Pricing Scarcity

The rapid expansion of data centers—especially those supporting artificial intelligence (AI)—is expected to significantly increase electricity demand over the coming decade. If AI proves transformative, meeting its long-run energy needs will require expanding generation, transmission, and distribution capacity through sound, market-oriented policies that promote efficient power development.

The more immediate challenge, however, is short-run stress on a system where electricity must be produced and consumed in real time. Wholesale markets reflect this through prices that rise when supply is tight, but most retail consumers face fixed rates and do not see these signals, weakening incentives to reduce usage when the grid is under strain.

A simple example illustrates the problem. When cooking dinner at home, you typically do not know—or respond to—the real-time price of electricity. If prices were temporarily higher due to grid stress, you might delay cooking or reduce usage. But with fixed prices, there is little incentive to adjust. When this behavior is repeated across millions of users, demand remains high precisely when the system is most constrained.

In most markets, scarcity is reflected in price. When hotel rooms are scarce, prices rise. When flights fill up, tickets get expensive. Electricity markets, at the retail level, do not function this way. Without price signals reflecting real-time conditions, demand does not adjust when the grid is under the most stress. As a result, periods of high demand can lead to congestion, reliability risks, and, in extreme cases, outages. The core issue is therefore not just limited capacity, but the failure to transmit real-time price signals at the retail level.

Restoring price signals that reflect real-time conditions can help address this issue by encouraging consumers—especially large users—to adjust demand when the grid is under stress. This improves the use of existing capacity and eases the transition to large-scale expansion.

Taken together:

1. In the long run, rising data center electricity demand will require substantial investment in infrastructure, supported by continued innovation.
2. In the short run, reliability challenges can be alleviated by strengthening price signals at the retail level to better manage demand.

The central policy question is therefore not whether to accommodate data center growth, but how to facilitate investment in capacity and demand management so the electricity system can adapt efficiently while preserving reliability.

There has been growing concern about the rapid expansion of data centers and their increasing demand for electricity. Although the precise magnitude of this growth remains uncertain, it is highly likely that energy consumption from data centers will rise much faster than it did over the past decade. The key question is not whether demand will increase, but by how much.

According to Berkeley Lab’s 2024 United States Data Center Energy Usage Report, sponsored by the U.S. Department of Energy, data centers are projected to account for between 6.7% and 12% of total U.S. electricity consumption by 2028, up from 4.4% in 2023. Much of this projected growth is driven by the expansion of AI-related computing.

These projections have led some commentators to raise alarm, arguing that data centers could overwhelm already overburdened power grids and threaten reliability for many Americans. But this framing misidentifies the source of the problem. The relevant policy question is not whether to accommodate data center growth, but how to facilitate the transition toward expanded electricity capacity. The constraint is not demand itself, but the system’s limited ability to expand supply quickly. Because infrastructure takes time to build, rapid increases in demand are difficult to accommodate in the near term.

Data from the U.S. Energy Information Administration (EIA) show that, once extreme weather events are excluded, the frequency and duration of outages have not systematically worsened alongside rising consumption. In fact, average annual outage duration has remained relatively stable over the past decade.

This reflects the electricity sector’s ability to adjust on the supply side. Over time, investment in generation capacity, expansion of transmission networks, improvements in grid operations, and technological innovation have allowed the system to meet growing demand. This should not be surprising. From wartime industrial surges to postwar suburban electrification, the grid has repeatedly adapted to large structural increases in demand for new technologies that make our economy work and grow. Data centers should be no different: Although they represent a new and potentially large source of consumption, they are not the first major structural shift the system has faced. The challenge is not whether the system can adapt, but how quickly it can do so.

In the short run, however, data center growth still presents real challenges. Electricity systems cannot expand instantaneously. Power plants, transmission lines, and other infrastructure take time to build, meaning that capacity is largely fixed in the near term. As a result, periods of high demand can put significant strain on the grid. The average construction timeline for non-nuclear power generation is 8-24 months in the US. Additional restrictions based on state or local regulations may extend that timeline.

Recent market outcomes already reflect these pressures. For example, capacity auctions in the PJM Interconnection—a Regional Transmission Organization serving much of the Mid-Atlantic and parts of the Midwest—have shown sharply rising prices for future delivery years. These increases signal tightening supply-demand conditions, with growing demand, including from data centers, which contributes to the strain.

A central problem is how electricity is priced. Wholesale markets generate prices that rise when supply is scarce or the grid is congested, reflecting real-time system conditions. Yet, most consumers never see these real-time signals. Instead, they pay preset retail rates that change infrequently and are largely disconnected from short-run conditions. When prices don’t rise during scarcity, demand doesn’t fall.

In most markets, prices adjust to manage limited capacity. Hotel rates rise during peak travel seasons, airline tickets become more expensive during busy periods, and concert tickets increase in price or sell out when seating is limited. In each case, higher prices encourage some consumers to shift timing or forgo consumption, helping allocate scarce capacity efficiently.

Electricity markets operate differently. When demand increases but prices remain fixed, consumption does not adjust to reflect limited capacity. The result is higher peak demand, greater congestion—transmission constraints that prevent electricity from flowing efficiently—and an increased risk of reliability problems or outages.

Over the longer run, the adjustment occurs through two well-understood channels: improvements in energy efficiency and expansion of supply. Technological progress is likely to improve the energy efficiency of computing and data center operations. At the same time, continued expansion of generation—particularly renewables such as wind and solar, along with storage and more efficient thermal technologies—will increase available capacity. Historically, sustained investment and innovation have enabled the system to accommodate large increases in demand.

In the short run, however, the problem is less about building new capacity and more about using existing capacity efficiently. This requires improving how price signals are transmitted to consumers, particularly large users such as data centers. When prices better reflect real-time system conditions, they can encourage adjustments in demand that reduce stress on the grid.

This can be achieved through a range of market and contractual arrangements. For example, large users could face pricing structures that vary with system conditions, such as real-time pricing, peak-based charges during periods of high demand, or interruptible service agreements, where utilities can reduce a customer’s load during grid stress. These mechanisms would encourage users to internalize the costs their consumption imposes during constrained periods, helping align private incentives with overall system reliability.

It is also important to recognize that consumers themselves benefit from the services powered by data centers, including cloud computing, digital medical records, GPS systems, and, increasingly, AI applications. These technologies generate substantial economic value and productivity gains across the economy. Policies aimed at restricting or delaying data center expansion in response to near-term pressures risk being counterproductive and threaten to undermine the American standard of living.

The International Energy Agency’s (IEA) 2025 Energy and AI report highlights a key contrast: data centers still account for a relatively small share of global electricity use, but their demand is growing very rapidly. In 2024, data centers consumed about 415 terawatt-hours (TWh), roughly 1.5% of global electricity consumption. However, their impact is much more visible at the local level because they tend to cluster geographically, primarily in regions where infrastructure, transmission access, and favorable regulatory environments are already in place. The United States alone accounts for about 45% of global data center electricity use, followed by China (25%) and Europe (15%), and in some regions data centers already represent a large share of total demand.

Looking ahead, global electricity demand from data centers is expected to more than double by 2030, reaching around 945 TWh. Much of this growth is driven by AI, which requires far more computing power than traditional digital services. But AI should be understood not as an anomaly, but as the next phase of a long pattern of technological expansion—akin to railroads, electrification, and the internet—each of which placed new demands on infrastructure while ultimately expanding economic capacity. In the United States, data centers are projected to account for nearly half of total electricity demand growth through 2030. By the end of the decade, their electricity use could exceed that of major energy-intensive industries such as aluminum, steel, cement, and chemicals combined. Beyond 2030, projections become more uncertain, depending on how quickly AI adoption expands and how much efficiency improves.

A defining feature of data center demand is its geographic concentration. In the United States, nearly half of all data center capacity is in just a few regions, including Northern Virginia, Texas, and California. This means that even if national-level impacts appear manageable, local grids can experience significant pressure.

Data centers consume electricity primarily for two functions.

First, most of the electricity is used for computing itself—running servers and specialized hardware. AI workloads rely on high-performance chips such as GPUs that operate continuously and at high intensity. This is not inherently inefficient. Rather, it reflects the computational intensity required to deliver advanced digital services that is transforming the world for the better. To give a sense of scale, a large AI-focused data center can consume as much electricity as a small city—on the order of 100,000 households—while the largest facilities under development may reach the equivalent of 2 million households. Training advanced AI models often requires thousands of these chips running non-stop for weeks.

Second, a substantial share of the electricity is used for cooling. Servers generate large amounts of heat and must be kept within safe temperature ranges to function reliably. Cooling systems—along with power conditioning, backup systems, and networking equipment—add significantly to total energy use. As AI increases the power density of equipment, cooling becomes more demanding and technologically complex.

In summary, while data centers still represent a modest share of total electricity use, their rapid growth, geographic concentration, and high-power intensity make them a critical pressure point for electricity systems. Their demand is driven primarily by high-performance computing and cooling needs, and accommodating this growth will require coordinated expansion of generation and transmission to deliver reliable power at scale.

Unlike almost any other commodity, electricity cannot be meaningfully stockpiled. At every moment, supply must closely match demand to keep the system stable. This distinguishes electricity from most other goods and creates unique operational constraints that need to be considered when making policy decisions.

In many industries, firms can produce at a steady pace and rely on inventories to absorb fluctuations in demand—like a retailer stocking shelves ahead of busy periods. Electricity systems cannot operate this way. Because large-scale storage remains limited and costly, electricity cannot be stockpiled in advance. Instead, production must continuously adjust to real-time demand.

At the same time, consumers expect electricity to be available instantly and reliably. When we turn on a light or plug in a device, we expect it to be there instantly and without interruption. This combination of real-time balancing and high reliability expectations makes coordination essential.

In effect, the electricity system must function like a tightly synchronized network, where generation, transmission, and consumption are constantly aligned. The grid must operate within narrow frequency and voltage tolerances; even small imbalances between supply and demand can cascade into system instability or outages. Maintaining this balance requires not only physical infrastructure, but also well-designed market and pricing mechanisms that help match supply and demand at every moment.

Electricity markets in the United States are organized in different ways depending on the state. In some states, a single utility owns and operates power plants, transmission lines, and distribution networks, with prices set by regulators. In others, parts of the system—especially electricity generation—are opened to competition, allowing multiple firms to produce and sell power. Many states fall somewhere in between, combining regulated utilities with competitive wholesale markets.

Despite these differences, all electricity systems operate through two interconnected layers: wholesale markets and retail markets. The wholesale market manages large-scale generation and ensures that supply and demand are balanced in real time. The retail market is where electricity is sold to homes, businesses, and industrial users.

Wholesale markets function much like a centralized coordination system. Power plants submit bids based on their costs, and grid operators—such as Regional Transmission Organizations (RTOs) or Independent System Operators (ISOs)—decide which plants should run at any given moment. Prices reflect the cost of supplying one additional unit of electricity at a specific location, taking into account factors like congestion on transmission lines. These markets operate both a day ahead (planning production) and in real time (adjusting every few minutes to keep the system balanced). In some regions, additional markets exist to ensure enough capacity is available in the future—a crucial function for maintaining long-term reliability and investment signals.

Retail markets, by contrast, determine what final consumers pay. In many areas, prices are set through regulatory processes and change only occasionally. This means that most households and businesses pay stable, predictable rates rather than prices that fluctuate hour by hour. In states with more competition, consumers may choose among suppliers, but even there, many contracts smooth or fix prices over time.

This creates an important disconnect. Wholesale prices can change rapidly—rising when electricity is scarce and falling when it is abundant—but most retail customers do not see these fluctuations. As a result, consumers often have little incentive to adjust their usage when the grid is under stress.

Arizona provides a useful example of this structure. The state largely relies on vertically integrated utilities such as Arizona Public Service (APS), Salt River Project (SRP), and Tucson Electric Power (TEP), which generate, transmit, and distribute electricity. Retail prices are set administratively and remain relatively stable over time, although some time-of-use pricing reflects differences between peak and off-peak periods, such as higher rates on summer afternoons when air conditioning demand surges and the grid is under the greatest strain.

At the same time, Arizona participates in broader wholesale markets through the Western Energy Imbalance Market—a multi-state system spanning much of the western United States—which balances supply and demand across the region in real time. Wholesale prices in this market respond continuously to system conditions, but most retail customers are shielded from this short-term volatility.

A simple way to think about this is that the wholesale market is highly responsive—constantly adjusting prices to keep the system in balance—while the retail market is largely fixed and predictable. This gap between real-time conditions and the prices consumers face plays a central role in how electricity demand responds to periods of scarcity, particularly during times of peak stress on the grid.

Most electricity consumers are not exposed to real-time prices. A typical electricity bill arrives once a month, and even when time-of-use plans are available, consumers rarely track how prices vary hour by hour. In practice, most people face preset rates under which utilities commit to providing electricity on demand at administratively determined prices.

This is unusual compared to most markets, where prices change to reflect the current conditions. In a grocery store, for example, prices are clearly posted and often adjust when demand rises, helping prevent shortages. If prices did not adjust, shelves would empty quickly. Electricity does not work this way at the retail level. Instead, consumption decisions are largely disconnected from real-time system conditions.

A simple everyday example illustrates the issue. Suppose you are cooking dinner in the evening. You are not aware of the price of electricity at that moment—it is fixed in your contract. Now imagine you receive a notification that electricity prices are temporarily much higher due to a heat wave or grid congestion. You might decide to delay cooking or reduce usage, essentially paying time for money. Not everyone would respond this way, but some would, and that response would reduce pressure on the system as a whole during peak periods.

At its core, this is a basic supply-and-demand problem. Prices convey information about scarcity. When prices rise, some consumers reduce or shift demand, allowing the market to balance. When prices are fixed and cannot adjust, demand remains high even when supply is constrained.

In most markets, this leads to shortages—as in housing markets with rent controls, where below-market pricing results in persistent shortages or during the 1970s, when gasoline price controls produced long lines and empty pumps. In electricity, the consequences are more severe. When demand exceeds what the system can supply, the result is an outage—a system-wide shortage rather than a localized shortage.

Importantly, this problem is not simply about limited capacity. Many industries operate near capacity at times, but flexible prices help allocate scarce resources without disruption. In electricity markets, however, fixed retail prices prevent this adjustment, making the system more vulnerable during periods of peak demand.

Concerns remain about whether the scale of data center growth could overwhelm such adjustments. However, evidence from the IEA suggests otherwise. Their analysis indicates that if data centers shifted or reduced demand during just a small fraction of the most stressed hours—around 1%—existing systems could accommodate projected growth through 2035. In practical terms, this means that relatively small changes in when electricity is used—not how much is used overall—can materially reduce system strain.

These estimates should be interpreted cautiously, but they highlight an important point: the present challenge is largely one of coordination and incentives. With better price signals, some demand can shift away from peak periods, easing strain on the grid. Emerging technologies, including AI-driven load management systems, may further enhance this coordination by optimizing when and how electricity is consumed.

Electricity systems have adapted to large structural shifts before—rural electrification, postwar suburban growth, the consumer electronics boom. None of those transitions were seamless, but in each case, the system expanded because investment followed demand. The challenge is not to constrain growth of new technologies that make our lives better, but to ensure that markets and policies allow the system to respond efficiently.

Data centers are not overwhelming the grid. They are exposing a system that does not price scarcity. The solution isn’t to slow the data centers. It’s to fix the pricing. The solution in the near term is straightforward: better price signals, fewer regulatory barriers to flexible rate structures, and demand that responds when the system is under strain.