Power

About 12 gigawatts of U.S. data center capacity was scheduled to come online in 2026, but only about a third of it was under active construction in the spring, according to an analysis by Sightline Climate reported by Bloomberg in April. The analysis pointed to shortages of transformers, switchgear, and batteries as a main reason many projects were expected to slip, and the reporting put roughly half of the year's planned builds at risk of delay or cancellation. That equipment accounts for less than 10 percent of a data center's total cost, yet a building cannot be energized without it.

The mismatch is one of timing. Large power transformers took 24 to 30 months to deliver before 2020; Bloomberg reported that high-power units can now take up to five years, while AI data center deployment cycles run under 18 months. Wood Mackenzie's second-quarter 2025 supply chain survey, cited by POWER Magazine, put average lead times at 128 weeks for power transformers and 144 weeks for generator step-up transformers, the large units that connect power plants to the grid.

Wood Mackenzie's August 2025 analysis found that U.S. demand for power transformers had risen 116 percent since 2019 and demand for distribution transformers 41 percent, against projected supply deficits of 30 percent and 10 percent in 2025. Imports were expected to supply 80 percent of power transformers and 50 percent of distribution transformers. “Utilities are routinely turning to the import market to meet project timelines,” said Ben Boucher, a senior analyst at Wood Mackenzie, adding that this was escalating costs and lead times. Bloomberg reported that U.S. imports of high-power transformers from China rose from fewer than 1,500 units in 2022 to more than 8,000 in 2025. Data centers are one demand driver among several: POWER Magazine, summarizing Wood Mackenzie data, also cited roughly 40 million distribution transformers past their expected service life, about 55 percent of the U.S. fleet, along with renewable energy interconnections and a 77 percent rise in power transformer prices since 2019.

Manufacturers are building more capacity, but not quickly. Wood Mackenzie counted $1.8 billion in announced expansions by major equipment makers since 2023. Hitachi Energy said in September that it will build a $528 million transformer factory in Gallman, Mississippi, more than twice the size of its nearby Crystal Springs plant, with construction expected to begin late in 2026 and production scheduled to start in 2029. Siemens Energy announced a $150 million factory in Charlotte, North Carolina, its first U.S. large power transformer plant, citing $3.9 billion pledged to expand and update the U.S. grid within two years. A plant announced in 2026 that begins producing in 2029 does little for a campus being planned for 2027.

Washington has started to respond. On April 20, 2026, President Trump issued a determination under Section 303 of the Defense Production Act describing U.S. capacity to produce transformers, high-voltage transmission components, and other grid equipment as “dangerously limited,” opening the way for federal purchasing and financial support. Utility Dive reported that about $323 million remained available in fiscal 2026 DPA funds, a figure attributed to Jean Su of the Center for Biological Diversity that the White House did not immediately confirm. Spencer Pederson of the National Electrical Manufacturers Association called the move a “step in the right direction” while noting that funding and implementation details were unclear; Su said the funding is “just not that much.”

Not everyone agrees the shortage is a manufacturing problem. In POWER Magazine's reporting, Patrick Tarver of Bolt Electrical argued that “there is not a shortage,” claiming standard substation power transformers can be delivered in 12 to 14 months once engineering approval is complete, and that the real bottleneck is how utilities and engineering contractors structure procurement, with qualification rules and approved-vendor lists that keep alternative suppliers from reaching decision makers. He also said industry messaging was exaggerated to drive pricing. That view sits against the Wood Mackenzie survey data, and the two are not easy to reconcile from the outside.

Whichever account is closer to the truth, the gap that matters to data center developers is the one between schedules. AI data center deployment cycles run under 18 months, while the grid equipment that feeds them is ordered on a clock of two years or more, in a market where imports fill much of the shortfall and new domestic plants are years from output. For now, the pace of the AI buildout is being set less by chip deliveries than by how fast a handful of factories can wind copper around steel.

U.S. electricity transmission capacity is falling far short of projected needs, with analysts at ICF forecasting residential rates could rise another 15 to 40 percent over the next five years and potentially double by 2050. The Department of Energy concluded in 2024 that the country would need to double or quadruple transmission capacity by 2050, requiring roughly 5,000 miles of new high-capacity regional lines per year; the U.S. built only 5,000 miles total over the nine years from 2017 through 2025, averaging just 600 miles per year since 2017. In August 2026, Energy Secretary Chris Wright canceled three proposed National Interest Electric Transmission Corridors that the Department of Energy had advanced in December 2024. A bipartisan Senate bill to reform transmission permitting has been introduced, but the current congressional session is running short.

Troy, Michigan-based Virtus Solis Technologies and Chicago-based Brae Systems have signed a 20-year power purchase agreement under which Virtus Solis will supply 100MW of orbital solar power, delivering 876,000 MWh annually to Brae's underwater data center concept. Neither company has yet deployed the infrastructure required to fulfill the deal; Virtus Solis plans a 100kW pilot solar plant in low-Earth orbit within approximately 24 months and a public microwave transmission demonstration in March 2027. The contract includes an option for Brae to take up to 250MW within three years of initial commercial operations.

As utility interconnection timelines lengthen and large-load tariffs grow more demanding, data center operators are increasingly pursuing dedicated behind-the-meter generation through gas, batteries, fuel cells, renewables, or co-located power plants. Even facilities that supply most of their own power typically remain connected to the public grid for transmission, emergency imports, reserve capacity, and black-start capability, creating unresolved questions about cost responsibility. Cases involving Amazon and Talen Energy, along with regulatory proceedings in Ohio and Michigan, are shaping an emerging hybrid-grid model that distributes costs between self-supplying operators and the broader grid.

PG&E CEO Patti Poppe told CNBC that the company is seeing a sharp increase in data center electricity requests, with a reported pipeline of 12.7 gigawatts, as AI and cloud infrastructure expansion drives new demand on California's grid. At the same time, Poppe warned that the failure of California lawmakers to pass meaningful wildfire liability reform under Senate Bill 492 has elevated borrowing costs, adding an incremental $600 million to customer costs over the past two years. In response to the unresolved liability framework, PG&E reduced its planned full-year 2027 capital investment by approximately $2 billion to $11.4 billion. Fitch reportedly maintained PG&E's BBB- rating but lowered its outlook to Negative from Stable, citing wildfire liability risks.

Infineon Technologies will supply silicon carbide power devices for integration into Eaton's medium-voltage solid-state transformer 2.0 platform, targeting 800VDC data center power distribution architectures. Eaton says its MVSST 2.0 reduces conversion stages compared with conventional AC-to-DC architectures, improving efficiency, power density, and deployment flexibility, with SiC's superior thermal conductivity and higher switching speeds further improving performance. David Zheng, VP of R&D for power quality in APAC at Eaton, and Andreas Weisl, EVP and chief sales officer of industrial and infrastructure at Infineon, both cited rapidly growing AI data center power demands as the driver for the collaboration. The two companies said they plan to explore next-generation platforms based on 2.3kV and 3.3kV SiC power modules to support higher system voltages in future facilities.

DG Matrix and TerraFlow Energy have announced a commercial agreement to deploy a solid-state transformer paired with a vanadium redox flow battery to power a 1,200-core GPU cluster of Dell servers, in what would be one of the first SST architectures to power a high-performance computing cluster in the United States. DG Matrix CEO Haroon Inam said the system will replicate Nvidia's 800 VDC power architecture and replace between 11 and 18 separate pieces of equipment with a single unit. Jefferies analysts wrote in an equity research note that the SST market for U.S. data centers could reach $4.3 billion in 2030, up from $37 million today, with analysts identifying a qualification window between now and 2028 when vendors must demonstrate field results. TerraFlow CEO Jon Parrella also plans to deploy the combined SST and battery system at his company's 25-megawatt Center of Excellence.

National Grid Partners' third annual Utility Innovation Survey, drawn from 134 U.S. utility innovation leaders surveyed between May 20 and July 7, found that 74 percent say AI-driven data center load growth is impacting grid reliability, and 78 percent are deploying at least one AI application to manage interconnection demand. Grid reliability has overtaken net-zero goals as the top industry priority, with 73 percent placing reliability in their top three concerns, up from 43 percent in 2025, while net-zero priority dropped from 54 percent to 16 percent. The survey also found that 83 percent of respondents said infrastructure costs for AI data centers are being passed on to residential customers. National Grid recently joined the AI Energy Management Alliance alongside Google, NVIDIA, Anthropic, and Emerald AI to develop policies for flexible, demand-responsive data centers.

Argentina's government wants to attract large-scale data centers, but a regulation published September 25, Resolution 264/2026, now requires any new demand representing at least 0.5% of the Wholesale Electricity Market average, currently around 80MW, to secure its own generation and firm capacity covering 100% of maximum demand, rising to 115% for data centers. The country's roughly 15 large-scale data centers currently consume only 35MW to 40MW combined, according to SkyOnline COO Daniel Fiorda, compared with 280MW in Chile and 900MW in Brazil, meaning a single 80MW project would double the sector's existing demand. Argentina's grid operates at approximately 94% capacity during peak season, a figure Fiorda says leaves almost no buffer, and energy contracts can take a year or more to result in actual supply because renewable parks are built against already-sold capacity. Last-mile distribution reinforcement works in critical areas can take between one and three years, adding further delay for prospective data center developers.

Data center development in the United States is facing a wave of physical, regulatory, and community obstacles even as overall construction activity remains active. Goldman Sachs said in May that only 50% to 60% of planned data center capacity will come online as expected in the next two years, and Wood Mackenzie analyst Caitlin Connelly reported in July that the 36 gigawatts added to the development pipeline in the first quarter of 2026 was down 19% from the prior quarter. Specific setbacks include a $1 million fine levied against a Microsoft-backed facility in Vineland, New Jersey, a court-ordered work stoppage on a Google project in Xcel Energy's Minnesota territory, and Oracle moving to protect itself financially after New Mexico regulators blocked a gas pipeline for its 2.5-gigawatt Project Jupiter campus in Dona Ana County. Texas Gov. Greg Abbott successfully pushed for a pause on data center grid interconnections that BloombergNEF said threatens up to 20% of the total U.S. development pipeline.

Tom Garvens, vice president of advanced technology innovation at Google, outlined the company's full-stack AI data center strategy during a keynote at Data Center World Power 2026 in Houston on September 21-23. Google now processes more than 3 quadrillion AI tokens per month, a sevenfold increase in compute demand from May 2025 to May 2026, and has moved away from centralized UPS systems to rack-level battery backup operating at 54 VDC, yielding approximately a 1.5% efficiency gain. The company is pursuing a gradual migration toward 800 VDC power distribution, expected to take a couple of years to mature, while deploying battery energy storage systems for standby power and peak shaving under demand-response agreements with utilities. Garvens noted that Google, Meta, Microsoft, and Nvidia have collaborated to define an interface specification governing how utilities and hyperscalers communicate with each other.

Oracle has issued a force majeure notice to Blue Owl Capital in an attempt to shield itself from increased costs tied to Project Jupiter, a 2.5 GW data center campus near New Mexico, according to a Bloomberg report. The 1,400-acre campus, which includes four data center buildings being developed by Stack Infrastructure and BorderPlex Digital Assets with plans to invest up to $165 billion, was announced with Oracle as tenant in January 2026 and is targeted for a 2028 launch. The primary obstacle has been a state rejection in July of a natural gas pipeline extension that would supply the campus; Oracle had sought federal fast-tracking of a review to allow the pipeline to enter service by August 15, warning that missing that window would significantly raise costs. Despite the notice, Oracle publicly stated on X that Project Jupiter remains on its planned schedule.

New Era Energy & Digital has closed a 20-year power purchase agreement with Luminant ET Services Company, a subsidiary of Vistra, to supply between 200 MW and 207 MW of electricity for Phase 1 of the Texas Critical Data Center in the Permian Basin. The power will come from a Vistra natural gas-fired plant in Odessa, Texas, with a capacity of approximately 1,180 MW, located immediately adjacent to the 493-acre TCDC site. Supply is scheduled to be available in the third quarter of 2027, with the agreement subject to automatic annual renewals after the initial 20-year term. Chairman and CEO Charlie Nelson said securing the electricity contract reduces development risk for Phase 1, with the full TCDC eventually targeting a capacity of 1.4 GW.

Average data center rack power density has risen from 12 kW in 2024 to 26 kW in 2026, more than doubling in two years, driven by AI accelerator deployments, space constraints, and improved thermal options such as direct-to-chip liquid cooling and immersion. Individual GPUs in some configurations reportedly draw up to 700 watts per device, and some facility designs are now targeting as high as 1 MW per rack, though that figure remains an outlier. Higher per-rack loads create cascading demands on power distribution architectures, cooling infrastructure, and building structures, since facilities designed for lighter racks may require structural reinforcement and new floor systems. Operators facing permitting delays or moratoriums on new builds are likely to densify existing space, accelerating these pressures further.

At the AI Infra Summit in Santa Clara on September 16, 2026, industry figures including Ed Nelson, strategy director and co-founder of the summit, and Joe Wells, founder and chief growth officer of Data Center Signals, said securing electricity and local consent has become the central challenge for data center builders. Nelson noted the summit has grown from roughly 300 attendees at its launch to 9,000, with 12,000 expected when it moves to the San Jose Convention Center for its 10th anniversary. Wells said the industry lacks an established approach for managing community opposition around water use, electricity demand, and project secrecy, though he argued confidentiality agreements and open public discussion can coexist. Broadband Breakfast reporter Akul Saxena observed that the shift from AI training to inference workloads is driving demand for more distributed, lower-latency facilities that now touch nearly every U.S. state.

National Grid Partners released its third annual Utility Innovation Survey on September 18, 2026, drawing on responses from 134 innovation leaders at U.S. utility companies collected between May 20 and July 7, 2026. The survey found that 78% of respondents are deploying or operationalizing at least one AI application to manage interconnection demand, while 74% say AI-driven data center load growth is already impacting grid reliability. Grid reliability has overtaken net zero as the top industry concern, with 73% of utility leaders ranking it in their top three priorities, up from 43% in 2025, while net-zero goals fell from 54% to 16%. The survey also found that 83% of respondents say the cost of building infrastructure to serve AI data centers is being passed on to residential customers through higher electricity bills.

NVIDIA, Google, and Emerald AI have launched the AI Energy Management Alliance (AEMA), a coalition aimed at getting data centers connected to the power grid faster by making them more flexible resources. New US data centers can currently wait a decade or more for a grid connection; AEMA argues that if AI facilities could reduce load during peak demand hours, the existing grid could unlock 100GW of capacity for flexible data centers, according to Emerald AI CEO Varun Sivaram. The coalition is proposing standardized technical requirements, performance metrics, and operational data sharing, with proposed tools including on-site batteries, on-site generation, and software to slow or shift less urgent computing workloads. AEMA also plans to lobby state capitals and Washington, asking governors and regulators to offer faster, larger grid connections to data centers that commit to flexibility.

Amazon Web Services has determined it cannot restore access to resources and data hosted in its Bahrain region after Iranian attacks caused damage that exceeded what its multi-availability-zone infrastructure was designed to withstand. The UAE region also suffered partial loss, with availability zone mec1-az2 declared unrestorable, while work continues on mec1-az1 and mec1-az3. AWS opened the Bahrain region in 2019 and the UAE region in 2022; Iran's Islamic Revolutionary Guard Corps claimed a second attack on the Bahrain region in July. Amazon said it will share further updates on Bahrain restoration plans in early 2027.

A report commissioned by the Pennsylvania Public Utility Commission and conducted by consulting firm Synapse Energy Economics warns that rolling blackouts could occur more than 13 days per year in a worst-case scenario where data center growth is unrestrained and electricity supplies stay tight. Even with new power plant construction over the next 15 years, the PJM grid would still fall short of demand six times per decade on average, and the system's reliability would be between six and 100 times worse than PJM's planning standard of one loss-of-load event per decade by 2030. The only scenario in which PJM meets its reliability goal is one in which no new data centers are built after 2026. PUC Chairman Steve DeFrank called Pennsylvania's situation an "energy crossroads" and urged policymakers, utilities, and other stakeholders to act.

A Broadband Breakfast report examined the intersection of a California investor-owned utility, grid operator CAISO, and Google over how to plan for and price escalating AI-driven electricity demand. The discussion centers on whether existing rate structures and interconnection processes are adequate for the speed and scale at which hyperscaler load is arriving. California's grid is already under seasonal stress, and large load additions from AI facilities are compressing planning timelines.

A new analysis from Qz.com finds that U.S. electric grid operators are making major capacity planning decisions based on highly uncertain demand projections from AI data centers. Developers routinely submit interconnection requests that overstate actual power needs, and utilities lack standardized methods to verify claims. The gap between projected and actual consumption has led to significant misallocation of grid planning resources.

Pennsylvania utility regulators have warned that rapid data center growth could cause blackouts in the state within the coming years, citing mounting strain on the PJM interconnection grid. The Pennsylvania Public Utility Commission identified the pace of new large-load interconnection requests as outpacing the grid's ability to add generation and transmission capacity. Officials have not yet specified which utilities or data center operators are most responsible for the demand surge.

A new analysis details how AI data centers are imposing load characteristics on the US electric grid that were not anticipated when current transmission and generation infrastructure was designed. The grid was built around gradual demand growth and predictable industrial load curves; AI facilities draw large, concentrated, and often continuous power that strains both local distribution and regional transmission. Utilities and grid operators are being forced to revisit interconnection timelines and reserve margin calculations.

Pennsylvania's Public Utility Commission has raised alarms that the rapid growth of data centers in the state is putting grid reliability at serious risk, warning that PJM's capacity to serve existing customers could be compromised. The PUC has flagged the pace of new interconnection requests from data center developers as outstripping infrastructure upgrades. Officials have not yet specified what regulatory or legislative actions may follow.

Pennsylvania's Public Utility Commission released a study warning that rapid data center growth will cause the PJM mid-Atlantic electric grid to fail its reliability standard by 2027 if no corrective actions are taken. The report projects more than a dozen loss-of-load events annually in a worst-case scenario, meaning portions of the grid could go dark on a regular basis. PJM serves roughly 65 million people across 13 states and the District of Columbia, making the stakes for grid stability exceptionally high.

The Electric Reliability Council of Texas is stress-testing its grid against the accelerating megawatt demands of AI data centers now coming online across the state. Data Center Frontier reports that ERCOT is evaluating whether interconnection queues and real-time dispatch protocols can handle large, variable AI loads without triggering reliability events. Texas has become one of the fastest-growing AI infrastructure markets in the country, drawing scrutiny over whether its deregulated grid can absorb the pace of new connections.

The U.S. Department of Energy has directed that the Eddystone power plant in Pennsylvania remain operational, citing surging electricity demand driven in part by data center growth in the region. Eddystone, a coal and oil-fired facility that had been slated for retirement, will continue generating power under the emergency order to prevent grid reliability shortfalls. The decision reflects growing federal concern that planned generation retirements are outpacing the construction of replacement capacity needed to serve new large loads.

Data Center Frontier launched an industry poll asking how electric utilities should distinguish genuine data center load requests from speculative or placeholder reservations, sometimes called ghost load, that tie up grid capacity without resulting in actual construction. The poll surfaces a live operational problem for utilities that have seen interconnection queues swell with projects that never break ground. Responses are being collected from operators, developers, and utility planners.

SemiAnalysis published a deep-dive analysis examining how AI laboratories are addressing the grid power shortage by deploying onsite natural gas generation at scale. The piece examines specific configurations, fuel sourcing, permitting approaches, and the economics of bypassing utility interconnection queues. As interconnection timelines stretch to five or more years in many markets, onsite generation has shifted from a backup option to a primary power strategy for large AI operators.

The Pennsylvania Public Utility Commission is scheduled to address data center electricity demand at an upcoming proceeding, according to Lehigh Valley Business. Pennsylvania has seen significant data center growth, particularly in the Lehigh Valley corridor, placing new load pressures on regional transmission networks. A formal PUC proceeding could produce cost-allocation rules, interconnection standards, or demand-forecasting requirements that affect developers operating in the state.

Questionable load forecasts submitted by data center developers are driving up electricity bills for Ohio utility customers, according to a Canary Media report. Utilities in the state have been building or planning generation and transmission infrastructure based on demand projections that have not materialized at the scale promised. The gap between forecast and actual consumption shifts costs onto existing ratepayers who fund grid upgrades that primarily serve large commercial customers.

Solar energy projects in Georgia are encountering organized opposition as the state's data center buildout drives surging electricity demand, according to Bisnow. Landowners, local officials, and agricultural advocates are pushing back against large solar installations that utilities are pursuing to meet the new load. The tension highlights a conflict between clean energy siting and rural land use priorities that data center growth is accelerating.

Oracle has issued a request for proposals seeking 2 gigawatts of new renewable energy capacity in New Mexico, signaling a major expansion of its power procurement strategy. The RFP is among the largest single clean energy solicitations by a technology company in the state's history. New Mexico's abundant solar and wind resources make it an attractive target for large-scale procurement, and the outcome will shape which developers and utilities win long-term contracts to supply Oracle's growing data center operations.

Google announced a record $15 billion investment in AI infrastructure in Finland, which the company described in a blog post as part of a blueprint for responsible clean energy growth. The investment marks the largest single technology infrastructure commitment in Finnish history and positions Finland as a major European hub for AI compute capacity. Google's post outlined clean energy sourcing commitments tied to the buildout, including expanded renewable procurement deals in the region.

A megawatt-class solid-state transformer has been successfully demonstrated on a live utility feeder, marking a significant step toward commercial deployment of the technology. Solid-state transformers can manage power flow more precisely than conventional units and are seen as candidates for handling the variable, high-density loads common in AI data centers. The demonstration moves the technology from laboratory settings to real grid conditions.

Major technology companies are taking on direct financial risk in power generation projects, moving beyond standard power purchase agreements to become active participants in project financing and development. The strategy is driven by forecasts of sharp electricity price increases tied to surging data center demand, with companies betting that owning risk upstream is cheaper than paying elevated market rates later. Reuters reported the shift involves companies committing capital to generation assets rather than simply contracting for output.

Enphase Energy has advanced its IQ Solid-State Transformer program, with power modules now being manufactured at a Texas facility targeting AI data center applications. The solid-state transformer technology is designed to handle the high-density power loads that AI workloads impose, replacing conventional magnetic transformers with semiconductor-based alternatives. Enphase has not disclosed which data center operators or hyperscalers are evaluating the technology.

Texas has halted or frozen data center electricity demand commitments, according to 247wallst.com, creating ripple effects for utility-sector investors who had priced in aggressive AI-driven load growth. The XLU utility ETF is feeling the pressure as the AI power demand story faces scrutiny in the state. Texas had been among the fastest-growing markets for data center electricity load.

Simply Wall St identified three utility stocks that have signed long-term power purchase agreements or capacity deals tied specifically to AI data center demand, framing them as beneficiaries of the sustained infrastructure buildout. The piece does not name specific dollar figures in the snippet but points to multi-year contracted revenue as the key investment thesis. Utilities securing long-term AI-linked contracts are increasingly seen as a distinct sub-sector within the broader power market.

The Tennessee Valley Authority says electricity demand across its service territory has set a 2026 record during the current heat wave, with the utility reporting it remains prepared to meet load. TVA serves approximately 10 million people across seven states in the southeastern United States. The authority has not disclosed the specific peak demand figure but confirmed grid stability has been maintained.