emissions, alongside information on data center locations, capacity, purpose, and operational dates. Using a staggered difference-in-differences model, we explore (1) how generation and emissions profiles change over time as balancing authorities add data centers, and (2) how additional megawatts of installed data center capacity affect those outcomes on the margin. We estimate these effects by fuel type and pollutant to obtain a detailed view of the environmental impact of data center capacity additions.Our preliminary results suggest that additional data center capacity increases natural gas generation while reducing coal generation. Changes in emissions profiles suggest increases in NOX and CO2 emissions and a decrease in SO2 emissions. These results highlight how large-load growth can simultaneously accelerate fuel switching from the dirtiest resources, while increasing emissions overall through higher immediate- and short-run electricity demand. Such findings are relevant for policymakers and the general public discourse around the environmental impact of data centers–a key variable for informed decision-making.
Artificial Intelligence and Decarbonization: When Does Digitalization Reduce Power-Sector CO2 Emissions?
Abstract
Digitalization and the rapid diffusion of artificial intelligence (AI) are transforming energy systems worldwide, yet their net climate impact remains unclear. Does new technologies and digitalization increase carbon emissions through increased electricity demand or enable decarbonization through efficiency gains? Using a global panel of 184 countries from 2000-2021, we examine how digitalization affects power-sector CO2 emissions per capita. We construct a digitalization index and employ fixed effects models, heterogeneity analyses, and a difference-in-differences design around the 2015 Paris Agreement. Results reveal three main patterns. First, higher digitalization is associated with increased power-sector emissions. Second, this effect is substantially stronger in fossil-intensive electricity systems. Third, in highly digitalized countries with cleaner electricity systems, digitalization instead contributes to lower power-sector emissions, a finding reinforced by post-Paris Agreement trends. The findings suggest that digitalization and by extension AI adoption is not inherently green. Its climate impact depends on crossing both a digital readiness threshold and a decarbonization threshold. This highlights the importance of complementary energy policies in shaping whether digital transformation accelerates or hinders climate mitigation.Discussant(s)
Dana Golden
,
Stony Brook University
Xiaochen Sun
,
New Mexico State University
Reid Taylor
,
Federal Reserve Bank of Dallas
Soran Mohtadi
,
University of Houston
Jesse Kaczmarski
,
University of Alaska-Fairbanks
JEL Classifications
- Q4 - Energy
- L9 - Industry Studies: Transportation and Utilities