Macro

The United States plans to build up to 250 new nuclear reactors over coming decades to meet surging electricity demand from AI, data centers, and manufacturing. The push revives nuclear power as a low-carbon alternative, focusing on advanced reactors and small modular reactors with passive safety systems. Federal programs like the DOE's Launch Pad support development, with companies such as TerraPower, Kairos Power, and X-energy building demonstration reactors backed by the Advanced Reactor Demonstration Program. Tech giants Google and Meta have signed power purchase agreements. Critics cite unresolved safety, cost, and waste challenges, while experts stress the need for extensive testing and regulatory review of new designs.

Global data center demand could nearly triple from 82 to 220 gigawatts between 2025 and 2030, driven by AI. AI-related demand is expected to grow 3.5 times to 155 gigawatts, accounting for 70% of total demand, while non-AI demand rises 1.7 times to 64 gigawatts. Cumulative global investment in data center infrastructure (excluding IT hardware) may exceed $1.7 trillion through 2030. The United States leads in investment, with Northern Virginia a major hub, followed by China, where expansion shifts west under the 'Eastern Data, Western Computing' strategy. Europe sees growth in the UK, France, and Nordic countries, while Singapore expands into Malaysia. Levelized costs vary from ~$200 per MWh in Chinese demand centers to ~$380 in London, with electricity prices and cooling needs as key drivers. Export controls on advanced chips affect China's energy efficiency.

Data centers drove half of all new U.S. electricity demand growth in 2025, making them the single largest contributor to the country's rising power appetite, according to the International Energy Agency. Overall U.S. energy demand grew 2% in 2025, the second-highest rate since 2000. The IEA expects data centers to account for about half of demand growth through 2030. Goldman Sachs projects U.S. data center power demand will more than double from 31 gigawatts in 2025 to 66 gigawatts in 2027, with AI-driven servers consuming roughly 60% of facility electricity. However, only 50-60% of planned capacity is expected to come online on schedule, creating reliability risks in the Mid-Atlantic, Mid-Continent, and Northwest. Over $61 billion was invested globally in data center construction in 2025, but community opposition over power and water use is mounting, with a Pew survey showing mixed public sentiment.

US data center electricity demand could more than double by 2030, rising from 167 terawatt-hours in 2023 to about 376 TWh, an increase sufficient to power 20 million average US homes for a year. This shift is turning power from a background cost into a frontline constraint for AI, making battery storage a critical component of AI infrastructure. Storage acts as a buffer between energy producers and consumers, moving power through time to meet demand spikes. Developers plan to add 24 GW of utility-scale battery storage in 2026, second only to solar, as storage becomes a reliability tool for the AI age.

The Federal Energy Regulatory Commission voted unanimously on Thursday to order six regional grid operators serving nearly two-thirds of the U.S. to fast-track interconnection requests from AI data centers and other large electricity users, mandating that they justify or rewrite tariffs within 60 days using Section 206 of the Federal Power Act. Grid operators have 30 days to report spare generating capacity and 60 days to defend or revise rates, with FERC Chair Laura Swett calling the vote 'historic' and saying 'This FERC is not the old sleepy agency that it has been in the past. We can’t afford to be, and our country cannot afford for us to be.' The order requires data centers to pay full grid upgrade costs and, to qualify for fast-track processing, to bring their own power or curtail demand during high-stress periods. It also directs operators to consider 'alternative transmission technologies' like solid-state transformers or superconducting lines and to be more accommodating to behind-the-meter power. Energy Secretary Chris Wright, who had urged the action eight months ago to compete with China, hailed the move to 'remove barriers, accelerate development and ensure America has the affordable, reliable and secure energy needed to power a new era of prosperity,' aligning with President Donald Trump’s AI Action Plan. The order follows a December step allowing direct power plant connections. Concerns persist about rising electricity prices, water use, and environmental impacts as data centers now account for about 5% of U.S. electricity demand, potentially tripling by 2035. Wholesale electricity rates are up as much as 267% compared with five years ago, according to Bloomberg. PJM Interconnection raised power costs by 75.5%, largely blamed on AI data centers, and Maryland filed a complaint with FERC over a $2 billion charge for infrastructure upgrades not directly benefiting the state. At the end of 2023, grid connection requests for power plants exceeded the total capacity of the existing power plant fleet. Tech companies and developers, unable to connect in a timely manner, have been turning to on-site power out of desperation. FERC emphasized it does not intrude on state authority over siting, generation, or retail rates, leaving states to protect residential and small-business customers from cost shifts. Texas, outside FERC’s jurisdiction, approved ERCOT’s 'Batch Zero' process on the same day to speed large-load studies for projects of at least 75 MW, with ERCOT tracking over 438,000 MW of large-load requests, nearly 89% from data centers. Tech giants including xAI, Google, Microsoft, Meta, Amazon, Oracle, and OpenAI have signed a pledge to build new power sources and cover infrastructure costs. Over 4,000 data centers operate in the U.S., with 3,000 more planned, though construction lags due to permitting and supply delays. A J.P. Morgan report last month, based on satellite imagery, found that over 60% of planned data center capacity scheduled for 2027 has not yet broken ground, with another 7% already delayed. The Trump administration on Wednesday said it would pay $765 million to wind developer Invenergy to cancel offshore wind leases near California, Maine, and New York, with the company saying it would use the money to build natural gas plants in the Midwest and geothermal projects in the West; one of Invenergy’s wind projects would have generated as much as 2.4 gigawatts of power. Altogether, the Trump administration has now spent about $2.6 billion to scuttle offshore wind developments. Communities increasingly resist data center developments in drought zones and quiet rural areas, citing water shortages and noise pollution. FERC’s order requires data centers to pay full grid upgrade costs and, to qualify for fast-track processing, to bring their own power or curtail demand during high-stress periods. It also directs grid operators to consider 'alternative transmission technologies' like solid-state transformers or superconducting lines, and to be more accommodating to behind-the-meter power. The order builds on a December step allowing direct power plant connections. Tech companies and developers welcomed the faster connection process, but utilities, states, and clean energy advocates had worried about loss of authority, though FERC said it leaves states in control of retail rates and terms. Robert Montejo, a lawyer representing data centers, said AI 'has fundamentally changed the electricity landscape.' The Edison Electric Institute said the order builds on existing processes while supporting flexibility. Jeff Dennis of the Electricity Customer Alliance said it responds to big power users and state regulators. Rob Gramlich, an energy consultant, said states should quickly develop rules to prevent cost shifts to residential and business customers, warning FERC could assert broader jurisdiction if states don’t act. Texas, outside FERC’s jurisdiction, approved ERCOT’s 'Batch Zero' process on the same day to speed large-load studies for projects of at least 75 MW, with ERCOT tracking over 438,000 MW of large-load requests, nearly 89% from data centers. Tech giants including xAI, Google, and Microsoft have signed a pledge to build new power sources and cover infrastructure costs. The Trump administration on Wednesday said it would pay $765 million to wind developer Invenergy to cancel offshore wind leases near California, Maine, and New York, with the company saying it would use the money to build natural gas plants in the Midwest and geothermal projects in the West; one of Invenergy’s wind projects would have generated as much as 2.4 gigawatts of power. Altogether, the Trump administration has now spent about $2.6 billion to scuttle offshore wind developments. Communities increasingly resist data center developments in drought zones and quiet rural areas, citing water shortages and noise pollution. FERC’s order requires data centers to pay full grid upgrade costs and, to qualify for fast-track processing, to bring their own power or curtail demand during high-stress periods. It also directs grid operators to consider 'alternative transmission technologies' like solid-state transformers or superconducting lines, and to be more accommodating to behind-the-meter power. The order builds on a December step allowing direct power plant connections. Tech companies and developers welcomed the faster connection process, but utilities, states, and clean energy advocates had worried about loss of authority, though FERC said it leaves states in control of retail rates and terms. Robert Montejo, a lawyer representing data centers, said AI 'has fundamentally changed the electricity landscape.' The Edison Electric Institute said the order builds on existing processes while supporting flexibility. Jeff Dennis of the Electricity Customer Alliance said it responds to big power users and state regulators. Rob Gramlich, an energy consultant, said states should quickly develop rules to prevent cost shifts to residential and business customers, warning FERC could assert broader jurisdiction if states don’t act. Texas, outside FERC’s jurisdiction, approved ERCOT’s 'Batch Zero' process on the same day to speed large-load studies for projects of at least 75 MW, with ERCOT tracking over 438,000 MW of large-load requests, nearly 89% from data centers. Tech giants including xAI, Google, and Microsoft have signed a pledge to build new power sources and cover infrastructure costs. The Trump administration on Wednesday said it would pay $765 million to wind developer Invenergy to cancel offshore wind leases near California, Maine, and New York, with the company saying it would use the money to build natural gas plants in the Midwest and geothermal projects in the West; one of Invenergy’s wind projects would have generated as much as 2.4 gigawatts of power. Altogether, the Trump administration has now spent about $2.6 billion to scuttle offshore wind developments. Communities increasingly resist data center developments in drought zones and quiet rural areas, citing water shortages and noise pollution.

Global AI data centre power demand is straining electricity grids, with consumption reaching 415 TWh in 2024 and projected to exceed 945 TWh by 2030. The surge, driven by GPU-optimized facilities requiring up to 200 kW per rack, creates continuous baseload stress on local transmission nodes. In response, New York imposed a moratorium on hyperscale permits, the EU mandates annual energy and water reporting for data centres over 500 kW, Ireland requires on-site backup and demand reduction obligations, and Australia plans to require large operators to act as net-generators, underwriting new power supply.

The U.S. electrical grid must add capacity equivalent to over 20 New York Cities by the early 2030s to meet surging demand, primarily from AI data centers now up to 50 times larger than previous generations. A 10-percentage point increase in agentic AI adoption could boost power needs by another 25% by 2035, with leading AI companies' requirements expected to double by 2027 per IEA estimates. This has transformed utilities from a low-growth sector into a high-growth investment theme, with $600 billion in transmission and distribution spending projected through 2030. However, the grid faces severe hurdles: regulatory delays, equipment shortages, political opposition, and a lack of skilled labor. A key example is PJM Interconnection, where capacity auction prices surged over 1,000% in two years to a record $329.17 per megawatt-day in July 2025, driven by data center demand that could reach one-fifth of regional load. This has sparked price volatility, affordability concerns, and calls for price caps from governors. Investors see opportunities in power generators, engineering firms, and equipment manufacturers, but outcomes depend on local regulations and market constraints. The demand shock may be prolonged through the decade due to underinvestment since 1999, though efficiency gains in AI could temper growth. Data center operators are increasingly building their own dedicated power plants to ease grid strain and reduce consumer cost impacts, but still face multi-year waits for permits and approvals. The power sector has experienced about $3 trillion in underinvestment since 1999, and the cycle may be long-lived as the grid grows linearly while AI computing scales exponentially. Regulators are stepping in with transitional planning and emergency procurement mechanisms, but a clear path forward remains uncertain. Price volatility, demand-supply imbalance, and popular dissent are all visible at PJM, where governors push for price caps and the operator has sped up its interconnection queue. Investors also see opportunities in fast-start generation, storage for intermittent renewables, and grid build-out services, but selectivity is key as outcomes vary widely by local constraints.

NextEra Energy and Brookfield Corporation are developing a $100 billion AI data center campus at the former Cold War-era uranium enrichment site in Paducah, Kentucky—the largest economic investment in the state's history. The project, selected by the U.S. Department of Energy, is scheduled to begin operations in 2028 and reach full development by 2032, providing over 1.2 gigawatts of data center compute capacity and up to 1.8 gigawatts of electricity that could be supplied to the grid. NextEra will develop up to 2 gigawatts of natural gas-fired power and 2.6 gigawatts of battery storage in stages, while Brookfield will own and operate the campus, leveraging existing transmission, water, and fiber infrastructure. The broader coalition includes electric cooperatives Big Rivers Electric Power Corporation, Jackson Purchase Energy Cooperative, and Paducah Power System. The initiative aligns with the Trump administration's Ratepayer Protection Pledge to prevent energy costs from being passed to residential customers. For NextEra, the project supports a base case of 15 gigawatts of data center hubs by 2035, while Brookfield uses it as a seed for its $100 billion AI infrastructure fund backed by Nvidia and the Kuwait Investment Authority. The project remains subject to final agreements.

PJM Interconnection, the largest US power grid serving 67 million people across 13 states and DC, faces its first-ever capacity shortfall as AI data centers drive unprecedented electricity demand. The grid's capacity auction failed to secure sufficient supply for June 2027–May 2028, with data centers projected to add over 30 gigawatts of peak demand by 2030—equivalent to 30 nuclear plants. Capacity prices surged tenfold from $28.92 to $329.17 per megawatt-day. In response, the Department of Energy issued a May 2026 emergency order allowing PJM to curtail power to data centers with backup generation during emergencies. PJM also approved a $6.7 billion transmission investment plan targeting Northern Virginia congestion and plans a reliability backstop auction for Autumn 2026. The capacity price increases could cost consumers over $100 billion through 2033.

Global data center power demand is set to triple over the next decade, with CO2 emissions projected to double, driven by the AI boom. Grid interconnection delays are pushing short-term reliance on off-grid gas turbines, as seen with xAI's Colossus 2, but long-term solutions include solar-plus-storage microgrids, enhanced geothermal systems (e.g., Sage Geosystems' partnership with Meta), hydrogen fuel cells (Ballard and Plug Power with Microsoft and Vertiv), and even orbital data centers (SpaceX, Blue Origin, Starcloud). At grid scale, wind, solar, and nuclear power will deliver the most low-carbon electricity, with small modular nuclear reactors (SMRs) potentially supplying up to 15% of data center power by 2037. Energy efficiency improvements, such as liquid cooling and 800VDC architecture, are being adopted to manage rising rack densities, with component-level gains from NVIDIA, AMD, and Infineon. Scope 3 supply chain emissions remain the largest source of data center CO2, forecast to exceed 0.8 gigatonnes annually by 2036, prompting strategies like carbon credits, low-carbon construction materials (green concrete, steel, timber), and lower-embodied-carbon IT hardware.

AI data centers could triple U.S. electricity consumption by 2028, up from 4.4% in 2024, according to Penn State University. Bloomberg projects they may consume 20% of U.S. power by 2035. The UN and IEA estimate global AI energy demand could double to 945 terawatts by 2030, exceeding consumption of Japan, Brazil, Germany, and France. Rack densities have doubled since 2021, with next-generation systems pushing toward 200 kilowatts per rack. Power availability now drives data center siting, with half of 2026 projects facing delays. Virginia could see data center power use rise to 57% by 2030, while Wyoming and Iowa may triple to 30%. Residential rates have risen 8-15% near major clusters, prompting states like Florida, Tennessee, and Oregon to restrict utility cost pass-throughs. Developers are funding grid upgrades or building off-meter plants, though challenges include supply chain issues and regulatory approvals.

Data centers have become the dominant force driving U.S. economic investment and political volatility in 2026, representing the largest capital project in human history. The five biggest hyperscalers—Amazon, Microsoft, Google, Meta, and Oracle—are set to spend over $750 billion on capital expenditures this year, a 67% increase from last year, with roughly 75% earmarked for AI infrastructure. This boom is straining the electric grid, with utilities now forecasting a sixfold jump in peak demand growth for 2030 compared to predictions from just three years ago. Data centers used 4.4% of U.S. electricity in 2023, forecast to reach nearly 12% by 2030, and PJM, America's largest grid operator, has proposed cutting data centers without their own power plants first during high stress. Politically, the issue has become a defining flashpoint, with over 70% of Americans opposing local data center construction, according to Gallup. In Q1 2026 alone, at least 75 projects totaling roughly $130 billion in potential investment were delayed by political opposition, prompting governors in Pennsylvania, Michigan, and Texas to impose moratoriums or strict new requirements. Meta is committing over $50 billion to build Hyperion, a 5 GW data center in Louisiana powered by 10 new natural gas plants, while Nvidia announced over $100 billion in guarantees for an 8 GW Ohio project backing OpenAI. Competition for resources is fierce: Amazon paid $700 million for 188 acres in Northern Virginia, and John Deere linked an 18% jump in construction sales to the data center buildout. Data centers have added over $21 billion in new private construction spending over the past year, even as the rest of the industry shrank.

Global data-center electricity demand reached 787.8 terawatt-hours in 2025, a nearly 20% year-over-year increase from 658.2 TWh in 2024 and roughly 92% above 2020's 410.8 TWh, according to the Energy Institute's 2026 Statistical Review of World Energy—the first to track data-center power. The United States consumed 312.6 TWh, 39.7% of the world total and nearly half of global growth (adding 63.5 TWh in 2025 alone), with overall U.S. electricity demand rising 1.7% annually since 2020 after two decades of stagnation. China followed at 205.7 TWh (26.1%), Europe at 144.6 TWh (18.4%). U.S. data-center use surged 25.5% in 2025, ending flat demand and driving an 81% increase since 2020. Virginia's commercial sales rose 30 million megawatt-hours from 2019 to 2025, with PJM summer peak demand up 23% and winter peak up 45%; transmission congestion costs in PJM jumped 43% to $6 billion in early 2026. Data centers now represent about 2% of global electricity generation, with S&P Global Energy forecasting demand could double to 1,550 TWh by 2030, consuming 6% of world electricity, while U.S. data centers could use 6.7% to 12% of national power by 2028. The AI buildout increasingly faces energy constraints, driving demand for natural gas, renewables, storage, and nuclear power—exemplified by Microsoft's deal to restart a unit at Three Mile Island and Amazon's nuclear investments, while companies like Hut 8, Core Scientific, and Iris Energy allocate power capacity to AI workloads.

Global data center power demand is projected to surge 26% in 2026, reaching 565 terawatt hours, driven by AI workloads, according to Gartner. AI-optimized servers will account for 31% of that consumption, and by 2027, their power use will surpass conventional servers. In the US, data center power demand is forecast to climb from 31 gigawatts in 2025 to 41 GW in 2026 and 66 GW in 2027, consuming 8.5% of peak summer power by 2027, per Goldman Sachs Research. However, grid constraints mean only 50-60% of planned capacity will come online on time, with regions like the Mid-Atlantic facing reliability risks. Power availability now drives data center location decisions, and Gartner warns consumption could exceed 1,200 TWh by 2030, straining grids.

Hyperscalers are racing to secure gigawatts of clean power for AI, with global capital expenditure projected at $5.2 trillion by 2030 and power demand surging 165%. In the US, Microsoft signed a 20-year deal to restart Three Mile Island, Amazon bought a data center campus at Susquehanna nuclear station for $650 million, Google agreed to small modular reactors, and Meta seeks up to 4 GW of new nuclear. In Europe, Bitzero secured a 15-year, $2.6 billion lease for AI workloads in Norway, leveraging cheap hydroelectric power. Middle Eastern sovereign wealth funds are investing globally, while China builds a self-sufficient AI power ecosystem with state-backed coal, nuclear, and renewables.

AI data centers' electricity demand in the U.S. is set to triple by 2028, according to Penn State University, after accounting for 4.4% of national consumption in 2024. Bloomberg projects they could use 20% of U.S. electricity by 2035. Globally, the IEA estimates AI and data centers could consume 945 terawatts by 2030, exceeding the usage of Japan, Brazil, Germany, and France. The surge is driven by AI racks now exceeding 50 kilowatts, with next-generation systems pushing toward 200 kilowatts per rack. This has led to concentrated impacts in states like Virginia, where data centers could use up to 57% of electricity by 2030, and has sparked concerns over rising utility rates, with some states passing laws to prevent cost shifts to residential ratepayers. Developers are increasingly building their own power plants to meet demand.

The AI boom is driving surging data center power demand, as hardware efficiency gains trigger Jevons paradox: Nvidia's Vera Rubin platform offers 10x more inference performance per watt than Blackwell, yet rack power draw has jumped from 40 kW in 2023 to 210 kW in 2024, with 600 kW racks projected by 2027. Hyperscalers are securing gigawatt-scale energy contracts, including nuclear projects like Microsoft's Three Mile Island deal and fusion power purchase agreements—Helion Energy will deliver 50 MW to a Microsoft data center by 2028, and Commonwealth Fusion Systems is building a 400 MW plant in Virginia with Google buying 200 MW. The industry's planning assumption is that efficiency gains will be consumed by increased scale, not cost savings.

The S&P 500 fell 0.53% to 7,706.03 on 23 September 2026, while the Dow declined 0.18% and the Nasdaq dropped 1.05%, as the 10-year Treasury yield surged 8.7 basis points to 5.054%—its highest since 2007—and the two-year yield hit 4.862%, a level not seen since June 2024. The sell-off was driven by S&P Global’s flash US Composite PMI jumping to 58.4, the strongest business activity since July 2021, with manufacturing at 57.0 and services at 58.7, far exceeding forecasts. Strong demand and rising price pressures reinforced expectations for further Federal Reserve tightening; Fed funds futures priced a 73% probability of an October rate hike, up from 53% earlier, and Fed Governor Michael Barr stated additional increases are likely needed. Over 72% of US stocks declined, with utilities and consumer discretionary leading losses, while energy was the only positive sector as Brent crude rebounded above $103 after five consecutive declines. Megacap tech mostly fell, with Alphabet down 4.7% and Amazon 2.2%, though Meta and Microsoft gained. Global markets also declined, with the DAX and Hang Seng falling, as investors focused on upcoming Trump-Xi talks and whether yields will stabilize.

A new TrendForce analysis projects that global data center power demand will reach 161 GW in 2026, a 31% increase from 122.9 GW in 2025, driven primarily by AI server expansion. Demand is forecast to climb to 211 GW in 2027 and 490.7 GW by 2030, while global grid capacity available for data centers will only reach 222.6 GW, creating a nominal 268 GW shortfall. AI servers are the primary driver, accounting for 33.4% of demand in 2026 (up from 25% in 2025), over 40% in 2027, and 53.7% (about 264 GW) by 2030—a nearly fivefold increase from 53.8 GW in 2026. General servers' share declines to 25.6% by 2027. The supply-demand divergence begins in 2026 and widens sharply after 2028, particularly in the US, where TrendForce's shipment-implied demand reaches 272.4 GW in 2030 against only 100.7 GW of grid-deliverable capacity. This 170+ GW gap comprises about 124 GW of demand exceeding the Lawrence Berkeley National Laboratory forecast (which projects ~148 GW of interconnection capacity by 2030) and roughly 49 GW of grid deferrals from operator filings—led by PJM (19.5 GW), ERCOT (11.2 GW), and MISO (7.6 GW). Behind-the-meter self-generation and high-voltage direct current systems could offset some of the gap, but real power shortages from aging grid infrastructure and interconnection delays pose a critical constraint on AI expansion, with delay risk concentrated after 2028. Capital expenditure from major cloud service providers continues to escalate, with annual growth rates above 30% into 2027, and the concentration of demand in US grid regions could drive up wholesale electricity prices for all customers.

AI data center energy consumption surged 80% from 2020 to 2025, with Microsoft, Google, Amazon, and Meta investing over $200 billion in infrastructure in 2024 alone. U.S. data centers now consume about 176 terawatt-hours annually, but 2030 projections range wildly from 200 to over 1,000 terawatt-hours, creating grid chaos. Utilities in Northern Virginia and Texas face power requests triple entire state usage, prompting million-dollar deposits to filter speculative applications. Local opposition has blocked more projects in a single quarter than in the prior two years combined. The scramble is raising electricity prices for all customers, as grid operators lock in capacity costs based on uncertain forecasts, with higher rates set for at least three years.

Hyperscale AI data center expansion is straining power grids, with Microsoft, Meta, Amazon, and Google's combined spending projected to exceed $700 billion in 2026, up from $64 billion in 2018, and cumulative capex surpassing $7 trillion globally. In the US and Canada, over 4,250 planned or in-progress projects represent roughly $4.75 trillion, with Texas accounting for nearly $1.25 trillion. US data center electricity demand is expected to reach 125 GW by 2030, up from 35 GW in 2024, while operational load has grown to 53 GW from 23 GW in 2023. Grid planners' 10-year demand forecast surged from 24 GW in 2022 to 166 GW by 2025. Developers are building or co-developing power generation, with natural gas seen as the fastest path to new capacity, though GE Vernova, Mitsubishi Power, and Siemens Energy are sold out of larger turbines through 2029, with some orders into 2031. Nuclear small modular reactors (61 projects) and solar paired with battery storage (701 BESS projects worth $200 billion) are also gaining traction, while $228 billion in new gas plant projects are in play. However, over 500 data center moratoria across 42 states constrain development; New Jersey and Michigan lead with 41 and 36, respectively, New York has a statewide ban on data centers over 50 MW, and Texas paused new grid interconnections amid 200 GW of requests. $130 billion in projects were rejected in Q1 2026 alone, more than the total for all of 2025, though capital continues to pour in at over $400 billion per month in new global projects since April 2026.

The global AI data center power consumption market was valued at $10.2 billion in 2025 and is projected to grow at a 27.3% CAGR to $111.9 billion by 2035, driven by surging AI workloads. The IEA reported data centers used 415 TWh of electricity in 2024 (1.5% of global use), expected to reach 945 TWh by 2030, with accelerated server power demand growing 30% annually versus 9% for conventional servers. North America led with a 36.2% market share in 2025, anchored by the US, which accounted for 45% of global data center electricity use in 2024 and is projected to drive nearly half of US electricity demand growth through 2030. Geopolitical factors, including US tariffs on power transformers (up to 50%) and shipping disruptions via the Suez Canal, are raising costs and lead times for equipment. Grid constraints are pushing operators toward on-site power, exemplified by American Electric Power's agreement for up to 1 GW of fuel cells.

US data centers consumed 192 terawatt-hours (TWh) of electricity in 2024, representing 4.7% of total US electricity use, according to Lawrence Berkeley National Laboratory’s 2025 Update report. The lab’s Reference Case projects this could rise to 649 TWh by 2030, accounting for 11.8% of forecasted US electricity consumption, within a scenario range of 9.5% to 15.3%. Globally, the International Energy Agency estimates data centers used 485 TWh in 2025, roughly doubling to 950 TWh by 2030. AI-focused data centers drove a 50% surge in consumption in 2025, while overall global data center electricity demand grew 17%. The US holds the largest share of global data center electricity use at 45%, followed by China at 25% and Europe at 15%. Data centers are projected to be responsible for 33% of total US load growth from 2024 to 2030.

The U.S. power grid is struggling to keep pace with Big Tech's AI data center buildout, as private data center construction spending surged 64.5% after inflation from August 2025 to August 2026, reaching an annual rate of roughly $85 billion. In Texas, ERCOT was tracking about 474 gigawatts of large-power-user requests as of June 2026—roughly 90% from data centers and more than five times the state's all-time peak demand—up from 63 GW in December 2024. Governor Greg Abbott ordered an audit on August 3, and ERCOT paused new large data center connections until the review is complete. Nationally, gas turbines are booked through 2032, and industry leaders warn that electricity access will be the biggest constraint for new projects over the next one to three years. Natural gas supplies 40% of data center power in 2024, and demand could rise nearly 20% by 2030, while solar generation is projected to grow 21% by 2026. Developers plan $7.6 trillion in spending through 2031, but face grid strain with 70% of transformers over 25 years old. The Trump administration has eased regulations, allocated $1.9 billion for grid upgrades, and allowed off-meter power plants. Nuclear options like reopening Three Mile Island and Duane Arnold are not expected until 2030–2035, and some coal plants are being revived, raising emissions concerns. In Texas, Chevron and Microsoft announced a 20-year off-meter natural gas deal for a 2.67 GW plant. PJM Interconnection's December 2025 power auction showed data centers accounted for $6.5 billion (40%) of the $16.4 billion total. Utilities are spending heavily—from $104 billion in 2015 to a projected $220 billion in 2026—but supply chain backlogs for generators and turbines persist. Ratepayer cost concerns have prompted calls for data centers to pay their own way, though the Data Center Coalition's E3 study found no historical evidence that data centers increased residential electricity costs.

Virginia Governor Abigail Spanberger has formally intervened in NextEra Energy’s proposed $67 billion acquisition of Dominion Energy, calling on the State Corporation Commission to reject the merger unless it includes a six-year freeze on residential base electricity rates, a tripling of monthly bill credits to $29 for four years, and other binding consumer and clean-energy protections. The merger, announced amid surging AI and data-center demand, would create a combined utility with an enterprise value of roughly $420 billion, serving about 10 million customers across Florida, Virginia, North Carolina, and South Carolina, with a 130-gigawatt data center pipeline and plans for 260 GW of installed generation capacity by 2032. NextEra expects to spend $59 billion annually from 2027 to 2032. Spanberger, the first Virginia governor to intervene at the SCC, also seeks structural reforms to ensure large data centers bear their own energy costs, expand rooftop solar by eliminating the net-metering cap, increase Dominion’s debt share from 48% to 60% to lower returns, and strengthen disconnection protections. Expert testimony filed by the governor’s office concluded that the companies have not yet met the legal standard for approval, and the SCC should reject the petition unless these conditions are substantially addressed.