Increases in US electricity prices since 2020 have coincided with a sharp rise in demand from data centres. This column analyses the impact of the surge in data centre growth since 2021 on electricity prices across US utility service territories. It finds that data centres raise electricity demand and electricity prices, particularly for households. However, residential price increases are close to zero in territories served by nonprofit utilities. Policymakers should expand generation capacity, encourage data centres with flexible electricity usage, and consider tariffs on large-load customers.
Electricity affordability has moved to the centre of the US political debate (Saad 2026). Between 2020 and 2025, the average residential electricity price rose from 13.15 to 17.30 cents per kilowatt-hour, an increase of almost one-third in nominal terms (US Energy Information Administration 2026). The increase has many possible causes, including fuel costs, extreme weather, and an ageing grid infrastructure (Wiser et al. 2025). But rises in electricity prices have also coincided with the rapid emergence of a new source of demand: data centres.
Data centres accounted for 4% of US electricity demand in 2023, and their consumption is projected to triple by 2030 (Shehabi et al. 2024). Their expansion is also geographically concentrated, creating much greater pressure on electricity systems in some areas than national figures suggest (Ferriani and Gazzani 2026). This has raised important questions. When data centres connect to the grid, how does this affect local electricity prices? If prices go up, who pays?
These questions are already generating protests and political responses (Brown 2026, Goldmacher 2026). For example, 71% of Americans say they would oppose the construction of a data centre in their local area, citing several concerns, including higher utility bills (Jones 2026, McCabe 2026). New York has introduced a one-year moratorium on permits for new hyperscaler data centres (New York State 2026), while regulators and grid operators are developing special rules for connecting large loads and allocating the resulting costs (PJM Interconnection 2026, Federal Energy Regulatory Commission 2026, Ramanan and Donalds 2026).
Despite the concerns about rising electricity prices and frustration with data centres as a possible culprit, there is little evidence showing that one is causing the other.
Data centres often fall into the category of ‘large-load’ customers, which can also include manufacturing plants and other industrial facilities. The effect of a new large-load customer on local electricity prices is theoretically ambiguous (Bistline 2026). A utility with spare capacity can serve a high-demand customer with little effect on average rates. Prices may even go down if utilities can spread their fixed costs over a larger volume of electricity sales. Conversely, if capacity is scarce, the same load can raise peak demand, force procurement of higher-cost generation, or trigger grid investments that may be recovered from ratepayers. Hourly demand schedules also shape the costs that large-load customers impose on the system (Knittel et al. 2025).
Empirical studies have so far been mixed. Analyses of recent retail electricity price trends find little evidence that points to load growth, including from data centres (Wiser et al. 2025). On the other hand, ratepayer advocates argue that utilities are already passing Big Tech’s costs onto the public (Martin and Peskoe 2025). Recent working papers on data centres and retail prices report limited or conflicting effects (Feher et al. 2025, Alvarez et al. 2026, Watten et al. 2026, Meeks et al. 2026).
In Scalera et al. (2026), we provide causal estimates based on variation across utility service territories. The context is one of rapid and highly concentrated growth (Figure 1). In the US, an average of 216 new data centres became operational each year between 2021 and 2024, up 70% from the 2015-2020 pace. Private firms spent an average of $18 billion per year on data centre construction, triple the pre-2021 average. The growth has been extremely heterogeneous. The median utility territory added no data centre at all over the previous decade, while utilities in northern Virginia collectively added over 290.
Two features of our analysis matter. First, we work at the level of utility service territories. This is the level at which retail rates are set, not the county or state level. We match average electricity price data to roughly 1,200 geographically defined utility territories over 2015-2024 and merge that to a panel of about 2,700 operational data centres. Second, we address the fact that data centres are not located randomly. They select into places with cheap land, proximity to clients, and favourable electricity price trends, which biases naive comparisons. To address this endogeneity, we use a shift-share instrumental variable. We interact the national surge in data centre construction spending in the ‘post-AI’ years from 2021 onwards with each territory’s pre-existing share of national fibre infrastructure. Our rationale is that pre-AI fibre influenced where technology companies planned and would later build their data centres but did not directly affect trends in post-AI electricity prices. The identifying assumption is that pre-AI fibre affected post-AI prices only through its influence on data centre location. We find that pre-AI fibre does not predict earlier price trends or several other sources of post-AI local electricity demand.
Figure 1 National and regional trends in data centre growth
A) National
B) By utility territory
Figure 2 shows the descriptive pattern that motivates our research design. Utility territories with high and low pre-AI fibre shares had similar price trends before the AI period, with real prices declining steadily in both groups. During the post-AI period, the trends diverged. In high-fibre territories, where the data centre development was concentrated, real prices began to increase, while real prices in low-fibre territories continued to fall.
Figure 2 Trends in residential electricity prices: High vs. low pre-AI fibre territories
The divergence shown in the figure is descriptive and does not by itself establish causality. Under the identifying assumptions described above, our shift-share design allows us to estimate the effect of data centre growth on electricity prices. Three main results emerge.
The first result is that data centres meaningfully raise electricity prices and more so for households compared to other classes of customers. An additional data centre increases average residential prices by 0.182 cents/kWh. A one-standard-deviation increase in the post-AI data centre stock, corresponding to ten facilities, raises residential prices by 13% relative to the post-AI mean. Size matters. A single average hyperscaler facility of about 60 acres raises prices by roughly 4%. Industrial and commercial prices rise too. But the increase for residential customers is 30-40% larger in absolute terms, on top of already higher average rates. Households bear a disproportionate share of the burden relative to average business customers, which can include data centres themselves.
The second result points to demand as an important mechanism. An additional facility raises summer peak demand by about 34 MW, precisely when the grid is most strained. We also analyse a more limited sample of utilities for which grid spending data are available. Our estimates show that an additional data centre leads utilities to increase expenditures on distribution infrastructure by about $11.6 million. We expect this channel to matter more in the coming years as capital costs enter rate cases with a lag.
The final result is that the market and institutional context matters. The residential price effect is close to zero in territories served by nonprofit utilities (cooperatives, municipal, and state-operated providers). One possible explanation is that nonprofit utilities rely more heavily on fixed charges and less on volumetric rates exposed to market conditions. Effects are also smaller where utilities held surplus energy before the AI boom, since utilities with excess procured energy can serve a new large load without turning to the wholesale market. Conversely, territories whose prices were historically most sensitive to demand experience more than double the average price increase. In the three states with meaningful retail competition in our sample (Illinois, Ohio, and Texas), the effect is around 44% smaller, although this estimate is less precise. Lastly, we find little evidence that states with more wind and solar generation experience larger price increases, contrary to the concern that renewable-heavy grids are ill-equipped to absorb data centre demand. Causal inference for these heterogeneous patterns is limited, since we do not independently identify the modifying characteristics.
These results help explain the political unrest involving data centres and why dismissing the opposition as NIMBYism is overly simplistic. The gains from AI are real but concentrated. Recent usage-based estimates suggest that AI-related time savings are concentrated in rich countries and, within countries, in higher-paid occupations (Fan and Nguyen 2026). The costs, by contrast, are local and potentially regressive. Electricity expenditures typically absorb a larger share of the budgets of those with lower incomes (Brown et al. 2020). And our estimates imply that data centre growth raises the bills of households, many of which likely capture little of its direct value.
The policy implication is not that data centres should be blocked. Rather, policymakers should address the issues from both sides of the market. From the supply side, expanding generation capacity, including renewables, and reducing grid interconnection times may reduce energy scarcity. From the demand side, data centres with flexible electricity usage hours would place less stress on the grid. Lastly, large-load tariffs that ensure data centres pay for the grid infrastructure upgrades that they require would protect households from bearing the cost.
The lesson travels. The mechanisms underlying our US estimates are relevant for Europe, despite the institutional and market differences. Rapid load growth is more likely to create pressure where spare capacity is limited, and utilities will need to build new infrastructure. Europe therefore has an opportunity to grow its grid and establish how costs will be allocated before its planned data centre expansion proceeds. As the EU integrates data centres into its digital sovereignty agenda (von Thun 2026), grid upgrades, tariff design, and cost allocation deserve as much attention as permitting and investment.
Source : VOXeu
US President Donald Trump welcomes Chinese President Xi Jinping to the White House on Thursday…
- Australia said on Thursday an OpenAI agent breached a government health data portal in…
Europe's investment debate tends to ask how much capital can be mobilised and how quickly.…
Most of what we hear about artificial intelligence comes through the lens of advanced economies:…
It has long been argued that even relatively small increases in global migration could result…
Of the roughly 3.74 billion people in the global workforce, about 400 million, or 11 percent, work…