Data center power struggle

By Josh Nye

U.S. electricity demand is on the rise

U.S. electricity consumption is rising anew after more than a decade of stagnation. The biggest contributor to the increase is the proliferation of energy-hungry data centers, driven by growing demand for cloud computing, and AI model training and inference (using trained models to make predictions on new data).

Utilities and hyperscalers (companies investing heavily in AI computing power) are scrambling to add capacity, but supply is struggling to keep up with demand. Rising consumer electricity prices have become a political issue, and local resistance to data center projects is growing.

Power might be the single biggest constraint to tech companies’ ambitious capital expenditure (capex) plans: the International Energy Agency (IEA) thinks one-fifth of global data center investment is at risk of delay due to grid bottlenecks. China doesn’t face this challenge—it has invested heavily in new capacity and now generates more than twice as much electricity as the U.S.

The White House faces a tough balancing act to improve electricity affordability without hindering domestic investment and shrinking the country’s sizeable lead in the data center buildout.

After a decade of stagnation, U.S. electricity consumption is rising again

Electricity end use, 12-month moving average , terawatt-hours (TWh/yr)

Data centers key to rising power consumption

The U.S. Department of Energy (DOE) estimates data center power consumption grew by 18 percent annually between 2018 and 2023, while overall electricity demand was flat. The share of electricity used by data centers more than doubled to 4.4 percent from 1.9 percent over that period, and the DOE estimates this could rise to between 6.7 percent and 12 percent by 2028 depending on factors such as installed computing capacity, energy efficiency, utilization rates and cooling requirements. While cutting-edge semiconductors are becoming 38 percent more energy efficient every year, installed capacity is growing by 2.3 times annually according to nonprofit research firm Epoch AI, driving overall consumption higher.

In response to growing demand, electricity investment has accelerated, rising at an inflation-adjusted rate of six percent annually over the past five years. It accounted for nearly five percent of private capex in 2024, the highest share since 1985.

U.S. electricity investment has accelerated in recent years

Private fixed investment in electric power

That investment helped utility-scale generation capacity grow at the fastest pace in more than a decade; however, much of the increase came from solar and wind projects. The additional capacity is less impressive when adjusting for intermittency and reliability. Reduced federal funding for green energy projects could slow the pace of new investment.

Adding capacity remains a challenge

The Lawrence Berkeley National Laboratory estimates that more than 10,000 power projects were seeking grid interconnection in the U.S. as of the end of 2024. That represented 1,400 gigawatts (GW) of new generation (about 35 times the net capacity added in 2024) and 890 GW of storage.

History, however, suggests just a fraction of those projects will ultimately become operational. Only 13 percent of interconnection requests between 2000–2019 reached commercial operations by the end of 2024, while 77 percent were withdrawn and 10 percent are still active. Billions of dollars in clean energy projects were canceled last year amid federal funding cuts, according to national nonpartisan business association E2.

Making matters worse, the time it takes to add new power plants to the grid has lengthened considerably. It has risen from less than two years in the early 2000s to more than four years recently. Adding new transmission infrastructure can take a decade.

It takes roughly 4.5 years to bring new power generation online in the U.S.

Median time from interconnection request to commercial operations (months)

Meanwhile, electricity demand can rise much more quickly: data centers can be built in about two years. The Federal Energy Regulatory Commission (FERC) made rule changes in 2023 to address the backlog of interconnection requests and streamline the process for adding new resources to the grid, but in our opinion, it’s too early to judge their effectiveness. Further rule changes were proposed in 2025.

To get around these delays, more data centers are being built with their own local power generation. Natural-gas-fired turbines are particularly popular, but equipment prices have doubled in recent years, and delivery times now stretch out to several years. Nuclear is making a comeback, with Microsoft and Constellation Energy planning to restart a reactor at Three Mile Island. Google and NextEra Energy are looking to reopen a plant in Iowa—but there is only so much latent capacity. Other patchwork solutions include fuel cells and conversion of Bitcoin mining sites.

Rising electricity prices add to affordability headwinds

Most new data centers still get their power from the grid, and the cost of upgrading electricity infrastructure is generally spread across ratepayers, including households and small businesses. While some utilities charge higher rates for data centers, pricing doesn’t fully cover the cost of new infrastructure. And when utilities purchase electricity on the open market, data center-driven demand pushes prices higher for all customers.

That has contributed to a one-third increase in residential electricity prices over the past five years—more than 1.5 times faster than the overall Consumer Price Index (CPI). To be fair, the sharpest increase in prices occurred before data center construction really began to accelerate. Russia’s invasion of Ukraine, which caused a spike in natural gas prices, was a key source of energy inflation in 2022.

Data centers are contributing to higher electricity prices, but are not the only factor

The data center effect is notable. Bloomberg reports wholesale electricity prices increased by as much as 267 percent over five years in areas located near significant data center activity. A Carnegie Mellon study estimates data centers and cryptocurrency mining could lead to an eight percent increase in average electricity generation costs by 2030. The increase could exceed 25 percent in markets with a high concentration of data centers, such as Virginia.

Rising electricity prices are becoming a political issue. They likely contributed to Democratic victories in some state elections last fall, in our opinion. As part of a growing focus on affordability, the White House has called on tech companies to “pay their own way” and is reportedly working on an agreement that would see data center developers fully cover the cost of new energy infrastructure. Some hyperscalers are already pledging to do so, although an opaque rate-setting process makes that difficult to verify.

More broadly, implementing such an agreement will require coordination between grid operators, utilities, state regulators, tech companies and other data center developers.

For households, electricity represented just 2.5 percent of the CPI basket, so it only contributed 15 basis points to the latest year-over-year inflation figure. But it is a highly visible, recurring and necessary expense.

Thus, electricity prices might make a greater contribution to perceived affordability challenges, which have sapped consumer confidence. Just as electricity prices were an issue in last year’s state elections, they could play a role in this year’s midterms, particularly in states that have seen substantial price hikes.

Higher electricity prices aren’t just putting politicians’ jobs at risk. Industry research firm Data Center Watch reports that 20 projects representing nearly $100 billion in potential investment were blocked or delayed by local opposition in the second quarter of 2025 alone.

China’s power surge

The U.S.’s power challenges contrast with China, where electricity generation has grown by nearly nine percent annually over the past 25 years. Since 2021 alone, China has added more generation capacity than the U.S. has in its history. A major investment push, a streamlined approval process and an “all of the above” approach embracing both renewables and fossil fuel generation have helped the country add capacity at an unprecedented pace.

China’s rapidly rising electricity supply is an advantage in the AI race

Electricity generation, terawatt-hours (TWh)

BloombergNEF predicts China will continue to extend its lead over the next five years, adding almost six times as much new capacity as the U.S. While data center investment by Chinese tech companies hasn’t kept pace with U.S. hyperscalers and access to leading-edge chips is a challenge, we believe China’s significantly greater power capacity could be a key advantage in the race for AI supremacy.

Power constraints could crimp spending plans

Back in the U.S., the latest round of quarterly earnings saw yet another upward revision to AI capex estimates. Consensus for 2026 spending by the big five hyperscalers—Amazon.com, Alphabet (Google), Meta Platforms (Facebook), Microsoft and Oracle—now exceeds $675 billion, a more than 60 percent increase relative to last year. But if tech companies are unable to follow through on that investment due to power constraints—and thus unable to monetize growing cloud service backlogs in a timely manner—we believe valuations could be at risk.

This issue is well known and should be reflected to some extent in current share prices and earnings expectations, but there could be room for disappointment and repricing if markets are too optimistic about the potential for new capacity to resolve power bottlenecks.

Furthermore, with the computing power required to train the most advanced frontier models growing by a factor of five annually according to Epoch AI, capacity constraints could act as a headwind to model development and the pace of improvement in AI capabilities more generally. Less capacity for inference would also mean slower diffusion of productivity-boosting AI tools to other sectors of the economy.

---

---