AI data centers are reshaping America’s power grid faster than utilities can adapt, and this new analysis suggests the cost could be higher electricity prices, more emissions, and growing pressure to expand renewables and transmission.
The Federal Reserve Bank of Dallas warns that the expansion of artificial intelligence (AI) data centers could increase wholesale electricity prices by up to 50% by 2028 in a high-utilization model scenario.
In a recent working paper published by the Federal Reserve Bank of Dallas, the authors examined how high-utilization computing is reversing two decades of stable electricity demand in the United States. Using an hourly, unit-level model of wholesale power markets, they estimated that existing data centers have already raised average modeled wholesale prices by 2% to 6% nationwide. This reflects a shift in the grid from predictable residential demand cycles to continuous, high-intensity industrial loads that strain current generation capacity.
Disruption of Stable Demand Patterns by Data Centers
For nearly two decades, electricity demand in the United States remained relatively stable. The rise of hyperscale data centers has ended this trend. Unlike homes or businesses, which use power in variable patterns, AI infrastructure creates concentrated demand in large, single blocks. These facilities often cluster in specific areas to stay close to fiber networks and skilled labor, putting heavy pressure on local power grids.
Data centers can operate at high utilization rates, maintaining demand even during periods of grid stress. In the model, this can push utilities and power markets to rely on higher-cost marginal generation when supply is tight.
Advanced Techniques for Analyzing Energy Market Changes
To analyze these changes, the authors Owen Kay, Robert Reaser, and Reid Taylor built a unit-level model of the continental United States to track the power grid hour by hour. They integrated data from the Environmental Protection Agency (EPA)’s Continuous Emission Monitoring System, covering about 3,368 thermal generation units. This dataset provides details on fuel use and output, enabling cost estimation based on fuel prices, maintenance, and environmental permits.
The locations and power capacities of existing and planned data centers were tracked in the “Cleanview” database. This enabled simulations of how different levels of computing activity affect electricity markets. The report evaluated multiple operating scenarios, including “Full Capacity,” “Mid-Range,” “Mid-Range Peak,” and “Low Capacity,” to assess their effects. For example, in the “Full-Capacity” scenario, every facility operated at 100% capacity continuously, creating an upper-bound case rather than a likely real-world outcome.
The authors also accounted for outages due to maintenance and unexpected failures. By comparing current grid conditions with counterfactual scenarios without data centers, the model estimated the marginal effect of data centers on electricity prices and pollution levels. The authors note that these are partial-equilibrium models, not direct observations of a full-market outcome.
Implications of Data Center Growth
The analysis showed that data centers shifted about $12.6 billion from wholesale electricity buyers to power suppliers between 2021 and 2025Q3 in the report’s mid-range peak scenario. Price increases were highest in regions with dense tech infrastructure and tighter supply conditions. In Northern Virginia’s SRVC region, prices often rose by more than 10%. These spikes typically occur during spring and fall when utilities schedule maintenance outages, but steady AI demand leaves little spare capacity.
Environmental impacts were significant. By late 2025, data centers were linked to an additional 5 to 16 million tons of carbon dioxide (CO2) emissions per month. When grid capacity was strained, incremental demand was often met by fossil-fuel generation, depending on the region and hour.
Coal-fired generation fulfilled demand in parts of the Midwest, while natural gas served as the marginal source in regions such as California and New England. The report also emphasizes that slower-than-expected wind and solar build-out would substantially amplify these price and emissions effects, because fewer low-marginal-cost renewable resources would be available to absorb rising demand. Without faster expansion of renewable energy, the carbon footprint of AI infrastructure could grow alongside its capacity.
Optimizing Data Center Siting for Grid Efficiency
The analysis indicated that current data center siting is not optimized for the power grid. Using a “social planner” model, the authors identified areas where computing could be placed to minimize energy costs, highlighting regional imbalances.
The ERCOT (Texas) region could support about 49% of the national compute workload due to abundant energy resources. In contrast, Virginia, which currently handles about 39% of data center activity, would ideally host less than 1% under a cost-efficient allocation.
This approach, known as “virtual transmission,” shifts workloads to regions with surplus capacity and is mainly suited for non-time-sensitive tasks, such as training large language models. However, current siting decisions are driven more by tax incentives and fiber connectivity than by grid capacity. This creates a mismatch between where compute is deployed and where power is available. Addressing this gap would require greater flexibility in workload placement, moving compute away from congested hubs toward regions with underutilized energy resources.
Strategic Considerations for Future Energy and AI Integration
As the world looks toward 2028, the economic impact of AI infrastructure is becoming more visible. Rising electricity costs could add up to 0.6 percentage points to annual inflation in an extreme full-capacity scenario if data center growth continues without a matching expansion in power capacity. Under the report’s more moderate central scenarios, the estimated inflation effect is much smaller. Governments must balance the benefits of AI leadership against the potential for higher household energy costs.
The report highlights the need for closer coordination between the AI and energy sectors, including long-term planning for generation and transmission to ensure that computing capacity does not outpace grid expansion. Transmission constraints were identified as a key driver of recent price increases. Without new interstate power lines, some regions are likely to face sustained cost pressures. For developers and operators, energy has become a core strategic constraint shaping the future of AI.
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.