← All stories

Digest Original

Policy Digest Original

State Audits Find Data Center Tax Breaks Return Cents on the Dollar in State Revenue

A Data Center Digest Original StoryThis article was researched and written entirely by AI, without human review or editing, as part of the Data Center Digest's ongoing experiment in AI-powered journalism.

When states have audited their data center tax breaks, the auditors have reached the same result on revenue: the breaks do not pay for themselves in state taxes. Virginia's legislative auditors estimated in November 2026 that the state's sales tax exemption for data centers brought back 22 cents in state revenue for every dollar it cost from fiscal 2021 through 2025. A December 2025 evaluation for Georgia found the state's exemption produced a net revenue loss in every year it studied.

Whether the promised jobs showed up is harder to say. Most states count the jobs companies commit to, or the jobs an economic model estimates, rather than jobs anyone has checked. Where official figures exist, the cost per job ranges from about $34,000 to roughly $900,000, depending on what is counted. Most of the work is construction, which ends when the buildings are finished.

This article draws on reports from Virginia's Joint Legislative Audit and Review Commission (JLARC), Georgia's Department of Audits and Accounts, Washington's legislative auditor, and the Illinois Department of Commerce and Economic Opportunity (DCEO). It also uses testimony from the Texas Comptroller's office, given as Texas lawmakers prepare to take up a bill that would repeal that state's exemption.

How the breaks work and who checks them

The largest state data center incentives are sales and use tax exemptions. A use tax is the companion tax owed on taxable goods bought without paying sales tax, for example from out of state. The exemptions cover servers, networking gear, cooling systems and generators. Texas also exempts the electricity the facilities use.

In return, companies must invest a minimum amount and create jobs. In 2019, JLARC described Virginia's terms as a $150 million investment and 50 new jobs paying at least 1.5 times the local average wage, or 25 jobs in high-unemployment areas. Texas requires 20 jobs at 120 percent of the county median wage and $200 million invested over five years, or 40 jobs and $500 million for the largest projects. Illinois requires $250 million, 20 jobs at 120 percent of the county average wage, and a project labor agreement, which is a pre-hire contract with construction unions. Georgia requires 25 jobs and $250 million in counties with more than 50,000 people, and as few as five jobs and $25 million in counties with fewer than 30,000.

Local property tax breaks are a separate layer, negotiated county by county. John Diamond of Rice University's Baker Institute told CBS Austin that under one Milam County, Texas, agreement, the developer would pay 25 percent of its would-be property taxes over a 10-year abatement period. None of the state evaluations reviewed here measures these local breaks.

Each evaluation depends on a "but for" assumption: an estimate of how much of the investment happened only because of the tax break. A Good Jobs First report, as summarized by DatacenterDynamics, found that 12 of the 32 states with data center subsidy programs do not disclose how much revenue they give up. In Texas, the Comptroller's office turned down a Texas Tribune records request for facility-level data, citing state confidentiality law.

Virginia's three verdicts

Virginia has the longest record, and its verdict has grown less favorable with each report. JLARC's 2019 study covered fiscal 2010 through 2017, when the exemption cost an average of $52.2 million a year.

The November 2026 report found 22 cents and 29 jobs per $1 million for fiscal 2021 through 2025, when the exemption averaged $811.5 million a year and reached an estimated $1.4 billion in fiscal 2025. That made it 60 percent of everything Virginia spent on economic development incentives from fiscal 2016 to 2025. That figure includes supplier and spending-driven jobs and subtracts jobs lost to the taxes needed to cover the exemption's cost.

Ellen Miller, JLARC's chief economic development analyst, told lawmakers that power availability and community acceptance may now weigh more than the tax break in companies' location decisions.

This article uses the 2026 figures.

The state figures leave out local revenue, which is where the fiscal case is strongest. Loudoun has separately reported $1.2 billion in data center real and personal property tax revenue, 39 percent of its budget, according to the Virginia Mercury. The 2026 budget kept the exemption and added an energy consumption tax on data centers.

Georgia, Washington, Illinois and Texas

Georgia's December 2025 evaluation was prepared by the University of Georgia's Carl Vinson Institute of Government under the Tax Expenditures Transparency Act of 2024 (Senate Bill 366). It cut the share of data center construction attributed to the exemption from the 90 percent assumed in a 2022 study to 30 percent. The new figure came from a statistical model built on data for 34 Georgia data centers and on national facility data. The evaluation estimated $450.6 million in forgone state revenue in 2024 and a net loss of $416.6 million after counting the new taxes data centers generate. It projects a net loss of $780.2 million in 2030. The same report also calculated a return of $2.86 in added economic output (value added, roughly the contribution to state GDP) for each dollar of forgone revenue. The researchers cautioned that repealing the incentive could signal an unfriendly business climate. In 2024, House Bill 1192, which would have paused new exemption certificates, was vetoed.

Washington's legislative auditor reviewed a 2022 exemption for data centers in King, Pierce and Snohomish counties in a preliminary report in July 2026. Fewer than 10 businesses used it, all for refurbishing existing buildings, and no new urban data centers were built with it. Beneficiaries saved an estimated $42.4 million from 2023 to 2026. Property tax payments rose after the refurbishments. The auditor recommended letting the exemption expire, and the 2026 Legislature narrowed it to new data centers only.

Illinois publishes an annual ledger rather than an evaluation. DCEO's 2025 report lists 34 agreements with companies including Microsoft, Oracle, Equinix and Digital Realty, mostly in Elk Grove Village, Chicago and other suburbs. Those agreements commit $10.67 billion in investment and 731 new jobs, with estimated tax benefits of $666.6 million, calculated at 6.25 percent of committed investment. The report gives commitments rather than verified jobs. Governor JB Pritzker paused new agreements starting July 1, 2026, saying the state needed to learn whether incentives were driving development without regard to consumer costs.

Texas has not published a formal evaluation of its exemption. At a July 2026 Senate Finance Committee hearing, Comptroller officials said companies certify their own eligibility, and that six of the 20 projects that had entered the audit process were found noncompliant or withdrew. Asked how many jobs data centers create, chief revenue estimator Brad Reynolds said he did not know but that some projects struggle to reach 20. Lawmakers expected the exemption to cost about $14.6 million in the 2014–15 budget cycle. Current estimates for the coming two-year budget are $3.2 billion according to the Texas Tribune's reading of Comptroller figures, and about $3.3 billion according to Senate Finance chair Joan Huffman. Dan Diorio of the Data Center Coalition, an industry group, testified that the exemption encourages continuous reinvestment. Shannon Halbrook of the advocacy group Every Texan argued that the industry no longer needs the subsidy.

Construction jobs, permanent jobs and cost per job

Every evaluation that separates the two kinds of work finds that construction dominates. Industry representatives told JLARC that a typical 250,000-square-foot data center employs about 50 full-time workers, half of them contractors, while about 1,500 workers are on site at the peak of a 12- to 18-month build.

Cost per job depends on what is counted. JLARC's figure of 29 jobs per $1 million works out to about $34,000 per job each year, but that count includes construction, supplier and spending-driven jobs. Illinois's $666.6 million in estimated benefits divided by 731 committed jobs gives about $912,000. That calculation is this article's own, based on officials' numbers. The jobs in it are still commitments, and the benefits extend over many years. Georgia's report does not calculate a cost per job.

So the question of whether the jobs and revenue arrived has a partial answer. No state's evaluation found the exemption recovered its cost in state taxes. Large counties such as Loudoun collect substantial property taxes from data centers. Permanent, verified job counts mostly do not exist. Residents weighing a local deal can ask the questions the audits raise: what share of the project is assumed to depend on the break, who verifies the jobs and when, whether local property tax abatements are counted, and what the company must repay if it misses its targets.

What to watch

Texas: Huffman filed Senate Bill 293 on November 9, 2026, with a majority of senators as joint authors. It would repeal the exemptions in Tax Code Sections 151.359 and 151.3595 and deny pending applications. It is a proposal, not law, and the 90th Legislature convenes in January 2027.

Virginia: some legislators have said they will revisit the exemption when the General Assembly meets in January 2027.

Illinois: the pause on new agreements has been in effect since July 1, 2026. The governor asked lawmakers to act during the fall veto session, and these sources do not show the outcome. DCEO lists July 1, 2029, as the statutory cutoff for issuing certificates.

Washington and Georgia: Washington stops issuing new urban exemptions after July 1, 2028, and existing exemptions fully expire July 1, 2038. Georgia's exemption is set to end in 2033 after House Bill 1291 extended it in 2022.

Sources
  1. Joint Legislative Audit and Review Commission (Virginia) — “Economic Development Incentives 2026”
  2. Joint Legislative Audit and Review Commission (Virginia) — “Data Centers in Virginia (2024)”
  3. Joint Legislative Audit and Review Commission (Virginia) — “Data Center and Manufacturing Incentives (2019)”
  4. Henrico Citizen / Virginia Mercury — “JLARC: Data center tax break accounts for 60% of Virginia's incentive spending”
  5. Georgia Department of Audits and Accounts / Carl Vinson Institute of Government — “Tax Incentive Evaluation: Georgia Data Center Sales & Use Tax Exemption (December 2025)”
  6. Washington Joint Legislative Audit and Review Committee — “2026 tax preference review: Data Centers in Urban Counties (preliminary report)”
  7. Illinois Department of Commerce and Economic Opportunity — “Data Center Investment Program 2025 Annual Report”
  8. Office of the Governor of Illinois — “Gov. Pritzker Pauses New Data Center Tax Incentives”
  9. Texas Senate, Office of Sen. Joan Huffman — “Senator Joan Huffman Files Legislation to Repeal Data Center Sales Tax Exemptions”
  10. Texas Public Radio — “Texas lawmakers question billions in tax breaks for data centers”
  11. The Texas Tribune — “Texas losing a billion dollars a year on data center tax break”
  12. CBS Austin — “Could Democrats' election wins reshape Texas data center tax policy?”
  13. Rice University's Baker Institute for Public Policy — “How the Data Center Boom Affects the Texas Tax Base (working paper)”
  14. DatacenterDynamics — “American data center tax exemptions "out of control," according to report”
A Data Center Digest Original Story · 2 days ago
Power Digest Original

SpaceX's Starbase Data Center Plan Puts a Texas Air Permit, Not the Grid Queue, at the Center of Review

A Data Center Digest Original StoryThis article was researched and written entirely by AI, without human review or editing, as part of the Data Center Digest's ongoing experiment in AI-powered journalism.

When a data center builds its own gas plant in Texas, the main state review of that plant is an air authorization from the Texas Commission on Environmental Quality (TCEQ), not a grid study. The Electric Reliability Council of Texas (ERCOT, which runs most of the state's grid) and the large-load rules in Senate Bill 6 (2025) mostly reach projects that plug into the grid, so a self-powered campus shifts oversight toward an air registration and local tax-break votes.

SpaceX's proposal near Brownsville is the clearest current example. A City of Starbase public notice says Space Exploration Technologies Corp. "proposes to construct, own, and operate a gas-fired power plant and super-compute facility" on about 444 acres between Boca Chica Boulevard and the Port of Brownsville. It puts the initial phases at about $24 billion. Texas Data Center Watch, which tracks the project, summarizes a Nov. 12 SpaceX announcement as saying the campus will generate its own power off the public grid.

SpaceX is not the only developer taking this approach. The Texas Tribune, citing an Earthjustice report, wrote that roughly 80% of the power secured by data center projects coming online in the next five years would come from gas-fired generators. In August, Gov. Greg Abbott told grid regulators to stop advancing data center connection requests until an audit was done. The Tribune noted that projects supplying their own electricity could still move forward. This article examines that gap.

What SpaceX has asked the state to approve

On Sept. 8, 2026, Trinity Consultants filed a registration with TCEQ on SpaceX's behalf for a "Boca Chica Facility," according to The 956 Report, a local outlet that obtained and published the document. The filing seeks coverage under TCEQ's standard permit for electric generating units (a pre-written authorization a project registers under, instead of getting a custom permit). It lists the site's industry code as computer processing and data preparation. Politico, citing the same application, described it as 35 mobile gas-fired turbines producing roughly 750 megawatts.

The equipment list shows where the turbines come from. The 956 Report lists an initial configuration of six GE Vernova TM2500 Gen 6 units (25.4 MW each), two TM2500 Gen 8 units (27.4 MW), nine Mitsubishi FT8 units (27.1 MW) and 18 Solar SMT-130 units (16.48 MW), about 748 MW combined. A final configuration would replace them with eight GE Vernova 7E turbines of 88.56 MW each, about 708 MW. The filing calls the first set temporary and lists start dates as to be determined. It says the units are used or refurbished, were not bought directly from the manufacturers and generally lack manufacturer performance guarantees.

The records reviewed for this article do not say who will supply the gas. Politico, citing East Daley Analytics, reported that SpaceX is building an 8-mile gas transmission line to Starbase that is scheduled to open in January. Using the filing's heat-input figures, The 956 Report estimates that at full load the 35 turbines would burn about 197 million cubic feet of gas a day and emit roughly 4.3 million tons of carbon dioxide a year.

The official figures for the project do not match. The Starbase notice puts initial phases at about $24 billion. A Cameron County agenda item, as reported by The 956 Report, lists a $17.9 billion capital cost, $5.2 billion in estimated taxable value and 500 permanent jobs. That county draft notice describes the plant as providing "behind-the-meter power generation to limit the burden of new load on the ERCOT grid" (behind-the-meter means generation on the customer's side of the utility meter). That wording leaves open whether the campus will eventually connect to the grid.

The air permit is the main state gate

TCEQ's standard permit for electric generating units, effective May 16, 2007, covers units that make electricity for the owner's own use, for sale to the grid, or both. The air rules therefore do not loosen because the power stays on site. The permit requires registration and bars construction or operation without written approval from TCEQ's executive director. Gas turbines must also meet federal performance standards for stationary turbines. A newer TCEQ standard permit for natural gas generating units, effective Jan. 30, 2025, covers engines only, not turbines.

The bigger federal review depends on tonnage. New major sources must go through Prevention of Significant Deterioration review, a federal preconstruction review that includes air-quality modeling and the strictest controls. The 956 Report says the threshold for a simple-cycle plant like this one (turbines without a steam unit) is 250 tons a year of any regulated pollutant. SpaceX's proposed cap for nitrogen oxides (NOx, a pollutant that helps form ozone) is 248.86 tons, 1.14 tons under that threshold, and particulate matter is capped at 242.11 tons. The filing says that if NOx runs higher than represented, the plant's maximum capacity factor will be cut proportionately. The 956 Report calculates that at full output under the permit's own NOx rate, emissions could reach about 458 tons.

The filing's own tables mark several pollutants as above the 100-ton threshold for Title V, the federal operating-permit program, so that permit will also be needed, The 956 Report found. The Title V process typically includes public notice and comment. The public notice section of the standard permit registration itself was left blank.

State politics has added a second layer. On Sept. 21, Abbott directed TCEQ to "halt all permits sought by data centers" until ERCOT finishes its audit. The law firm Bracewell reads the order as covering fully off-grid projects and expects the audit to finish in mid-December. On Oct. 19, TCEQ's interim executive director told the governor that the agency had paused permits directly supporting data centers. Public Citizen said some permits had kept moving, and the governor's office said none had been approved. The public record reviewed here does not show a TCEQ decision on SpaceX's registration.

What grid rules reach, and what they miss

SB 6, enacted in 2025, sets standards for loads of 75 MW or more that seek to connect to the ERCOT grid. Among them, such customers must disclose on-site backup generation, and ERCOT may direct them to run that generation or cut load during an energy emergency. For transmission-level loads connected after Dec. 31, 2025, utilities must have a way to cut them off during firm load shed (rotating outages).

The law's co-location section governs pairing a new large load with a power plant. It applies only to an existing plant registered with ERCOT as a stand-alone generator as of Sept. 1, 2025. Those deals require notice, an ERCOT study within 120 days and a Public Utility Commission decision within 60 days after that. The conditions must keep the plant's capacity available to the grid in emergencies. A new, privately owned plant built to serve only its own campus does not fit that description.

The grid queue itself is stalled. On June 18, 2026, the Public Utility Commission approved ERCOT's Batch Zero process, which studies large-load requests as a group. After Abbott's Aug. 3 directive, ERCOT postponed it, according to the law firm Akin Gump, which said the audit will examine how far data centers supply their own generation. Politico reported that Batch Zero projects are slated to receive power allocations in the spring and that SpaceX would likely need to make its own power for years before connecting.

Outside Texas, federal regulators are handling the same question differently. On Dec. 18, 2025, the Federal Energy Regulatory Commission directed the PJM grid operator to create new transmission services. As Commissioner Rosner explained in a concurrence, these let a load co-located with a generator pay for grid service in line with its actual net use. That framework assumes the plant and load still touch the grid.

Local control moved with the city line

Brownsville gave up its say over the land before the project became public. In late August, the city removed the 444.45-acre tract from its limits at SpaceX's request in exchange for $220 million for water and wastewater projects, and the City of Starbase later annexed it, the Tribune reported. Brownsville's 90-day data center moratorium took effect in early November and covers the city and its extraterritorial jurisdiction (a buffer zone outside city limits). Had the land stayed in the city, the moratorium would likely have pushed the project elsewhere, according to the Tribune. The city said it stands by the decision, which it based on the information available at the time.

The remaining local decisions are about taxes. Starbase's three-member commission, two of whom are SpaceX executives according to Politico, will weigh a property tax abatement and a related economic development agreement. Cameron County commissioners were asked to authorize notice of a separate abatement hearing. Neither body issues the air authorization; as The 956 Report noted, that decision belongs to TCEQ.

The Tribune reported that the South Texas Environmental Justice Network said the facilities would pollute the air, use water and raise energy bills, while some residents welcomed the jobs. Memphis offers a contrast in who decides. There, the Shelby County Health Department issued xAI a permit for 15 turbines totaling 247.2 MW through Jan. 1, 2027, after campaigners claimed the company had run turbines without proper permits, according to Data Center Dynamics.

Oversight lost and gained

Based on these records, TCEQ reviews a self-powered campus in Texas for air emissions under the same rules that apply to plants serving the grid. Local governments are involved only through annexation, tax incentives and water deals. If the campus never connects, ERCOT's system-impact studies drop away, along with SB 6's emergency obligations for co-located and backup generation and the visibility those rules give grid planners. In Brownsville's case, the city lost its zoning and moratorium powers through the de-annexation, not because the plant is self-built.

What is gained depends on whom you ask. SpaceX's county filing frames on-site generation as a way to keep new load off the ERCOT grid, and Abbott has said data centers must pay their own way. Critics point to the limited public notice: Public Citizen training materials cited by The 956 Report list standard permits for gas turbines among authorizations with no public notice. The records that would settle the remaining questions are TCEQ's technical review of the registration, the terms of the abatement agreements and any future ERCOT interconnection request from SpaceX.

What to watch

Dec. 9, 2026: the Starbase City Commission holds its public hearing on SpaceX's Chapter 312 tax abatement and Chapter 380 agreement. Mid-December 2026: ERCOT's data center audit is expected to finish, according to Bracewell; the governor's directive ties TCEQ's permit pause to that audit.

Dec. 29, 2026: Cameron County's draft notice sets a public hearing on SpaceX's county abatement, according to The 956 Report. January 2027: Governor-elect Gina Hinojosa, who has said she would ban data center approvals until lawmakers act, takes office, and the gas line to Starbase is scheduled to open, according to Politico. Spring 2027: Batch Zero power allocations are scheduled to be issued.

States in this story

Sources
  1. City of Starbase, Texas — “Notice of Public Hearing (tax abatement, Dec. 9, 2026)”
  2. The Texas Tribune — “SpaceX to build $24 billion data center in South Texas, sidestepping moratorium”
  3. The Texas Tribune — “Greg Abbott ordered a pause on data center permits. Critics say some plans still moved forward.”
  4. Office of the Texas Governor — “Governor Abbott Directs TCEQ To Halt Data Center Permits”
  5. Bracewell LLP — “Governor Abbott Orders TCEQ to Pause All Permit Issuance for Data Center Projects”
  6. Akin Gump — “Texas Pauses Data Center Interconnections Pending Statewide Audit”
  7. Texas Legislature — “89(R) SB 6 - Enrolled version - Bill Text”
  8. Texas Commission on Environmental Quality — “Air Quality Standard Permit for Electric Generating Units”
  9. Texas Commission on Environmental Quality — “Air Quality Standard Permit for Natural Gas Electric Generating Units”
  10. The 956 Report — “35 Gas Turbines, One Data-Processing Code: Inside SpaceX's Filing for a Power Plant at the Port of Brownsville”
  11. The 956 Report — “UPDATE: SpaceX Is Now Asking Cameron County for a Tax Break Too. The County's Paperwork Says $17.9 Billion and 500 Jobs.”
  12. Politico (via Yahoo Finance) — “Musk seeking tax breaks for $24B data center in SpaceX's hometown”
  13. Texas Data Center Watch — “SpaceX Starbase gas-fired power plant and super-compute facility”
  14. Federal Energy Regulatory Commission — “E-1: Commissioner Rosner's Concurrence on PJM Co-Location”
  15. Data Center Dynamics — “xAI granted permits for 15 gas turbines at Memphis data center”
A Data Center Digest Original Story · 4 days ago
Policy Digest Original

China Planned Its Data Centers Nationally. Germany Just Did. The U.S. Has Not. Should It?

A Data Center Digest Original StoryThis article was researched and written entirely by AI, without human review or editing, as part of the Data Center Digest's ongoing experiment in AI-powered journalism.

On September 8, 2026, Representative Suhas Subramanyam, a Virginia Democrat, introduced what he calls a National Data Center Plan: four bills covering where data centers are built, how their energy and water use is measured, who pays for the grid upgrades they need, and how to protect nearby communities from foreign threats. It is the nearest thing in Congress to a national strategy, and its siting piece is guidance, not a requirement. The Digest found no executive branch document that calls itself a national data center strategy. What the federal government has instead is a set of separate moves: orders that speed approvals, deals on federal land, a grid regulator still deciding how large loads connect, and 61 bills that the Digest tracks, of which one has passed a chamber.

What Washington has done so far

The main executive action is Executive Order 14318, signed July 23, 2025. It covers projects that need more than 100 megawatts of new load for AI work. It tells agencies to find existing exclusions from environmental review under the National Environmental Policy Act, allows projects to be placed on the federal permitting dashboard under the FAST-41 process, and has the Interior, Energy and Defense departments offer federal land, including leases at military installations. It also gives EPA 180 days to issue guidance on reusing brownfield and Superfund sites. The order’s stated priority is “the rapid and efficient buildout” of the infrastructure. It speeds federal approvals and points to federal land and former industrial sites, but it sets no national rules for where projects must or must not go.

On siting authority, the administration has so far left the decision with the states. Executive Order 14365, signed December 11, 2025, directs the White House to draft legislation overriding many state AI laws, but says the proposal “shall not propose preempting” state laws on “AI compute and data center infrastructure, other than generally applicable permitting reforms.” The framework the White House sent Congress in March 2026, as Governing reported it, asks for faster permitting and for protection of households from higher power costs, and leaves state zoning power in place.

Federal land is the administration’s most direct lever. The Energy Department picked four of its sites in July 2025 for private AI data center and power projects: Idaho National Laboratory, Oak Ridge Reservation, the Paducah site in Kentucky and the Savannah River Site in South Carolina. In late July 2026 two of them moved. DOE selected Brookfield to develop a 1.8-gigawatt campus at Paducah, with NextEra planning a 2-gigawatt gas plant and more than $100 billion in private investment expected, and the National Nuclear Security Administration chose Amentum to negotiate a lease for a 1-gigawatt campus at Savannah River, which is not final. Critics have pushed back. Senator Ron Wyden’s S. 5658 would ban selling or transferring federal public land for new data centers. His office says Senate Republicans blocked a request for fast passage on September 30, and it cites reporting that the administration is weighing more than 17,000 acres in six Western states for 12 data centers, which the release describes as under consideration.

The grid is where Washington has come closest to national rules, and it is contested. On October 23, 2025, Energy Secretary Chris Wright directed FERC to assert jurisdiction over connecting large loads, those above 20 megawatts, to the high-voltage grid and to set standard procedures. State utility commissions have historically handled that job, and the proposal said it would leave states’ authority over retail sales and generation siting alone. The Secretary asked for final action by April 30, 2026. FERC did not issue a single national rule. On June 18, 2026, it ordered six regional grid operators to file a report within 30 days and, within 60 days, to justify their current tariffs or file changes on five topics, including cost shifting, co-located power plants and flexible large loads. The rulemaking remains open.

What Congress has

As of the Digest’s November 9 data pull, 61 federal bills mention data centers. By a keyword count of titles and descriptions, about a dozen deal with ratepayer protection or who pays for grid costs, and more than 20 deal with disclosure or reporting. Only two, by the Digest’s reading, direct a federal agency to study or set guidance on where data centers should go: Subramanyam’s H.R. 10321 and H.R. 10041, a Commerce Department study of sites near military installations, rail hubs, airports and industrial zones.

Costs have moved furthest. H.R. 9340, the Ratepayer Protection Act, passed the House 417 to 3 on September 16, 2026. It would require state regulators to consider standards, for loads over 100 megawatts, that make large users cover the full incremental cost of new generation and grid upgrades. It does not require states to adopt them. In the Senate, Republican Jon Husted asked to pass it by unanimous consent and Democrat Martin Heinrich blocked the request. Heinrich, the top Democrat on the Energy and Natural Resources Committee, has said the bill does not go far enough and wants rules on community engagement, water use and air pollution as well.

A bipartisan Senate bill introduced September 30 by Heinrich, Mike Lee, Shelley Moore Capito and Sheldon Whitehouse takes a different route. It would widen FERC’s power to approve certain interstate transmission lines and would require data centers of 20 megawatts or more to pay the full transmission and incremental costs of serving them, without passing those costs to other customers. It still needs committee action.

Subramanyam’s package covers more of what a strategy would include. The siting bill, H.R. 10321, would have the Energy Secretary, consulting FERC and the Defense Secretary, publish best practices within a year of enactment and revise them at least every two years. The factors include grid and transmission effects, water, noise and light, workforce, tax revenue and national security, including the risk of several data centers in one place, and the practices would name “ideal locations.” The text places no requirements on developers, states or localities. A second bill, H.R. 10322, would require data centers and other large energy users to pay the full cost of the grid and generation upgrades built for them, and would cut highway funds to states that do not adopt large-load tariffs. A third, H.R. 9372, has the National Institute of Standards and Technology develop standards for measuring data center energy and water use. A fourth, H.R. 8711, calls for a strategy to protect communities near data infrastructure from adversaries.

Another House bill points the other way. H.R. 10119, introduced August 20 by Republican Tom Barrett, would bar federal agencies from overriding state or local land-use, zoning, siting or permitting authority over data centers on non-federal land, and from conditioning federal funds on a state or locality giving up that authority.

The questions a strategy would have to answer

Read together, the proposals show what a national strategy would have to decide: where data centers go, how they are built, who pays, and who decides.

Where is the least settled. Current federal policy steers projects toward federal land and former industrial sites, and the one proposal for broader guidance is non-binding. A report published August 11, 2026 by the Center for Strategic and International Studies recommends encouraging development on brownfields, former industrial sites, federal property and places with existing energy and fiber, and argues that planned siting can cut permitting delays.

How they are built has so far been mostly a matter of measurement and reporting. The NIST bill would create measurement standards, and the disclosure bills would add reporting. On reliability, FERC ordered the North American Electric Reliability Corporation on July 16 to submit the first enforceable standards for computational loads by December 31, 2026.

Who pays has the most momentum, through the House-passed bill, the bipartisan Senate bill and FERC’s June orders. Who decides is the hardest. Land use has been a state and local matter, the White House framework keeps zoning with the states, and DOE’s request that FERC take over large-load connections reaches into a field that state commissions have historically held.

What other countries have tried

China has the longest record with a national siting plan. Introduced at the end of 2020 by four agencies, its “Eastern Data, Western Computing” program designates eight national computing hubs, three in the east and five in the west, and ten data center clusters. Power-hungry facilities were meant to go west, closer to renewable energy, while latency-sensitive work stayed near eastern cities. Targets included latency under 20 milliseconds for clusters, average availability of at least 65 percent and power usage effectiveness below 1.25.

Results are disputed. A July 10, 2026 ChinaTalk analysis argues that compute has not moved west in any meaningful way. It cites a Chinese state press report on a western computing center whose servers ran below 30 percent utilization, and says electricity was only about 5 percent of three-year costs at one 400-megawatt facility, so power prices did not decide where capacity went. It also points to skilled labor concentrated in the north, east and south. The plan itself assigns latency-sensitive work to the east and heavier, delay-tolerant work to the west, so how much of the western capacity is in use is the open question.

Germany, which like the United States divides power between federal and state governments, adopted its first national data center strategy on March 18, 2026. It aims to at least double capacity by 2030, calls for a concept of priority areas on brownfield sites such as former power stations, sets standardized siting criteria as a practical guide for local authorities, and looks at giving host municipalities a larger share of tax revenue. The law firm summary of the strategy lists no binding efficiency or waste-heat requirements, and says measures that need new federal money depend on funding. The European Commission separately proposed a Cloud and AI Development Act on June 3, 2026, aimed at speeding deployment across the EU.

Where the race stands

On one measure the United States is far ahead. The Brussels think tank Bruegel estimated in a September 10, 2026 brief that the United States holds about 35 gigawatts of operational AI compute, 78 percent of the world total, against about 5 gigawatts (11 percent) in China and 2 gigawatts (5 percent) in the EU. Planned projects would change those shares only modestly by 2031, to 68 percent for the United States, 15 percent for China and 6 percent for the EU. Bruegel’s subject is Europe, where it says speed to operation and grid access are the main limits and treats coordination among permitting authorities as the more binding constraint. These are one group’s estimates, and published figures for Chinese capacity vary.

The case for and against acting nationally

Supporters of a stronger federal role come at it from two directions that do not agree on what a strategy should do. One argues for speed. The executive order sets rapid buildout as a priority, CSIS calls domestic data centers increasingly a prerequisite for technology leadership, and House Speaker Mike Johnson and Majority Leader Steve Scalise urged Congress to act on the White House AI framework so the United States can beat China in the global AI race. The other argues for protection. Subramanyam’s package, Heinrich’s call for stronger rules and the bills on disclosure and community impact would use national standards to limit costs and harms to residents. Both treat Washington as the place to set terms. They differ on whether the terms should mainly clear the way or mainly set conditions.

Arguments for caution are also on the record. Land use has been a state and local matter, the administration’s own framework keeps zoning with the states, and H.R. 10119 would write that into law. In the Chinese record, as its critics read it, a plan built around power prices left capacity idle where demand and skilled workers were not. Germany’s strategy shows another possibility: a plan that is mostly guidance and depends on follow-through by states and towns. Wyden argues that Congress should act before the administration sets a precedent for handing public land to developers, and Heinrich’s objection shows that some who want national action think the House cost bill is too weak. Bruegel’s finding is a reminder that a strategy can stall on power and approvals even with a good siting list.

What to watch

The 119th Congress ends on January 3, 2027, and bills not enacted by then expire, so the Ratepayer Protection Act, the Senate affordability bill and the Subramanyam package must move in the lame-duck session or be reintroduced next year. FERC’s next step after the grid operators answer its June orders will show how much of the grid side becomes national and how much stays regional. NERC’s first-phase standards for computational loads are due December 31, 2026. The Paducah and Savannah River deals still have to clear negotiations and approvals.

Whether any of this becomes a single strategy is open. At present the federal government has a speed policy, a cost debate and one non-binding siting proposal, and no document that ties them together.

Sources
  1. White House — “Accelerating Federal Permitting of Data Center Infrastructure (Executive Order 14318)”
  2. White House — “Ensuring a National Policy Framework for Artificial Intelligence (Executive Order 14365)”
  3. Governing — “White House AI Framework Pushes for Broad Preemption of State Laws”
  4. Rep. Suhas Subramanyam — “Rep. Subramanyam Introduces National Data Center Plan”
  5. GovInfo — “H.R. 10321, Responsible Data Center Siting Act of 2026”
  6. GovInfo — “H.R. 10119, Protecting Local Control of Data Centers Act”
  7. Utility Dive — “House passes ratepayer protection bill for data centers”
  8. Newsweek — “Data center bill hits roadblock in the Senate: what happens next”
  9. Shale Magazine — “FERC Transmission Siting Bill Targets Data Center Costs”
  10. Mayer Brown — “FERC Large-Load Interconnection Preliminary Rulemaking: Key Takeaways”
  11. Duane Morris — “FERC Acts to Advance Data Center and Large Load Integration in Six RTO Regions”
  12. Utility Dive — “PJM, NERC data center and crypto reliability standards”
  13. American Nuclear Society — “Private-sector data center plans advance for Paducah and Savannah River sites”
  14. Sen. Ron Wyden — “Senate Republicans Block Wyden Bill to Ban Data Centers on Public Lands”
  15. Center for Strategic and International Studies — “Powering Innovation: Data Centers, Compute, and U.S. Competitiveness”
  16. China Briefing — “Eastern Data, Western Computing: China’s Big Plan to Boost Data Center Computing Power Across Regions”
  17. ChinaTalk — ““Eastern Data, Western Compute” is Fake”
  18. Gleiss Lutz — “Data centre update: Federal Government adopts first national Data Centre Strategy”
  19. European Commission — “Proposal for the Cloud and AI Development Act (CADA)”
  20. Bruegel — “How can Europe address its pressing AI compute infrastructure shortfall”
A Data Center Digest Original Story · 9 days ago
Market Digest Original

Old AI Chips Cost More to Rent This Year, Not Less. How Long Can That Hold?

A Data Center Digest Original StoryThis article was researched and written entirely by AI, without human review or editing, as part of the Data Center Digest's ongoing experiment in AI-powered journalism.

Our October 9 story on what happens to a data center's hardware at the end of its life covered the accounting side of a big question: how long does an AI chip last? It reported on Meta's and Amazon's changes to the assumed lives of their servers, the dispute between critics such as investor Michael Burry and defenders such as Yardeni Research, and used-chip prices from a dealer. This follow-up covers what that story left out: what renters are paying for older chips, how Nvidia's release schedule compares, and how lenders are treating chips that are booked as lasting six years. The evidence from 2026 leans toward the longer-life side so far, but nearly all of it traces back to a shortage.

What renters are paying

The clearest data point runs against the expectation that new chips make old ones cheap. SemiAnalysis reported that the one-year contract rate for an Nvidia H100, a chip that launched in 2022, rose from $1.70 per GPU-hour in October 2025 to $2.35 in March 2026, an increase of almost 40%. It said demand for H100s held firm even as newer Blackwell chips ramped up, and that some H100 contracts were renewing at the same rate as two to three years earlier, with some extended to 2028. Half the providers it contacted were sold out, and most had no Hopper-generation capacity coming off contract.

SemiAnalysis tied the squeeze to heavy demand from agentic AI workloads, open-weight models and AI lab fundraising, along with a surge in memory prices that led some operators to delay or drop deployments. Its authors drew a conclusion for the accounting debate: higher rental rates extend the economic life of existing GPUs, because the money invested keeps earning for longer before it needs replacing. That is the part to read carefully. The price rise happened because supply was short, so it shows an old chip is worth more when new ones are scarce. It does not show how much the chip will be worth when supply catches up.

What used chips sell for

Resale data gives a second, less tidy reading. Silicon Data, which tracks GPU listings, measures used and refurbished H100s against new ones, counting age from the H100's October 2022 launch. Its main table shows used H100s at a median of about 61% of the contemporaneous new price in 2024, when the chips were roughly two years old, and about 69% in 2025, at about three years. That sits a little below the 72% to 75% that dealer Caladan Semiconductor reported in the October 9 story. Both are listing prices from sellers, not audited sales, and Silicon Data says its numbers are not an accounting depreciation schedule. A different section of Silicon Data's page gives lower figures, 45% to 55% for used units at three years, so the page itself does not settle on one answer.

The baseline matters too. Silicon Data also compares resale prices with the highest price a new H100 ever listed at. By that yardstick, three-year-old chips sit at about 20% to 30%. A buyer who paid the peak price and a buyer who paid today's price have had very different experiences, and the share of a chip's value that survives depends on which one you are.

Nvidia's calendar

The supply side keeps moving. At its March 2025 developer conference Nvidia said it was “striving to announce new chip families on an every-year basis,” where before the AI boom it released new architectures every other year, according to CNBC. The company said Blackwell Ultra would ship in the second half of 2025, Vera Rubin systems in the second half of 2026, a Rubin Next generation in the second half of 2027, and a chip called Feynman in 2028. It also claimed cloud providers could earn up to 50 times the revenue from Blackwell Ultra compared with the Hopper generation, a company claim.

Whether older chips keep their rental value once newer ones arrive in volume is the test that has not yet been run. theCUBE Research argues they will, describing a cascade in which the newest hardware does foundation-model training for its first years, older chips move to real-time inference, and fully depreciated ones handle batch work. It expects hyperscalers to settle on a five-year schedule, down from six, but not two or three. That is an analyst framework with no measured utilization data behind it. The counterpoint the Motley Fool cited is a late-2024 Tom's Hardware report that data center GPUs running AI workloads at 60% to 70% utilization may last one to three years.

How lenders are treating the chips

The clearest example of money riding on the answer is CoreWeave, the specialist GPU cloud. It extended the assumed life of its technology equipment from five years to six at the start of 2023, a couple of years before its IPO, the Motley Fool reported. Its annual report shows what depends on it: technology and infrastructure depreciation and amortization rose from $843 million in 2024 to $2.3 billion in 2025, and total indebtedness stood at $21.6 billion at the end of 2025, with $4.4 billion of debt service paid during the year, including $3.4 billion of principal.

In March 2026 CoreWeave closed an $8.5 billion delayed-draw term loan that it described as the first investment-grade rated financing secured by high-performance computing infrastructure and an associated customer contract. Moody's rated it A3 and DBRS rated it A (low). The initial draw is up to about $7.5 billion, rising to $8.5 billion as the assets reach stabilization. Pricing is SOFR plus 2.25% on the floating portion and about 5.9% on the fixed portion, and the loan matures in March 2032, about six years after closing. Blackstone Credit & Insurance anchored the deal, with MUFG and Morgan Stanley as structuring agents, and the release says it was meaningfully oversubscribed. It describes the loan as non-recourse and secured by substantially all assets of a CoreWeave subsidiary, and it points to previously contracted cloud services without naming the customer.

The release does not say how the loan is sized against the chips. That matters because a lender repaid out of a long customer contract cares less about what the hardware is worth, so long as the chips keep running and the customer keeps paying. A lender counting on resale value cares a great deal. The sources reviewed here do not show which of the two the A3 rating leans on, and they do not include the lenders' or rating agencies' own analyses.

What to watch

Three things will say more than any forecast. Vera Rubin's shipments, due in the second half of 2026, will show whether the yearly cadence holds. Rental rates for H100s will show whether the 2026 increase reflects a lasting need for older chips or a shortage that eases once new supply arrives. And the next round of annual reports, due in early 2027, will show whether the big companies keep, stretch or shorten the lives they assign to their chips. For now the market is paying more for older chips than SemiAnalysis said many had expected, for reasons that may not last.

A Data Center Digest Original Story · 13 days ago
Power Digest Original

Washington Says Data Centers Hold 35 Gigawatts of Backup Power. In 2026, the Grid Began Asking for It.

A Data Center Digest Original StoryThis article was researched and written entirely by AI, without human review or editing, as part of the Data Center Digest's ongoing experiment in AI-powered journalism.

At about 7:56 a.m. on July 22, 2026, a mechanical failure took a 230-kilovolt line out of service in Northern Virginia. The grid handled the fault correctly. The data centers did not wait to find out. About 3,800 megawatts of their load disconnected within moments and moved to on-site backup generation, the largest such transfer in PJM's history, according to PJM. Over the same year, the federal government went the other way and asked data centers to run those same generators so the grid could stay up. In 2026 the data center was, in the space of a few months, both a source of grid trouble and a reserve for it. This piece sets out what happened, what the rules say about running the generators, and the alternative of asking data centers to use less.

When data centers leave on their own

The July event was the third of its kind in PJM's Dominion zone, the territory that holds Northern Virginia's data center cluster. PJM says similar events of about 1,500 megawatts each occurred on July 10, 2024, and February 17, 2025. The 2024 event began when a lightning arrestor failed on a 230-kilovolt line during a thunderstorm. According to GridStatus, six rapid voltage drops followed, and NERC found that data centers shifted from grid power to uninterruptible power supply systems. The facilities stayed off the grid for hours, and reconnecting them took manual intervention. Voltage and frequency moved outside normal ranges, and utilities heard complaints about flickering lights and noisy appliances, the Union of Concerned Scientists noted. NERC's president called it a “five-alarm fire when it comes to reliability.”

The 2026 event was larger. PJM reports the first wave moved 2,970 megawatts to backup generation as system voltage rose, and a second wave moved another 1,099 megawatts, causing further voltage disturbances. PJM's total load fell from 99,984 megawatts to 96,205 megawatts. The grid operator reported no significant reliability impacts and restored its area control error limit within nine minutes, against a NERC standard of 30. Operating Committee Chair Emanuel Bernabeu said the fault was cleared normally and the data centers “are too sensitive” to voltage changes, disconnecting “too early.”

Regulators are now writing rules. On July 16, FERC ordered NERC to submit the first phase of enforceable standards for computational loads by December 31, 2026, according to PJM, and minimum voltage and frequency ride-through requirements, which would keep a facility connected through a brief disturbance, are expected in 2027. PJM is also evaluating interconnection requirements for large loads, including ride-through standards, Utility Dive reported. Until those exist, an emergency that starts with one transmission fault can end with several thousand megawatts of demand disappearing at once.

When the grid asks them to run the generators

The second thread runs in the opposite direction. In December 2025, Energy Secretary Chris Wright argued that backup generators at data centers were sitting idle while utilities struggled to meet demand. “We have 35 gigawatts of backup generators that are just sitting there today,” he said, according to E&E News. In late January 2026, as Winter Storm Fern brought severe cold, DOE sent emergency orders to PJM and Duke Energy that could require data centers to run backup diesel generators, covering about a dozen states. PJM did not need to use the measure during that storm, Data Center Knowledge reported.

DOE issued a second order on May 18, 2026, running through May 20, after PJM requested it the day before. The agency cited a heat wave and transmission and generation outages tied to seasonal maintenance, and repeated its estimate of more than 35 gigawatts of unused backup generation nationwide. Another order covered the early-July heat wave. PJM issued backup-generator warnings on July 2 and July 3, the first of their kind, and no generators were dispatched, Data Center Knowledge reported. PJM called the warnings precautionary. Served peak load reached 162,713 megawatts, and unrestricted demand, which counts emergency actions and customer load reductions, was estimated at 168,158 megawatts, an all-time unrestricted peak. Independent Market Monitor Joseph Bowring said PJM “was getting very close to a shortage of primary reserves.” The coverage reviewed for this piece did not report that any data center generators ran in the January or May events either, so the 35 gigawatts remains an estimate of what is available, not a record of what has been used.

What the air rules allow

Running an emergency generator outside a true emergency runs into air rules. EPA's current framework, summarized by Sidley, sets no hourly cap on emergency operation and limits nonemergency operation to 100 hours a year. Demand response under a financial agreement with a third party counts as nonemergency, with up to 50 of those hours allowed if the local balancing authority or transmission operator dispatches the engine and the operator keeps records, among other conditions. In a May 2025 fact sheet, EPA said grid operators can require customers to start generators during demand response events, and that an Energy Emergency Alert Level 1 can justify emergency demand response when escalation to Level 2 is imminent. The guidance is nonbinding, and EPA said it is evaluating more substantive changes to the underlying regulations.

States are moving the other way on permits. Virginia's environmental agency said the “limited operating hours” basis for exempting emergency generators from stricter controls “is no longer appropriate,” and expects most future data center backup fleets to be permitted as non-emergency units. For permit applications filed on or after July 1, 2026, it set a presumptive limit of 0.60 grams per horsepower-hour, which effectively requires selective catalytic reduction. A separate law, HB 507, requires emissions limits at least as strict as the federal Tier 4 standard. The numbers behind the dispute differ by who is counting. A state legislative review found data center generators account for less than 4% of regional nitrogen oxide emissions, and the Data Center Coalition cited a finding that they often run 10 to 30 minutes a month. The Piedmont Environmental Council counts thousands of diesel generators representing more than 11 gigawatts of backup capacity in Virginia. The Southern Environmental Law Center said tighter controls are an improvement but may let data centers run generators far more often. Compass Data Centers said it keeps using older Tier 2 engines because the catalyst systems on Tier 4 units can slow or shut down an engine if the controls fail during an outage.

The alternative: use less, not generate more

A different approach asks data centers to cut demand briefly. A Duke University study from 2025 estimated that the 22 largest U.S. balancing authorities could add 76 to 126 gigawatts of new flexible load if it could be curtailed for 0.25% to 1% of the time, or roughly 85 to 366 hours a year, Utility Dive reported. PJM, at 18 gigawatts, had the most headroom at a 0.5% curtailment rate, ahead of MISO at 15 and ERCOT at 10. The study did not account for transmission constraints, which could shrink the figure.

Some operators are trying it. Google announced demand response agreements with Indiana Michigan Power and the Tennessee Valley Authority in August 2025, the first to cover machine learning workloads, which can be shifted away from peak periods. Google's Michael Terrell wrote that “there are limits to how flexible a given data center can be,” because services such as Search and Maps need reliability. Texas made curtailment a condition of connection. Senate Bill 6, signed in June 2025, requires large loads of 75 megawatts or more that connect to ERCOT to install shutoff equipment, so utilities can disconnect them during firm load-shed events, and creates a separate voluntary program that pays loads to reduce demand with at least 24 hours' notice.

The trade-offs

The two approaches pull against each other. Asking a data center to run generators adds local emissions and relies on equipment built for rare outages. Asking it to curtail leaves the facility's own reliability promises in doubt. Staying connected through a disturbance, the goal of ride-through standards, makes a facility less likely to add to the problem but also keeps a large load on a stressed grid. Virginia has added a wrinkle: according to Data Center Knowledge, a separate state law bars operators from meeting voluntary demand flexibility standards by running carbon-emitting backup generators, so in that state the generator route and the flexibility route cannot count as the same thing.

For a community near a campus, the practical questions are narrow. Is the site covered by a shutoff or curtailment obligation? Are its generators permitted as emergency or non-emergency units? Will it ride through the next voltage dip, or trip off? The answers now differ by state and by grid operator.

What to watch

NERC's first-phase standards for computational loads are due by December 31, 2026, with ride-through requirements expected in 2027. EPA could change the underlying generator rules. Texas is still building the programs SB 6 created. And next summer will test whether the warnings PJM issued in July 2026 stay precautionary.

States in this story

A Data Center Digest Original Story · 19 days ago
Construction Digest Original

Can a Steel Mill Become an AI Data Center?

A Data Center Digest Original StoryThis article was researched and written entirely by AI, without human review or editing, as part of the Data Center Digest's ongoing experiment in AI-powered journalism.

A closed industrial site comes with something an AI developer cannot buy quickly: a connection to the high-voltage grid. That is the pitch behind a wave of proposals to turn idle steel mills, aluminum smelters and retired power plants into data center campuses. In July 2025, Cleveland-Cliffs chief financial officer Celso Goncalves told analysts the company was exploring sales of idled mills, including to data center developers, because the sites offer access to power and water with the infrastructure already in place. Since then the idea has moved from an earnings call to signed deals, a federal guidance document and a long run of zoning hearings. This piece asks three questions: is it a good idea, has it worked, and are there better options. The reporting reviewed here points to an uneven answer. The conversions furthest along are aluminum smelters and power plants. The steel mills are, so far, proposals.

What the old site is selling

EPA's January 2026 guidance on redeveloping Superfund and brownfield sites as AI data centers says such sites may be strong candidates because of their preexisting infrastructure, likely industrial zoning and proximity to energy lines. It also lists what a site needs: typically at least 100 acres for a 100-megawatt campus, enough grid capacity and substation access with redundancy, a reliable water source, redundant fiber, and a cleanup status compatible with construction. For a brownfield, a state regulator's “No Further Action” letter may be required. The guidance adds that construction takes thousands of workers while operations need a small number of skilled staff. The cases below test that list.

The furthest along is a smelter in Kentucky

Century Aluminum shut its Hawesville smelter in 2022 because of high energy costs, a closure that cost about 600 jobs, according to WKU Public Radio. TeraWulf closed on the site in February 2026. Data Center Frontier reports it paid $200 million in cash plus a 6.8% interest in the development subsidiary, about $301.9 million in all. The smelter reportedly drew about 480 megawatts around the clock. The property has multiple high-voltage lines, an energized on-site substation and more than 250 buildable acres. TeraWulf's chief operating and technology officer, Nazar Khan, said the site “already had five independent transmission lines for redundancy.”

About five months after closing, TeraWulf announced a 20-year lease with Anthropic for roughly 401 megawatts of critical IT capacity, worth an estimated $19 billion. The campus has access to up to 480 megawatts of gross capacity, so the old smelter load and the new data center load are close in size. First capacity is due in the second half of 2027, with the full amount by early 2028. Nothing is operating yet. Data Center Frontier notes that the schedule leaves little room for delay in transformers, chillers and backup power, and that the $19 billion headline works out to roughly $950 million a year on a simple average, with actual revenue depending on phased rent starts. Local officials hope the project will offset the lost jobs; WKU Public Radio noted that data centers typically employ fewer than 100 people.

Other smelters show the limits

Alcoa shut its Massena East smelter in New York in 2014. On October 15, 2026, the company disclosed a definitive agreement to sell about 1,300 acres for a data center, for up to $315 million: $50 million to $65 million at closing, and $60 million to $250 million more tied to permitted data center capacity over as many as seven years. Alcoa has not confirmed the buyer. The price is a ceiling, not a figure. The site's interconnection historically supported about 435 megawatts of smelting load, while the proposed three-building campus would draw about 635 megawatts and need a new 200-megawatt substation, according to the county planning director, who has questioned where the power would come from. The developer has no power purchase agreement with the New York Power Authority. Data centers are not yet a permitted use in the town's industrial zone, and an earlier town revision meant to allow them in industrial zones was never officially approved.

Water is a second question there. At a May county planning board review, the application described about 2.7 million gallons a day withdrawn from the river for evaporative cooling, with about 1.6 million gallons a day lost to evaporation, and a planner said staff were told the applicant would not use a closed-loop system. The board chair raised groundwater concerns because of contaminated landfills already on the property.

In Maryland, the 2,100-acre former Alcoa Eastalco site shows how long a conversion can take even with an industrial pedigree. The plant opened in 1969 and closed permanently in 2010, and groundwater contamination had been found there in 1983. Quantum Loophole and TPG Real Estate bought it in 2021, and the planned power demand across the campus is about 2,400 megawatts, including 260 megawatts for Aligned and 300 megawatts for Amazon Web Services. Aligned, AWS and Rowan have all broken ground. Even so, a June 2026 tally by the environmental group Nature Forward found none of the data centers operating, and a referendum petition against the county's data center overlay zone drew more than 21,000 signatures.

The steel mills: proposals, not buildings

In the reporting reviewed for this piece, no AI-scale data center was operating on a former steel mill site. The steel projects sit at the zoning, cleanup and utility-study stage.

Alton, Illinois, is the newest. Alton Steel, which operated its mill for more than a century, ceased operations in January 2026. On October 21, developer TerraWatt proposed the Riverbend Technology Center on the 381-acre site: a $3 billion first block of 200 megawatts, growing to as much as 1 gigawatt in 200-megawatt steps. The first phase would demolish 1.26 million square feet of mill buildings. The 200-megawatt block is sized to what the mill used to draw, about 160 megawatts according to consultant Cecilia Dvorak, who called 200 megawatts roughly what Ameren Illinois could deliver in the short term. She described the full buildout as a “chicken and egg” problem. Cleanup is expected to be extensive, with scrap and slag piles and regulated materials that may include chromium, lead, manganese, mercury, nickel and zinc, and a 2023 environmental assessment needs updating before costs can be set. TerraWatt says it will pay for the cleanup. No formal permit or zoning application has been filed, and the contingent purchase agreement, which TerraWatt expects to close in 2027, depends on rezoning and an Ameren Illinois study. Alton Steel chief executive Chris Ervin said: “It's not a done deal.”

In Kentucky, Rubix Data Centers, part of Spain's Submer Group, has proposed a 2-gigawatt campus on roughly 500 acres of the former AK Steel Ashland Works, spread across Russell, Boyd County and Ashland. The company plans closed-loop cooling and about 200 full-time jobs. It also faces moratoria: Russell has a six-month pause that the company wants lifted, Boyd County finalized a 180-day pause, and Ashland's was set to end in August, according to The Daily Independent in July.

In Pennsylvania, developer Brian O'Neill wants a 2-million-square-foot data center in ten existing buildings on 66 acres at the Cleveland-Cliffs mill in Conshohocken, with natural gas generation on site. Plymouth Township's planning board unanimously refused to recommend it in October 2025. The zoning hearing board met on August 6, 2026, in a session the Philadelphia Inquirer described as raucous, and made no ruling; the hearing was expected to resume in September. In Gibraltar, Michigan, Raeden proposed reusing buildings at the former McLouth Steel plant for a 100-megawatt inference data center. The city approved a 12-month moratorium, and Raeden and the property owner sued in federal court, saying the pause unlawfully stopped their application.

Is it a good idea?

The numbers in these cases point to the main limit. The two aluminum smelters carried about 435 and 480 megawatts, which is close to the size of an AI campus. The one steel mill with a published figure, Alton, used about 160 megawatts, and its developer is planning in blocks because that is what the utility can deliver. A mill's historic draw records what the site once used, not what the utility will supply today. As Dvorak put it, final capacity depends on what the utility can realistically provide. Seen this way, an old connection is a down payment on the power supply, not the whole thing. Massena would need a new substation and a power contract, and Ashland's 2 gigawatts is far beyond any mill's past load.

Cleanup is the least visible cost. EPA's guidance warns that redeveloping contaminated sites may carry federal liability for contamination, with protections for buyers who complete the required inquiries and follow continuing obligations, while state programs set standards for brownfields. Alton's cost cannot be set until an assessment is updated, and Massena's board chair pointed to landfills already on site. Jobs and politics matter too. A mill that once employed hundreds is replaced by a facility that needs a small operating staff, and every steel proposal above has met a moratorium, a zoning dispute or organized neighbors, or still faces a permit it has not yet filed. On this evidence, the idea works best for sites that come with a large, still-valid grid connection, workable zoning and a cleanup that is finished or fundable, and works poorly when the old load is small or the permits are far off.

Has it been done? An older precedent

Conversions are not new, only newly large. By July 2015, Keystone NAP was serving customers from a 180,000-square-foot building that was once part of U.S. Steel's Fairless Hills operation in Pennsylvania, using the site's existing electrical infrastructure and a nearby microgrid. At that time it had 13 megawatts of capacity to the building, expandable to 30, according to Data Center Frontier. That is a fraction of the campuses above, and its record since then is outside this piece.

Are there better options?

The retired power plant is the other template. In April 2025, Homer City Redevelopment and Kiewit announced a gas-fired campus on the site of Pennsylvania's 1,884-megawatt Homer City coal plant, which closed in mid-2023. The 3,200-acre site has interconnections to the PJM and New York grids and sits about five miles from a major gas pipeline. The plan calls for up to 4.5 gigawatts of generation and a $10 billion first investment. FracTracker, which visited on April 1, 2026, reported that developers had announced the start of vertical construction alongside demolition at the old plant. The coverage reviewed named no tenant.

Federal land is another. On July 29, 2026, the Department of Energy announced a $100 billion campus at the former Paducah Gaseous Diffusion Plant in Kentucky, which closed in 2013. Brookfield would develop and operate a 1.8-gigawatt data center on about 600 leased acres, with NextEra building a gas plant of about 2 gigawatts and up to 2.6 gigawatts of battery storage. First components could arrive in 2028 and construction would finish in 2031, but the Kentucky Public Service Commission still has to approve the power service agreement, and DOE cleanup of the plant is expected to run through 2065.

What these sites share is not steel. It is a large, energized connection, a single seller, enough acreage and, at Homer City, its own fuel supply. By that measure, the better options are the sites that bring more of those traits: retired power plants with generation and fuel, smelters whose load matches an AI campus, and federal land under one owner. A steel mill is a good candidate when it has them and a poor one when it does not. The sources reviewed do not offer a head-to-head cost comparison with new greenfield sites. Data Center Frontier says existing connections can shorten utility interconnection waits that often delay new campuses by years, which is the case for the old sites.

What to watch

Hawesville's first capacity is due in the second half of 2027, an early test of an AI lease on a former industrial site. In Alton, watch whether a rezoning request is filed and what the Ameren Illinois study finds. Massena's zoning and its power contract will decide whether the $315 million ceiling is reached, with closing expected within six months of the October 15 disclosure. Conshohocken's zoning hearing board, Gibraltar's lawsuit and the moratoria around Ashland will show how much local politics, not the old power lines, decides which mills become data centers.

Sources
  1. Manufacturing Dive — “Cleveland-Cliffs looks to sell idle steel plants to data center developers”
  2. U.S. Environmental Protection Agency — “Guidance on the Redevelopment of Superfund and Brownfield Sites as AI Data Centers (January 2026)”
  3. Data Center Frontier — “TeraWulf's $19B Anthropic Lease Puts Its Brownfield AI Strategy to the Test”
  4. WKU Public Radio / Louisville Public Media — “Data center developer purchases former western Kentucky aluminum smelter”
  5. Baxtel — “Alcoa To Sell Idle Massena East Smelter Site For Up To $315M Data Center Build”
  6. NNY360 — “County Planning Board: Massena data center site plan contains more questions than answers”
  7. Nature Forward — “Quantum Frederick, Frederick County, MD: Maryland Data Center Series, Pt. 3”
  8. St. Louis Magazine — “Former Alton Steel property would become a data center under TerraWatt plan”
  9. St. Louis Public Radio — “Armory data center developer wants a similar project in Alton”
  10. The Daily Independent — “Rubix to Russell: “It's ultimately up to you””
  11. WHYY — “Conshohocken data center proposal garners pushback”
  12. The Philadelphia Inquirer — “Plymouth Twp. zoning meeting turns raucous over Brian O'Neill's plans for a data center”
  13. WXYZ Detroit — “Gibraltar approves data center moratorium amid former McLouth Steel site plan”
  14. WXYZ Detroit — “Data center developer sues Gibraltar, says moratorium unlawfully halted project”
  15. Data Center Frontier — “From Steel to Data: Keystone NAP's Modular Power Play”
  16. Utility Dive — “Largest US gas-fired power plant planned for data centers in Pennsylvania”
  17. FracTracker Alliance — “From Coal Plant to AI Campus: FracTracker Documents Construction at Homer City”
  18. WKMS — “$100B data center, gas-fired power plant to be developed around Paducah DOE site”
A Data Center Digest Original Story · 22 days ago
Cooling Digest Original

A Plain-English Guide to Data Center Cooling

A Data Center Digest Original StoryThis article was researched and written entirely by AI, without human review or editing, as part of the Data Center Digest's ongoing experiment in AI-powered journalism.

Water is the data center fact most likely to arrive without its context. A county says a project is contracted for 2 million gallons a day. A company says its new campus consumes essentially no water for cooling. A researcher says the industry uses billions of gallons. These statements can sit side by side because they measure different things: different parts of the cooling system, different points in the water's trip, and sometimes different power plants. This piece starts with a car and a house, then builds up to the vocabulary that sorts the numbers out.

Start with the car and the house

Every data center has one job that never stops: moving heat from the chips to somewhere else. Two familiar machines show the basic choices. A car engine circulates coolant in a sealed loop. The coolant picks up engine heat and carries it to the radiator, where passing air takes the heat away. Nothing is used up. A home air conditioner works on the same idea: a sealed loop of refrigerant carries heat outdoors, and a fan blows air across the outdoor unit to release it. Neither machine needs a water supply.

In data center terms, these are closed loops that hand their heat to the air, usually through equipment called dry coolers. The cost is electricity for fans and compressors, and the work gets harder on hot days, when the air that has to take the heat is already warm.

The other approach is closer to sweating. A cooling tower sends warm water through a stream of moving air, and a small portion of it evaporates, carrying heat away. The same effect makes a swamp cooler work in a dry climate, and it is why a breeze on wet skin feels cold. Evaporation removes heat very effectively for little electricity. The price is the evaporated water, which leaves as vapor and does not return to the pipe or the river. This is what people mean by evaporative cooling, and it is often what “open loop” means in data center talk; University of Georgia Extension material uses the label that way.

“Closed loop” describes only half the system

Researchers who study data center water describe a two-stage process. In the first stage, heat moves from the servers to a heat exchanger inside the facility, through air or through a liquid loop that is sealed and uses no water up. In the second stage, the facility sends that heat outdoors, either through wet cooling towers that evaporate water or through air-cooled dry coolers. That description comes from a March 2026 preprint by Yuelin Han, Pengfei Li, Adam Wierman and Shaolei Ren of UC Riverside, RIT and Caltech. A building can have a sealed liquid loop in the first stage and still use cooling towers in the second. Margaret Cook of the Houston Advanced Research Center put the consequence plainly in CBS Austin's reporting: “Whatever that technology is will dictate the water consumption.”

That is why the same words appear on both sides of the debate. QTS Data Centers told Texas lawmakers that every facility it has built since 2019 uses a closed-loop system that needs essentially no water for cooling once operating. Berkeley Lab's national model found the opposite end of the range: data centers with water-cooled chillers and no economizers, the equipment that lets a chiller rest when the weather allows, had the highest on-site water use, because of cooling tower evaporation. Hyperscale sites, it found, commonly use outside air for much of the year with only occasional water-assisted cooling, which keeps their water use minimal.

The trade: water here, electricity there

Neither choice is free. Evaporative cooling uses substantially more water on site than dry cooling, and it uses less electricity. The UC Riverside preprint cites a large technology company's statement that at its Louisiana campuses, evaporative cooling cuts electricity demand by 25 to 35% compared with dry cooling during summer peaks. Google says water cooling suits sites where the local source is not at high risk of depletion, and reports that its Oregon campuses in The Dalles needed only 6% to 10% extra energy for overhead in the fourth quarter of 2025, against a 54% industry average, which it ties in part to water cooling. Those are company figures.

Dry cooling moves part of the footprint somewhere else. Cook says closed-loop and air-cooled designs raise power needs. Microsoft says its newest design, which circulates coolant at the chip and uses no evaporation, raises annual energy use by a small amount compared with its evaporative designs, partly offset by running warmer. It says the design avoids more than 125 million liters of water a year per data center, based on its fiscal 2024 fleet average, with pilot designs planned for Phoenix and Mount Pleasant, Wisconsin, in 2026 and new sites coming online from late 2027.

Power plants use water too. Berkeley Lab estimated that U.S. data centers consumed about 66 billion liters (about 17 billion gallons) directly in 2023, and nearly 800 billion liters (roughly 211 billion gallons) indirectly through the electricity they bought, about 12 times as much. That estimate is based on regional grid mixes and does not account for power purchase agreements or on-site generation. The comparison depends on the local grid and the local water source, so a design that saves water on site is not automatically better or worse once the power plant is counted.

Withdrawn, consumed, discharged

The second source of confusion is the verb. Water withdrawal is water taken from a river, an aquifer or a utility. Consumption is the part that does not come back, usually because it evaporates. Discharge is the difference: the part returned right away, typically to a sewer or treatment plant. A shower and a lawn sprinkler show the gap. Nearly all the water from a shower goes down the drain and back to treatment, while much of what a sprinkler puts out evaporates or is taken up by the lawn. The UC Riverside preprint estimates that homes consume 5 to 15% of what they withdraw, public water supplies as a whole about 12%, and data centers 70 to 90%.

Google's reporting for 2025 gives a rare side-by-side. Across its data centers, it withdrew 13,562 million gallons, discharged 3,039 million and consumed 10,523 million, about 78% of what it took. Google calculates consumption by subtracting discharge from withdrawal. By site, Council Bluffs, Iowa, withdrew 1,746 million gallons and consumed 1,346 million. Ashburn, Virginia, withdrew 76.5 million and consumed 53.2 million. The share returned runs from about 9% at its Berkeley County, South Carolina, site to about 30% at Ashburn. A reader who sees only “withdrew” or only “consumed” can be off by a large margin in either direction.

A third number is the contract. When Cardinal News wrote about Google's proposed site in Botetourt County, Virginia, the figure was 2 million gallons a day under contract, or 730 million gallons a year, with a possible future expansion that could require 8 million a day. A contract is a supply commitment. The column compares it with the annual withdrawals Google reports at its existing Virginia sites: 359.6 million gallons at Bristow, 258.4 million at Sterling, 238.1 million at Leesburg and 76.5 million at Ashburn. That is a fair comparison of withdrawal, but it says little about consumption.

Timing matters as much as volume. The UC Riverside preprint finds that water demand clusters on the hottest days, when a large evaporatively cooled data center can need more than 1 million gallons a day and some planned campuses up to 8 million, which strains local water systems even when annual totals are modest. It projects that by 2030 U.S. data centers could withdraw 80 to 150 billion gallons a year and consume 60 to 110 billion, about 4 to 7% of consumptive use in U.S. public water systems. Google's framing is that data centers' water consumption is small nationally, less than 1% of what Americans use on lawns each year, but that local watersheds can still feel it. Both can be true.

Five questions for the next headline

When a data center water number appears, five questions sort out most of the confusion. Is it withdrawal, consumption or a contract limit? Is it a yearly total or a peak day? Does it count only the building, or also the power plants behind it? What cooling design is it, and is “closed loop” only the inner stage? And what is the local water source and its condition? Google says it uses a water risk framework to match cooling to each site, and chose water cooling at The Dalles because local sources were not at high risk of depletion or scarcity.

What to watch

The policy debate is moving toward these distinctions. In Texas, state Sen. Charles Perry said at a Senate hearing that he expects evaporative cooling for data centers to be barred after the next session, and lawmakers plan to take up the issue in January; CBS Austin's report describes no enacted rule or bill text. Cook argued for efficiency standards over technology mandates, because a mandate could become outdated and regulators should weigh a facility's full water and energy footprint. That is the trade in this piece in policy form: a rule that cuts water on site can raise electricity demand.

On the company side, Google says its water projects replenished about 7.7 billion gallons in 2025, roughly 78% of its freshwater consumption, against a goal of 120%. Microsoft's chip-level pilots will show how its design performs outside the lab. None of the sources claims one design is right everywhere. What they describe is a choice that depends on the local climate, the local water source and the local grid, and a set of numbers that mean little until it is clear which one is being counted.

States in this story

A Data Center Digest Original Story · 25 days ago
Market Digest Original

Insurers Aren't Fleeing Data Centers. They Are Running Out of Room to Hold Them.

A Data Center Digest Original StoryThis article was researched and written entirely by AI, without human review or editing, as part of the Data Center Digest's ongoing experiment in AI-powered journalism.

A common story about data centers and insurance is that insurers are pricing the industry out of the market. The numbers do not show that. Marsh's index for the third quarter of 2026, released October 21, has US property insurance rates down 10%, matching a 10% drop globally that was the largest of any line of business. Aon's Ryan Barber told an industry forum in early October that there is ample capacity for data center projects valued at $3 billion to $5 billion, and that placements above $10 billion are now possible. The strain is somewhere else: in the size of the biggest sites, and in how many of them an insurer is willing to hold in one place.

A soft market, and a ceiling at the top

Marsh's release does not mention data centers, but it describes ample capacity and strong competition across major lines of insurance, which is the backdrop to the data center story. Within that market, the largest campuses are a different kind of risk. Swiss Re Corporate Solutions says a traditional data center once needed about $2 billion in limits, while one hyperscale campus can now require $20 billion for the buildings alone. S&P Global Ratings puts the insurable value of the largest projects at as much as $30 billion. Brokers say Munich Re will take roughly $250 million net on a single data center construction project, so reaching $10 billion on one site means stacking a tower across 40 or more carriers.

Zurich's figures show how fast the scale moved. The average insured data center project there rose from about $150 million to $3 billion in five years, and Zurich's head of construction and surety, Kelly Kinzer, has said the market does not have enough capacity to insure these projects at full value. Zurich's 2026 US report, as summarized by a construction law firm, found that many projects are insured for only one-third to one-half of their total value. The same summary says some projects reportedly carry limits of $2 billion to $5 billion on campuses worth many times that. Insurers increasingly size coverage to an estimated or probable maximum loss, not to the full replacement cost.

The problem is concentration

Barber called tracking accumulation the biggest struggle right now for insurers and reinsurers. Campuses are built in phases, which leaves an insurer holding builders risk on one building and operating property exposure on the next, at the same address. The contents add to it. Computing equipment is often several times more valuable than the buildings that house it, Facilities Dive reported in its coverage of MSIG USA, and S&P's Patricia Kwan said insurers are not designed for that level of concentration in one location. The geography is also shifting: Zurich reports that 64% of capacity under construction in 2026 is outside traditional hubs such as Northern Virginia, so insurers are being asked to underwrite new locations as well as larger sites.

What goes wrong: fire, weather, and waiting for equipment

Allianz Commercial's analysis of 221 data center claims, worth about $782 million, found that fire drives severity. Fire accounted for well over half the value of those claims even though it was only 14% of claims by number. Water damage was the most frequent claim, at 21%, and willful acts, including crime and cyber crime, made up 19%. Allianz points to lithium-ion batteries built into server racks, in place of separate battery rooms, as a source of thermal runaway, and the report notes a 2025 fire at a South Korean government data center in Daejeon, believed to have started with a battery explosion, that shut down 600 servers. Tyler Ahrenhold of MSIG USA said on-site turbines or batteries raise the risk of fire or explosion tenfold compared with a grid-only site, and that his company asks for at least 100 feet between a power source and a building. He added that insurers rely on best practices because no code addresses the spacing.

Weather is the other large driver. Severe weather has been the leading cause of loss in Zurich's US data center builders risk portfolio for three years. Swiss Re's modeling puts more than a quarter of US data center capacity in areas with three or more large-hail days a year, and over 40% in zones with significant to very high tornado risk. Then there is waiting. Allianz cites Schneider Electric modeling in which business interruption makes up almost a third of climate-related losses, with more than 90% of that coming from upstream supply chain disruption rather than direct damage, and it notes switchgear lead times of up to 80 weeks and transformer lead times of up to 50 weeks. It also describes a startup delay caused by power disturbances that produced a loss of 50 million to 100 million euros at one hyperscale facility. Coverage for delay in startup has become harder and more expensive to buy at meaningful limits, according to the construction law firm's summary of the Zurich report.

Who holds the gap

When insurance falls short of full value, someone else holds the difference. S&P's Kwan said lenders generally want full-value protection and resist coverage sized to maximum loss, and that any uninsured gap flows back to the sponsor and shows up in the ratings of the data center's debt. The construction lawyer Kenny Cantrell argues that contracts, more than policies, are becoming the main way the financial consequences of a major loss get divided, with owners and contractors both scrutinizing indemnities, liability caps and delay provisions.

The industry is also looking for more capital. Zurich's Kinzer said he expects talks about securitizing data center insurance risk to become more necessary as project values rise, though he said the market is not there today. Euler ILS Partners is working with insurers to underwrite specialist data center policies, and investors in insurance-linked securities, who typically take damage risk but not the risk that a data center proves unprofitable, have asked Aon how to take part. Premiums are expected to grow. Swiss Re's Michael LaRocca projects data center premiums rising toward $24 billion from about $10.6 billion today, S&P estimates that construction coverage alone could generate about $10 billion in premiums this year, and Barber estimates more than $100 billion in new premiums over the next five to ten years. Liberty Mutual's Marc Orloff cautioned that the premium is not fully recurring, because builders risk premium falls if construction slows.

What to watch

Several open questions follow from this. One is whether lenders keep accepting coverage below full value as campuses grow past what the market can place. Another is whether securitization and specialist capital arrive in time to matter. A third is whether fire standards for on-site power catch up, since insurers are filling a gap that codes leave open. And the equipment and supply-chain exposure ties back to the long lead times the Digest covered in its piece on transformers. None of the sources predicts a particular outcome. What they describe is an insurance market with plenty of room for the average project, and a hard limit on the largest ones.

A Data Center Digest Original Story · a month ago
Opposition Digest Original

Split in 2025, Opposed by Most in 2026: Inside the Fast Turn in Public Opinion on Data Centers

A Data Center Digest Original StoryThis article was researched and written entirely by AI, without human review or editing, as part of the Data Center Digest's ongoing experiment in AI-powered journalism.

In August 2025, Heatmap Pro and Embold Research asked 3,741 registered voters whether they would support or oppose a data center being built near them. The country was evenly split: 44% in favor, 42% opposed. A year later, with the question worded the same way, 75% were opposed and only 4% were strongly in favor, down from 13%. Other pollsters, using their own questions and samples, have recorded the same direction through 2026, with opposition to a local data center landing between 61% and 75% depending on the poll. Gallup's first reading, in March, put it at 71%, compared with 53% for a nuclear plant in the same survey; opposition to a local nuclear plant has not passed 63% since Gallup began asking in 2001.

The turn, in five polls

Heatmap's own series is the cleanest because the wording never changed. Opposition was about 42% in August 2025, 51% in February 2026, about 70% in May (with 55% strongly opposed), and 75% in August, when more than six in ten were strongly opposed. Emerson College found a similar slope among likely voters: 42% opposed to an AI data center in or near their community in December 2025 and 63% in a July 19 to 20 poll, with 27% in support and 10% neutral, according to Newsweek's report of the survey. The Annenberg Public Policy Center at the University of Pennsylvania found 49% opposed in a spring wave and 61% by mid-July, with support falling from 21% to 14%; the center called it the largest shift on any AI question in its surveys. In mid-September, PBS News, NPR and Marist found 65% of registered voters opposed to construction in their own community, 43% of them strongly. That is higher than the 58% who said new construction by large tech companies would have a mostly negative effect on the country.

The turn shows up at the state level too. A Washington Post and George Mason University Schar School poll of 1,101 Virginia registered voters, taken March 26 to 31, found 35% comfortable with a new data center in their community, down from 69% in 2023. Northern Virginia is the largest market on the Digest's own map. Support for tax breaks for data centers that create at least 1,000 jobs fell from 61% to 37%, and 67% of those polled said the state's sales-tax exemption should end, against 26% who would keep it.

What changed in people's minds

Pew Research Center's American Trends Panel shows the shift as a change in perceived effects. Between January and a July 20 to August 9 survey of 10,548 adults, the share calling data centers mostly bad for the environment rose from 39% to 54%, for home energy costs from 38% to 50%, and for the quality of life of people nearby from 30% to 49%. About 4% called any of those three mostly good. Local jobs and local tax revenue, the benefits developers cite most, drew a split verdict: about 22% called each mostly good, while the share calling them mostly bad rose from 15% to 24% for jobs and from 12% to 21% for tax revenue. The cost side of the ledger gained new believers; the benefit side did not.

Pew also found 60% not too or not at all comfortable with a new data center in their area, and 7% extremely or very comfortable. Awareness rose over the same months: 35% said they had heard a lot about data centers, up from 25% in January, and the share who had heard nothing at all fell from 25% to 12%. The survey cannot say whether attention drove opinion, but the two moved together. The movement crossed party lines: the share of Republicans calling data centers mostly bad for nearby quality of life rose from 24% to 42%, and among Democrats from 37% to 57%.

What people say worries them

Gallup followed its March poll with an open-ended question in April to 2,054 panel members, asking supporters and opponents to explain themselves in their own words. About two-thirds of supporters cited economic benefits, including 55% who named jobs and 13% who named tax revenue. Opponents gave a wider range of reasons. Half cited excessive resource use, including 18% who named water and 18% who named energy. Another 16% cited pollution, such as noise and air or water contamination. About one in five raised quality-of-life effects such as population growth, traffic or land use, and a similar share cited economic costs such as higher utility bills and building costs. Most of the remaining opposition came from general or specific concerns about AI itself.

A November 2025 YouGov survey of 1,000 Americans, designed by Harvard's Stephen Ansolabehere just before the turn, suggests why quality of life can outweigh the electric bill. Two-thirds of respondents expected a nearby data center to raise electricity prices and 58% expected more pollution, yet support sat near neutral on a zero-to-one scale, at 0.51, between an auto factory (0.57) and a petrochemical plant (0.37). Nearly two-thirds also expected a data center to create long-term jobs, and a similar share expected economic growth. When the analysis weighed all the factors together, perceived effects on community quality of life were about three times as important to support as any other single factor, electricity prices included. People who expected a data center to lower prices were 36 percentage points more likely to support it, but that gap roughly halved once the other factors were included.

Heatmap's August 2025 poll tested arguments directly. The most convincing case for a data center was construction and operating jobs, which 63% found very or somewhat convincing, followed by local tax revenue for schools and services. The argument that data centers are needed to win the AI race with China scored far lower, at a net of plus 4. The most convincing case against was large electricity use raising utility bills (net plus 46), followed by heavy water consumption (net plus 34). Heatmap's reading was that tangible local effects mattered more to respondents than abstract ones. In Virginia, the Post and Schar School poll found 57% saying data centers are hurting home energy bills and 59% saying they are harming the local environment.

Who opposes, and where the politics moved

The opposition is bipartisan but uneven, and the Republican shift is the sharpest. In Heatmap's August 2025 poll, net support for a nearby data center was plus 14 among Republicans, minus 5 among independents and minus 8 among Democrats. A year later, Republicans were underwater by 43 points, independents by 65 and Democrats by 75. Annenberg's July numbers show a narrower gap, with 69% of Democrats, 54% of Republicans and 53% of independents opposed, and Gallup found 56% of Democrats, 39% of Republicans and 48% of independents strongly opposed.

Education draws a sharper line than geography does. Emerson found 79% of respondents with a postgraduate degree opposed, up 27 points since December, against 72% of college graduates (up 25) and 37% of those with a high school education (up 9). Annenberg found opposition at 70% among adults under 30 and 57% among those 65 and older. Place matters less than might be expected: Heatmap found rural voters underwater by 63 points, with urban and suburban voters only a few points more supportive, and PBS, NPR and Marist found opposition above 60% in big cities, suburbs, small towns and rural areas alike. Familiarity with AI does not predict it either. Annenberg found 64% of people who had not used AI opposed to local data centers, against 60% of heavy users, and said baseline AI use did not predict later opposition.

The politics are already visible, weeks before the November midterms. In Pew's September 28 to October 4 survey, 61% of registered voters said they dislike political leaders who support building data centers near them, and 8% said they like such leaders. Among Democratic candidate supporters the dislike figure was 75%; among Republican candidate supporters it was 44%, with 42% feeling neither way.

What the polls can and cannot say

Question wording explains part of the spread between 61% and 75%. Annenberg asks about new local data centers, Emerson about AI data centers in or near the community, Gallup about AI data centers in the local area, Heatmap about a data center near the respondent, and Pew about comfort. Only a few of these are true trend lines, meaning the same question asked repeatedly: Heatmap's four waves, Emerson's two, Annenberg's two, Pew's two, and the Virginia comparison with 2023. The rest are single readings that add context but not direction. All of them record attitudes toward hypothetical or generic projects, not reactions to a specific proposal with specific terms, and none shows how those attitudes translate into approvals or denials. On that side, PBS reported that Data Center Watch counted at least 75 projects worth roughly $130 billion blocked or delayed in the first quarter of 2026.

Little in the published polling tests what could move opinion back. The Harvard survey found that people who expect lower electricity prices are more supportive, and Heatmap's 2025 poll found that renewable power and local jobs persuaded some respondents, but those results predate the turn, and whether they still hold is an open question. What the data do show is a majority position that formed within about twelve months, across parties, ages, education levels and kinds of place, and that rests, in Gallup's open-ended answers, on resource use, pollution and local quality of life more than on any single issue.

States in this story

A Data Center Digest Original Story · a month ago
Market Digest Original

Where the Next Big Data Center Markets Will Be, According to the Digest's Own Data

A Data Center Digest Original StoryThis article was researched and written entirely by AI, without human review or editing, as part of the Data Center Digest's ongoing experiment in AI-powered journalism.

The Digest's map already shows where the old hubs are running out of room. It flags six regions as constrained: Northern Virginia, where Dominion's interconnection queue has reached 26 GW and new large loads wait two to four years; Phoenix, with 18 to 36 month waits; the Bay Area and Los Angeles; New York and New Jersey; and New England. Demand does not disappear when a hub fills up, so it moves. This article asks where it is moving next, using three things the Digest tracks: power headroom and pipeline size from the map, the policy climate from the Legislation tracker and the map's moratorium layer, and momentum from the more than 600 stories and posts the Digest has published since Aug. 15. The calls below are judgments drawn from that data, not a forecast model.

Start with the backdrop. The map covers 41 markets, six of them primary, with roughly 119 GW of capacity and pipeline listed across them. The Legislation tracker follows 255 bills, and 62 percent are classified as anti-development against 18 percent pro. The moratorium layer shows local restrictions, enacted or proposed, in 27 states, 18 of them already in effect. A separate Digest report on stalled projects cited Goldman Sachs saying only 50 to 60 percent of planned capacity will come online as expected over the next two years. Friction is the normal condition almost everywhere, so the markets most likely to grow are the ones where power is available and the friction is lower, already settled, or still ahead.

One pattern runs through the newest big projects: they sit where power infrastructure already exists or is being built alongside them. Homer City, Pennsylvania, is a former coal plant site where a 4.5 GW gas plant is planned, and AWS filed in October for a 36-building, 1,100-acre campus there. Aligned broke ground in September on a 2 GW campus at Shippingport, Pennsylvania, next to a retired coal plant, with its own on-site power. In Maryland, the Quantum Frederick park is built on a former aluminum smelter site and carries about 2.4 GW of combined planned demand. Crusoe's Goodnight campus in the Texas Panhandle, with Google as the end user, connects directly to an adjacent wind farm. The next hubs look less like the incumbents' dense fiber-and-colocation clusters and more like power sites that gained data centers.

The strongest case is Pennsylvania. The map lists three planned projects there that together exceed 7.7 GW: Homer City at 4.5 GW, Shippingport at 2 GW, and Nebius's Schuylkill County campus at 1.2 GW. Only Texas was mentioned in more Digest stories since Aug. 15 (22 for Pennsylvania against 39 for Texas, by a count of state mentions). The policy picture is mixed: the tracker holds 30 Pennsylvania bills, 8 pro and 21 anti, with every pro bill still in committee, and on Oct. 1 the state House Energy Committee advanced three ratepayer bills, including one that would require developers to pay for the utility upgrades that serve them. The moratorium layer lists Pennsylvania as proposed, not enacted. If the projects deliver, Pennsylvania has the pipeline of a top-tier market; the open question is whether new cost rules slow it.

Next is West and Northwest Texas. The map shows West Texas at about 2.5 GW with the 1.2 GW Stargate campus at Abilene leading, and the Panhandle's Goodnight campus is under construction with about $29 billion of investment expected, more than 1 GW by one estimate. Crusoe has also filed plans for two data centers in Jayton. Texas led all states in Digest coverage, with 24 of its 39 stories tagged to construction, power, AI, market, or investment topics. The risk is executive, not legislative: the tracker has no Texas bills, but on Sept. 21 Gov. Greg Abbott ordered the state's environmental regulator to halt all data center permits until audits of the grid and water board finish, expected in mid-December, and the Digest's report cited BloombergNEF saying the pause threatens up to 20 percent of the national pipeline. Expect Texas to keep growing, with the December audit as the checkpoint.

The third likely rise is the Gulf South. Northeast Louisiana is already shown as a primary market, with Meta's Hyperion campus targeting up to 5 GW and a separate $3.6 billion campus at Boyce, and the map rates its power availability as good. Across the border, DeSoto County, Mississippi, is shown at about 2.5 GW, anchored by xAI's roughly 2 GW campus, with AWS and Compass also committing billions elsewhere in the state. The political signal here is thin rather than favorable: the tracker has one Louisiana bill and none for Mississippi, and the two states drew only four Digest stories between them. The main exposure is permitting, since xAI's gas turbine approval in Mississippi is under legal challenge.

Three states have the pipeline but also the loudest pushback. Michigan's Stargate campus in Saline Township ($16 billion, 1.4 GW) broke ground in June, yet the tracker shows 37 Michigan bills, 22 anti and 8 pro, and Republican senators introduced a package in September for a one-year statewide moratorium, although such bills frequently stall in committee. Ohio's Pike County campus is described as up to 10 GW, but three of the four Ohio bills tracked are anti, the moratorium layer lists Ohio as proposed, and the state Supreme Court recently kept a citizen zoning measure off the Wilmington ballot, showing how contested local rules are becoming. Wisconsin's Port Washington campus (902 MW of IT capacity) sits in a state the map marks as having active moratorium action. These markets will keep building, but likely more slowly than their announced pipelines.

The quieter candidates are where the data shows headroom and little noise. The map rates Wyoming, Iowa, the Pacific Northwest, and the Southwest Power Pool states of Kansas, Nebraska, and Oklahoma as high availability, the best tier. Microsoft closed a land purchase for its Cheyenne campus in June and has framed it as the base for a multi-gigawatt buildout. Indiana lists more than 4 GW planned, including Amazon and Meta commitments of $36.8 billion, yet appeared in only two Digest stories. Kansas has a pro bill awaiting its governor, and Evergy reported advanced talks for 2 to 3.5 GW of new demand. The caution is that Missouri and Nebraska each show three anti bills out of three tracked, and a Digest report found Nebraska's grid under rate pressure. If a surprise market emerges, it is probably in this group.

Two wild cards could go either way. New Mexico's Project Jupiter, a 2.45 GW Stargate campus in Dona Ana County, is stalled on a gas pipeline the state land office rejected, and Oracle sent a force majeure notice on Sept. 24, while Google is exploring a closed-loop, air-cooled campus in Lea County. In Utah, the Stratos project in Box Elder County was approved unanimously by county commissioners and has drawn organized opposition from residents. The incumbents also deserve a caution against writing them off. PG&E told CNBC it has a 12.7 GW pipeline of data center requests even as the map calls California critically constrained, a reminder that constrained does not mean closed.

The limits of this analysis are worth stating. Pipeline gigawatts are announcements, not completed capacity. The tracker finds bills through a keyword search, so states with few tracked bills may simply have little visible legislation, and the absence of Texas bills from the tracker is a gap in what it captures more than a sign of calm. Story counts reflect what the Digest's pipeline published, which was interrupted by two outages in late September and early October, and they include some non-US stories published before the Digest narrowed its scope. The signs to watch are specific: the December ERCOT audit in Texas, whether Pennsylvania's ratepayer bills pass, whether Michigan's moratorium bills leave committee, and the new Jupiter pipeline timeline in New Mexico. The next big markets are probably already visible; the data suggests they will be the ones where power arrives before the politics catches up.

Sources
  1. The Data Center Digest: US Data Center Build Map (markets, power availability, grid constraints, moratorium layer)
  2. The Data Center Digest: Legislation tracker (255 bills, sentiment and status by state)
  3. The Data Center Digest: “Opposition, Permits, and Grid Scarcity Stall U.S. Data Center Pipeline”
  4. The Data Center Digest: “AWS Files Plan for 36-Building, 1,100-Acre Data Center Campus in Pennsylvania”
  5. The Data Center Digest: “Aligned Data Centers Breaks Ground on 2GW Pennsylvania Campus at Former Coal Site”
  6. The Data Center Digest: “Pennsylvania House Committee Advances Three Bills to Shield Ratepayers From Data Center Costs”
  7. The Data Center Digest: “Google Confirmed as End User for Crusoe's Amarillo Campus, With $29 Billion Investment Expected”
  8. The Data Center Digest: “Crusoe Files Plans for Two Data Centers in Jayton, Texas”
  9. The Data Center Digest: “Texas Governor Abbott Freezes All TCEQ Data Center Permits Pending ERCOT Audit”
  10. The Data Center Digest: “Michigan Republican Introduces Three Bills for One-Year Data Center Moratorium”
  11. The Data Center Digest: “Ohio Supreme Court Blocks Wilmington Data Center Zoning Ballot Measure”
  12. The Data Center Digest: “Nebraska Grid Faces Rate Pressure as Data Centers and Aging Infrastructure Drive Demand”
  13. The Data Center Digest: “Oracle Sends Force Majeure Notice to Blue Owl Over New Mexico Pipeline Rejection”
  14. The Data Center Digest: “Google Explores First Self-Built Data Center in New Mexico's Lea County”
  15. The Data Center Digest: “Utah Residents Fight Back Against 40,000-Acre Wonder Valley Data Center”
  16. The Data Center Digest: “PG&E Reports 12.7-Gigawatt Data Center Pipeline Amid Wildfire Liability Concerns”
A Data Center Digest Original Story · a month ago
Impact Digest Original

The Afterlife of a Data Center: Resold GPUs, Mounting E-Waste, and Who Pays for the Cleanup

A Data Center Digest Original StoryThis article was researched and written entirely by AI, without human review or editing, as part of the Data Center Digest's ongoing experiment in AI-powered journalism.

The equipment inside a data center has a shorter working life than the building around it, and the companies that own it do not agree on how much shorter. Meta extended the assumed life of most of its servers and network equipment to 5.5 years in January 2025, up from a previous range of four to five years, a change its annual report says cut 2025 depreciation expense by $2.92 billion and raised net income by $2.59 billion. Amazon moved part of its fleet the other way: effective January 2025, a subset of its servers and networking equipment went from a six-year life to five, which its annual report attributes to the pace of technology development in artificial intelligence and which added $1.4 billion in depreciation expense. Both figures are accounting estimates, not measurements, and they decide how much of the cost of AI hardware shows up in earnings each year.

The disagreement is sharpest for GPUs. Critics, including investor Michael Burry, argue that AI chips should be depreciated over less than three years, and that their true economic life may be one to three, a view Yardeni Research summarizes alongside the point that Nvidia has moved to a one-year product cycle. Defenders, including Yardeni, say older GPUs keep earning as they shift from training the largest models to less demanding inference and other work, so their useful lives run longer than the critics assume. The answer determines when hardware leaves a data center, and where it goes next.

Part of that answer is a resale market. According to price ranges published in May 2026 by Caladan Semiconductor, a used-GPU dealer, used Nvidia H100 SXM5 cards were trading at roughly $18,000 to $24,000 against $25,000 to $32,000 new, or about 72 to 75 percent of the new price. Used 80GB A100s fetched about 55 to 65 percent of new, down from more than 80 percent in 2022, according to the same source. Caladan reported that most used H100s were coming from failed startups and distressed sales rather than enterprise upgrades, while A100 supply was growing as companies that bought clusters in 2021 and 2022 moved to newer chips. The figures come from a dealer, and the article does not say whether they are asking or completed-sale prices, so they are a rough guide and not audited market data. Companies such as Reconext, a remarketer, now sell processes that wipe drives and test GPUs in place so racks can be resold without being pulled apart.

What does not get resold becomes electronic waste, a category already growing faster than the world's ability to handle it. The UN's Global E-waste Monitor 2024 counted a record 62 million tonnes generated in 2022, up 82 percent since 2010 and on track for about 82 million tonnes by 2030. Only 22.3 percent was documented as properly collected and recycled, and the report found generation rising five times faster than documented recycling. Researchers writing in Nature Computational Science in 2024 estimated that servers built for generative AI alone could add up to 5 million tons of e-waste cumulatively through 2030, and as much as 2.5 million tons a year by 2030 if nothing is done to reduce it, about 3 percent of the UN's projected 2030 global total. They also estimated that reusing and remanufacturing components, extending equipment life, and better recycling could cut that output by up to 86 percent.

The building itself is a separate problem. Ashwin Murthy of Columbia Law School's Sabin Center for Climate Change Law wrote in August that older abandoned data centers often sat vacant for long stretches, kept up by a skeleton crew while owners looked for buyers, and that modern hyperscale facilities are harder to repurpose because they can span millions of square feet and are specialized by design. Idle sites, he warned, risk becoming abandoned warehouses that waste land and leave hazardous material behind. There is also a cost that can fall on others: because data centers require heavy grid upgrades, he wrote, ratepayers may end up paying for infrastructure built for a company that fails and leaves a facility idle.

Some local governments are writing cleanup rules in advance. Murthy points to Susquehanna County, Pennsylvania, which requires decommissioning within twelve months after a data center's useful life ends, including removal of all hazardous materials and contents, backed by bonding, and notes that Smithfield Township has a draft ordinance. He argues that states could go further by requiring detailed decommissioning plans before permitting that bind any future owner, and bonds or similar financial assurances that cover the full cost, modeled on oil and gas bonding rules. Whether a locality can impose such rules varies by state, and he notes that rules aimed only at data centers could face equal-protection challenges, though he expects those to fail.

For now, how long a data center lasts depends on assumptions in annual reports, a resale market priced by dealers, and a patchwork of local ordinances. Who ends up paying when facilities are retired, whether shareholders, customers, or the town that hosted them, will depend on which of those holds up.

A Data Center Digest Original Story · a month ago
Market Digest Original

Six New Markets Join the Digest Data Center Map as Texas Pauses Permits and Chicago Weighs a Moratorium

The Data Center Digest has refreshed its interactive Data Center Map, adding six markets and updating the regulatory status of several existing ones. The map now covers 41 markets, up from 35.

The new markets reflect projects that have moved into the news in recent weeks. In the Texas Panhandle, Crusoe's Goodnight campus east of Amarillo is under construction with Google as the end user and about $29 billion of investment expected. In southern New Mexico, Oracle and OpenAI's 2.45 GW Project Jupiter campus is facing pipeline delays, and Oracle sent a force majeure notice to its developer on Sept. 24. The map also adds the Quantum Frederick park in Frederick, Md., where Amazon is building four data centers; Vantage's $15 billion-plus Lighthouse campus in Port Washington, Wis.; Aligned's 2 GW Project Phoenix campus in Shippingport, Pa., which broke ground Sept. 10; and the Stratos project in Box Elder County, Utah, which has drawn organized opposition.

Several existing entries changed as policy moved. Texas Gov. Greg Abbott directed state regulators on Sept. 21 to halt all data center permits until audits of the state grid and water board are complete, and the map's Dallas-Fort Worth, West Texas, and San Antonio entries now reflect that pause. In Chicago, the mayor proposed a 12-month moratorium ordinance on Sept. 23 that is before a City Council committee. Charlotte's City Council has set an Oct. 12 public hearing on extending its moratorium, which expires Nov. 5. Loudoun County, Va., is scheduled to vote Oct. 20 on a resolution pausing legislative data center applications for up to 12 months, and New York Gov. Kathy Hochul has until Dec. 31 to act on the Responsible Data Center Development Act. The Homer City, Pa., entry now includes Amazon Web Services' filing for a 36-building, 1,100-acre campus.

The map's legislation layer continues to update automatically from the Digest's legislation tracker.

Explore the map →
The Data Center Digest · a month ago
Power Digest Original

AI Data Centers Are Built in Under 18 Months. The Transformers They Need Take Years.

A Data Center Digest Original StoryThis article was researched and written entirely by AI, without human review or editing, as part of the Data Center Digest's ongoing experiment in AI-powered journalism.

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.

States in this story

A Data Center Digest Original Story · 2 months ago
Construction Digest Original

The AI Buildout's Biggest Bottleneck Isn't Chips. It's Electricians.

A Data Center Digest Original StoryThis article was researched and written entirely by AI, without human review or editing, as part of the Data Center Digest's ongoing experiment in AI-powered journalism.

Associated Builders and Contractors says the U.S. construction industry needs to attract 349,000 net new workers in 2026 and 456,000 in 2027 just to keep pace with demand, citing the ongoing artificial intelligence infrastructure buildout as one driver. No trade is harder to fill than electricians: 81 percent of firms with electrician openings reported difficulty filling them in the AGC and NCCER 2026 Workforce Survey, released in September, the highest difficulty rate of any craft tracked, ahead of mechanics (79 percent), HVAC technicians (77 percent), and concrete workers (76 percent). The survey's own framing put it plainly: construction workforce shortages remain acute even as data centers strain the labor supply.

Electrical work makes up an estimated 45 to 70 percent of a data center's total construction cost, and hundreds of electricians can be on a single site at peak. Microsoft president Brad Smith has called the shortage “perhaps the single biggest challenge” to the company's U.S. data center expansion, telling a Senate Commerce Committee hearing in 2025 that Microsoft has electricians commuting as far as 75 miles or temporarily relocating from across the country just to keep projects staffed. Smith has estimated the country needs roughly half a million new electricians over the next decade, against a current base of about 780,000 active electricians and another 80,000 open positions today.

Age is compounding the gap. About one-fifth of electricians are over 55, according to ABC, and Smith has put the share of union electricians between 50 and 70 years old at nearly 30 percent. The Bureau of Labor Statistics projects 9 percent job growth for electricians through 2034, faster than the average occupation, with roughly 81,000 openings a year, and cites the AI-driven data center boom explicitly as one reason electricity demand, and the workforce needed to build for it, keeps rising. The cost of getting this wrong is steep: a month's construction delay on a 60-megawatt data center can cost a developer roughly $14.2 million in lost lease revenue and penalties, according to research firm STL Partners, and a 2026 Kelly staffing survey found 90 percent of data center operators call staffing shortages a critical constraint, with a quarter of data center personnel hired away by competitors every year.

Labor unions remain the most established pipeline, and data center construction has visibly swelled their ranks. IBEW Local 24, which covers the Baltimore-Washington corridor, has roughly doubled its annual apprenticeship class from around 100 to 120 apprentices to a 230-person class this fall, with plans to grow it to 275 to 300, business representative Rico Alba Carries said; the local has added about 1,000 members, roughly 50 percent growth, over the past several years. A single project in Frederick County, Maryland, has kept 700 to 800 electricians on site plus about 150 more staffing a prefabrication shop, according to union leaders. That pipeline still runs on a five-year apprenticeship timeline, a pace the AI buildout's own schedule cannot compress.

Data center operators and their contractors have built shorter, targeted programs alongside the union track. Compass Datacenters and Meridian Community College in Mississippi launched a tuition-free, 14-week program in 2026 tied to Compass's roughly $10 billion campus investment there, teaching hands-on mechanical, electrical, and IT infrastructure work with no prior degree required; Compass CEO Chris Crosby called it “tailor-made to ensure Meridian has the workforce to meet that future.” In central Ohio, home to more than 50 data centers run by companies including Amazon, Meta, and Cardinal Health, Columbus State Community College offers a Data Center Technician certificate backed by Amazon-funded scholarships of $1,000 per semester. Microsoft's Datacenter Academy has built a similar model at more than 20 community colleges and technical schools across the country, donating server racks and networking equipment for hands-on labs and running roughly 16-week internships into roles like datacenter technician and critical environment technician.

For workers, the opening is real. The median electrician earns $62,350 a year, according to BLS, without the four-year degree and student debt a comparable salary would typically require elsewhere, and entry-level data center technician postings often start lower, commonly cited in the $18 to $26 an hour range, as a lower-barrier way into the same industry.

Even so, the programs now running are measured in the hundreds of students at a time, against a national gap counted in the hundreds of thousands. How fast the buildout continues may depend less on how many GPUs a company can buy than on how quickly the country can train the people who wire, cool, and maintain the buildings those GPUs sit in.

States in this story

A Data Center Digest Original Story · 2 months ago
Market Digest Original

Research Lab, Cloud Factory, or Colo Warehouse: A Field Guide to Data Centers

A Data Center Digest Original StoryThis article was researched and written entirely by AI, without human review or editing, as part of the Data Center Digest's ongoing experiment in AI-powered journalism.

When people hear that a “data center” is coming to their county, they often picture one generic warehouse full of blinking servers. In practice, the industry builds at least three genuinely different kinds of facilities under that name, and the difference matters: who owns it, what it's for, and how much say a community will ever have in what happens there next.

The rarest and most narrowly defined type is the research computing center: publicly funded supercomputers built to answer scientific questions rather than sell a product. The U.S. Department of Energy operates the world's three fastest supercomputers, Frontier at Oak Ridge National Laboratory, Aurora at Argonne National Laboratory, and El Capitan at Lawrence Livermore National Laboratory, all housed inside government-run national labs rather than commercial campuses. Frontier reached 1.1 exaflops on the standard performance benchmark and roughly seven exaflops of AI performance when it launched in 2022, and DOE says its exascale systems support work including climate modeling, fusion reactor materials design, and COVID drug research. Aurora, released to outside scientists in January 2025, has since been used to discover new battery materials, design drugs, and accelerate fusion energy research. These systems are open to outside academic researchers who apply for time on them, a structure almost nothing else on this list shares, and they are essentially never what a community is fighting over at a local zoning hearing, since national labs sit on federal land rather than land a private developer is trying to rezone.

Most of what actually gets proposed in a town today falls into a second category: commercial data centers built and owned by a single company to run its own products, most often a hyperscale cloud or AI campus owned outright by Amazon, Microsoft, Google, Meta, or an AI lab like OpenAI or xAI. This is also the fastest-growing category by far. Hyperscale operators controlled 44% of the world's data center capacity in the first quarter of 2025, according to Synergy Research Group, which projects that share will reach 61% by 2030. A commercial hyperscale campus typically has one tenant, meaning the company that builds it controls every future decision about what gets added, cut, or repurposed on that site.

The third category is colocation: facilities built and operated by a specialist company that leases out space, power, and cooling to many different customers rather than running its own product on the servers inside. Equinix and Digital Realty dominate the two main flavors of this business, retail colocation (many smaller tenants) and wholesale colocation (fewer, larger ones), respectively, with QTS, Iron Mountain, CoreSite, and newer entrants like Vantage Data Centers and Applied Digital also competing for share. The U.S. colocation market is on pace to reach $46.84 billion in 2026 and grow to $72.37 billion by 2030, increasingly driven by AI-focused tenants such as CoreWeave and Lambda Labs renting GPU-dense space rather than traditional enterprise IT. A colocation facility's tenant list can and does change over its lifetime, which makes it harder for a community to know years in advance exactly who will be operating inside.

A fourth category, enterprise-owned data centers that a bank, hospital, or retailer builds purely to run its own internal IT, still exists but is shrinking fast: Synergy pegs on-premise enterprise capacity at 34% of the world total in early 2025, down from 56% just six years earlier, and forecasts it will fall to 22% by 2030 as more companies shift workloads to the cloud. That leaves hyperscale commercial campuses and colocation as the two types a growing share of new local projects will actually be, with research computing centers remaining a rare, federally sited exception most communities will never encounter directly.

For a resident trying to make sense of a groundbreaking announcement nearby, the fastest way to tell these apart is asking one question: who owns it, and who's actually going to work there. A single company's name on the permit generally means a hyperscale commercial campus. A property developer or REIT's name usually means colocation, with tenants to be named later. And a federal agency's name means, almost certainly, that it isn't open to the public at all.

A Data Center Digest Original Story · 2 months ago
Policy Digest Original

Data Centers Just Became Military Targets. Washington Hasn't Decided What to Do About It.

A Data Center Digest Original StoryThis article was researched and written entirely by AI, without human review or editing, as part of the Data Center Digest's ongoing experiment in AI-powered journalism.

Iranian drones struck two Amazon Web Services data centers in the United Arab Emirates on March 1, 2026, and a third AWS facility in Bahrain on April 1, in what Rest of World and other outlets described as the first known military strikes on a hyperscaler's infrastructure. The attacks, launched amid Iran's broader retaliation for U.S. and Israeli strikes on its territory, knocked out services for regional apps including the ride-hailing platform Careem and payment service Alaan, and the April strike on Bahrain's ME-SOUTH-1 facility, operated by telecom company Batelco and described as Amazon's largest Middle East data center, caused a fire, according to Bahrain's interior minister. Days later, Iran's Islamic Revolutionary Guard Corps named 18 U.S. companies, including Microsoft, Apple, Alphabet, Meta, Amazon, and Nvidia, as legitimate military targets in the region.

“Physical attacks are only going to become more common moving forward as AI becomes more significant,” Sam Winter-Levy of the Carnegie Endowment told Rest of World. IDC's Ashish Nadkarni described the shift more starkly: “Now suddenly, protecting data centers is like protecting top-security government offices.” The industry has historically built its defenses around cyberattacks and natural disasters; the drone strikes forced a reckoning with the idea that a data center could simply be bombed.

That reckoning has domestic roots too. On April 6, 2026, Indianapolis city-county councilman Ron Gibson said someone fired 13 shots into his home just after midnight while he and his 8-year-old son were inside; a note reading “No Data Centers” was left at the door. No one was injured, and police called it an isolated, targeted incident with the FBI assisting. Gibson had voted days earlier to support a rezoning petition for a Metrobloks data center project in his district. Jordyn Abrams, a research fellow at George Washington University's Program on Extremism, told the Associated Press that data centers have increasingly become a symbol for grievances spanning the political spectrum, citing energy and water consumption concerns and confidential utility power deals as recurring flashpoints.

Security researchers say the physical exposure compounds an already underappreciated vulnerability in data centers' operational technology. More than half of data center professionals identified human threats, internal or external, as their biggest security risk, according to a 2026 AFCOM survey cited by Bloomberg Law. Anthony Ferrante, global cybersecurity head at FTI Consulting, told Bloomberg Law that compromising the systems controlling cooling, fire suppression, and access control “could essentially shut down” a facility's computers entirely. Dave Wulf, co-founder of the Center for Cross-Sector Coordination, warned that a data center failure can cascade into banking, pipelines, hospitals, and defense systems that depend on it.

Congress has not settled on how to respond. The House Homeland Security Subcommittee on Cybersecurity and Infrastructure Protection has held hearings on whether data centers, or AI infrastructure more broadly, should become the 17th federally recognized critical infrastructure sector. “If a major data center is attacked, disrupted, or taken offline, the consequences can reach far beyond one company,” Rep. Andy Ogles said, arguing the current framework lacks a clear, unified approach. Industry witnesses split on the remedy: USTelecom's Robert Mayer and the Space Information Sharing and Analysis Center's Samuel Visner backed a standalone sector designation, following the United Kingdom's own move to designate data centers as critical infrastructure, while the IT-ISAC instead created a “special interest group” for data center providers rather than pushing for new federal machinery. The debate is complicated by capacity at the agency that would have to run it: CISA's headcount has fallen from roughly 3,300 to about 2,200 employees over the past year, including 96 of 189 people in its Stakeholder Engagement Division, according to a letter five House Democrats sent the Government Accountability Office asking it to investigate what capabilities were lost.

A year ago, data centers were mostly discussed in terms of power bills and zoning fights. Between a foreign government's drones, a councilman's front door, and a Congress still arguing over jurisdiction, the industry is now confronting a threat model it did not build for, and no clear consensus yet on who is responsible for closing the gap.

A Data Center Digest Original Story · 2 months ago
Policy Digest Original

As Data Center Opposition Hits Record Highs, Both Sides Have a Playbook to Bridge It

A Data Center Digest Original StoryThis article was researched and written entirely by AI, without human review or editing, as part of the Data Center Digest's ongoing experiment in AI-powered journalism.

Community opposition, not power or land, has become the U.S. data center industry's central bottleneck in 2026. Data Center Watch, a project of AI intelligence firm 10a Labs, found that opponents blocked or delayed at least 75 projects worth about $130 billion in the first quarter of 2026 alone, the most in any quarter since the group began tracking in 2023. The number of active opposition groups more than doubled over the same stretch, from 396 at the end of 2025 to 833 across 49 states by March. Separately, Carbon Direct found that at least 46 AI data center projects worth $170 billion were publicly delayed or cancelled between January 2024 and May 2026, with a lack of transparency around ownership, power, and water needs the single most common cause. JLL's own research captures the underlying gap: 93% of communities agree data centers matter in the abstract, but only 35% support one built near them.

That gap is reshaping state and local law at a record pace. State lawmakers introduced at least 375 data center bills by mid-July 2026, according to The Washington Post, up from 243 in all of 2025, 75 in 2024, and 54 in 2023. Nine states have proposed statewide moratoriums, and 27 states are advancing “large load” legislation that requires developers to cover the cost of new energy infrastructure, with California, Ohio, and Utah already enacting versions of it. Seven major AI companies, including Amazon, Google, Meta, Microsoft, and Oracle, signed a voluntary Ratepayer Protection Pledge with the White House in March 2026, committing to cover their own power and infrastructure costs, though the pledge carries no legal enforcement mechanism. As the Digest has reported, that legal uncertainty is also opening a new front in litigation: law firms including ArentFox Schiff, Davis Wright Tremaine, and Taft have built dedicated data center practice groups in 2026 as opposition groups increasingly challenge project approvals on procedural grounds rather than waiting on new legislation. The real friction now sits locally: state-level moratorium bills have faced resistance in statehouses, while dozens of municipalities have moved ahead with local construction pauses instead. New Hampshire is a case in point: Governor Kelly Ayotte says she will seek a multi-year statewide moratorium after the head of ISO New England told her a large data center would raise regional energy prices.

Across the Digest's own coverage and trade outlets including Data Center Dynamics and Data Center Frontier, the same idea keeps surfacing: trust, not physics, is now the binding constraint on growth. “The next constraint on data center growth, we thought it was going to be power, but it may be community acceptance and political durability,” Loudoun County Economic Development's Buddy Rizer said on a Data Center Frontier podcast, alongside PR strategist Adam Waitkunas of Milldam Public Relations. Some community opposition is genuinely merited; a lot of it stems from mistrust of unfamiliar corporate entities, a lack of basic information, or a project becoming a stand-in for broader anxiety about AI. Below are five moves each side can make to close that gap.

Five things developers can do

  1. Lead with radical transparency on power, water, and ownership before site selection becomes public. Carbon Direct's research found that opaque ownership structures, NDAs with local officials, and undisclosed end users consistently triggered faster, sharper opposition; the firm recommends treating transparency as a siting strategy rather than a legal afterthought.
  2. Negotiate real, enforceable Community Benefit Agreements, not PR gestures. As one CBA advocate wrote for Data Center Dynamics, a strong CBA “is not a public-relations gesture or a list of voluntary commitments”; it is a negotiated, enforceable partnership that defines responsibilities and impacts, giving developers predictability and communities tangible, long-term value.
  3. Map opposition and engage stakeholders before acquiring a site. Identifying local advocacy groups, environmental organizations, and civic leaders early, before land is under contract, can prevent costly delays once a project becomes public.
  4. Fund the infrastructure a community actually needs, not just tax revenue. Communities increasingly expect developers to pay for the power and water infrastructure their projects require; some operators have pledged to replenish more water than they consume or to fund local education and workforce programs.
  5. Drop the NDA-heavy playbook with local officials. Confidentiality agreements that limit what elected officials can say publicly create a perception of secrecy that can poison a relationship before a developer has formally introduced itself, even when there are legitimate commercial reasons behind them.

Five things communities can do

  1. Organize early and understand real negotiating leverage before a developer shows up. Communities that organize ahead of time can negotiate more effectively, aligning corporate needs with local priorities on things like workforce pipelines, rather than reacting only after a project is already announced.
  2. Push for a formal Community Benefit Agreement process instead of ad hoc protest. A CBA process typically moves through public education, negotiation, and contract drafting, and once signed it obligates a developer to deliver specific, monitored commitments rather than vague promises.
  3. Use zoning and permitting tools to shape projects rather than simply block them. Mason, Michigan is a working example: rather than an outright ban, its city council adopted a new zoning framework in February 2026 that tightened local rules on data centers after a contentious public meeting.
  4. Demand transparency as a condition of engagement, not a courtesy. Since opaque disclosure of ownership, power, and water needs was the single most common reason cited across Carbon Direct's 46 delayed or cancelled projects, communities that insist on upfront disclosure are better positioned to evaluate real tradeoffs instead of relying on rumor.
  5. Separate legitimate, project-specific concerns from generalized anti-AI sentiment. Distinguishing real site-specific harms, like a documented water or noise impact, from broader distrust or misinformation strengthens a community's credibility and its actual negotiating leverage with both developers and state regulators.

The throughline across nearly every source here is the same: both sides do better with a formal, enforceable Community Benefit Agreement negotiated early, before a site is acquired and before NDAs harden positions. Tax revenue alone is no longer enough to earn public support, and a blanket moratorium is not a substitute for the transparent, ongoing oversight that actually resolves these conflicts, which is exactly the middle ground a well-built CBA is designed to occupy.

States in this story

Michigan (2)New Hampshire (2)
A Data Center Digest Original Story · 2 months ago
AI Digest Original

As Healthcare Rushes Into AI, It's Running Into a Global Compute Shortage

A Data Center Digest Original StoryThis article was researched and written entirely by AI, without human review or editing, as part of the Data Center Digest's ongoing experiment in AI-powered journalism.

The National Institutes of Health's All of Us Research Program illustrates how much computing power modern medical research already demands. In its June 2026 data release, the program made genomic and health data from more than 747,000 participants available to registered researchers, bringing total enrollment past 883,000. The dataset now includes more than 535,000 whole genome sequences, 1.3 billion genetic variants, and nearly 482,000 linked electronic health records, all housed on a secure Researcher Workbench built on Google Cloud Platform. “There’s a paradox at the heart of precision medicine,” NIH Director Jay Bhattacharya said in announcing the release. “To tailor treatments to individuals, you actually need very large populations to uncover the patterns that connect genetics, lifestyle, and the environment to health outcomes.”

That kind of population-scale genomic analysis runs on infrastructure most people never see. Biowulf, the NIH’s in-house Linux computing cluster, is described by the agency’s Center for Information Technology as the world’s most powerful supercomputer dedicated solely to biomedical research. It has grown to more than 100,000 processor cores and 60 petabytes of storage, is used by roughly 75% of NIH principal investigators, and in fiscal year 2024 alone consumed more than 1 billion core-hours and 6 million GPU-hours. The system helped the Telomere-to-Telomere Consortium publish the first complete, gapless human genome sequence in 2022, and NIH says COVID-19 research on Biowulf consumed more than 87 million CPU hours and produced more than 50 peer-reviewed publications.

That demand is emerging just as the broader market for AI compute tightens. Apollo Global Management’s wealth insights team described on-demand GPU capacity as “effectively sold out” in a June 2026 analysis, noting that rental rates for Nvidia H100 GPUs rose from roughly $1.85 an hour in late 2025 to about $2.40 an hour by March 2026, while spot prices for high-bandwidth memory chips climbed roughly eightfold since early 2025. Apollo named healthcare specifically as one of the sectors “only beginning to deploy AI at scale” just as that squeeze sets in, alongside legal services and financial analysis. Data Center Knowledge reported the binding constraint has shifted from the power shortages that dominated 2024 and 2025, when Microsoft CEO Satya Nadella said the company had “a bunch of chips sitting in inventory that I can’t plug in,” to semiconductor manufacturing capacity itself. “Silicon is the binding short-term constraint. Power is the binding long-term constraint,” HyperFrame Research’s Stephen Sopko told the outlet.

Drug makers with the capital to do so are responding by building and owning compute capacity rather than competing for it on the open market. Recursion Pharmaceuticals brought online BioHive-2, a supercomputer built with Nvidia using 504 H100 GPUs delivering 2 exaflops of AI performance, which debuted at No. 35 on the TOP500 list of the world’s most powerful supercomputers. Recursion’s chief technology officer said the company can now get “80% of the value with 40% of the wet lab work” that traditional drug discovery required. Roche has gone further, announcing an AI factory spanning more than 3,500 GPUs across facilities in Europe and the United States, built on Nvidia’s newest Blackwell chips and expected to be fully operational by early 2027. “Everybody wants to get their hands on Nvidia chips,” biopharma AI consultant Christian Hein told SWI swissinfo.ch.

Hospitals, by contrast, mostly cannot build their own supercomputers and instead depend on outside vendors and cloud infrastructure for the same AI capacity pharma companies are racing to own. In 2026, the U.S. Department of Veterans Affairs began scaling its ambient AI scribe tool from a 10-site pilot to all of its more than 130 medical centers nationwide. Mount Sinai Health System integrated OpenEvidence’s AI clinical decision support directly into its Epic electronic health record system across all seven of its hospitals. HonorHealth rolled out Abridge’s ambient documentation platform to roughly 3,000 physicians and advanced practice providers, and the University of Texas System’s earlier pilot with Qualified Health generated more than $15 million in run-rate financial impact within six months before expanding across all eight of its health institutions. Each of these systems now depends on rented, high-capacity cloud infrastructure to keep those tools running, in a market where that same infrastructure is getting harder and more expensive to secure.

That dependence is starting to show up in how healthcare organizations plan their own facilities, too. Commercial real estate firm Colliers has noted that hospitals and health systems increasingly compete with data center developers for the same scarce inputs, reliable electricity, available land, and network connectivity, while their own IT infrastructure has to support higher computing density and GPU-ready systems that older hospital data centers were never built for. For an industry historically organized around clinical staffing and bed capacity, the amount of processing power behind a diagnosis, a drug candidate, or a documentation tool is quickly becoming a resource healthcare organizations can no longer take for granted.

A Data Center Digest Original Story · 2 months ago
Policy Digest Original

As Data Center Moratorium Fights Escalate, Law Firms Build Practices on Both Sides

A Data Center Digest Original StoryThis article was researched and written entirely by AI, without human review or editing, as part of the Data Center Digest's ongoing experiment in AI-powered journalism.

A wave of local moratoriums and lawsuits over data center development is creating a new legal specialty, and law firms are positioning on both sides of it. Major firms including ArentFox Schiff, Davis Wright Tremaine, Jenner & Block, Norton Rose Fulbright, and Taft have all built or expanded dedicated data center practice groups in 2026, while a separate and newer trend has opposition groups and developers alike turning to litigation as moratoriums spread across dozens of jurisdictions.

ArentFox Schiff operates a group it calls Data Center Legal Solutions, led in part by Amy Antoniolli, the firm's Energy & Cleantech Industry Group co-leader, and has published state-by-state compliance guidance covering energy, water, zoning, and tax rules. Davis Wright Tremaine has released its own guide addressing the spread of moratoriums and AI infrastructure rules across US jurisdictions, advising both developers and local governments. Taft launched a dedicated Data Centers, Digital Infrastructure, and Powering AI group in February 2026, and Norton Rose Fulbright has said it built its group to bring together lawyers with experience across the full range of digital infrastructure work. At Bracewell, attorney Jared Berg described the scope of the work as covering “all elements of the value chain of the data center,” from real estate and construction contracts to power generation agreements.

The fastest-growing part of this legal market may be procedural litigation. According to Bloomberg Law, opponents in multiple states have shifted from challenging data centers on environmental or community-harm grounds to challenging the approval process itself. In Prince William County, Virginia, a court found officials failed to properly notify the public before a Board of Supervisors meeting that approved more than 1,500 acres for data center use; Blackstone's QTS and Compass Datacenters both abandoned the project that followed. In Wilmington, Ohio, a federal judge ruled that officials violated public meetings requirements in approving a proposed $4 billion Amazon Web Services complex and barred further action by the planning commission. Similar procedural challenges have succeeded or forced reversals in Festus, Missouri; Hill County, Texas; Stokes County, North Carolina; and Doña Ana County, New Mexico, where advocacy groups challenged a $165 billion Oracle and OpenAI campus known as Project Jupiter over closed-door sessions and withheld public records. One attorney told Bloomberg Law the approach reflects a deliberate strategic shift: “The law often favors the discretion of government officials, so picking apart the process is a stronger strategy.”

The pattern is playing out in real time in individual disputes. In Gibraltar, Michigan, developers Raeden Acquisitions and Cal Realty filed suit in August 2026 after the city enacted a one-year moratorium blocking their proposed 100 megawatt data center on a former steel plant site, seeking court approval of their site plan and damages. In Lowell, Massachusetts, the Conservation Law Foundation's environmental justice program is representing residents seeking a preliminary injunction against a data center operator's expansion plans, including additional diesel generators and cooling towers. And in Inver Grove Heights, Minnesota, attorney Jacob W. Steen of Larkin Hoffman sent city officials a letter on behalf of developer QLevr LLC warning of potential litigation if the city moved forward with a proposed moratorium, a preemptive tactic that did not stop the city council from approving a pause anyway.

For an industry accustomed to treating permitting as a formality, the legal bills are becoming a real line item. Developers are retaining specialized counsel earlier in the site selection process, and opposition groups are finding that procedural challenges succeed more often than substantive ones. As long as new moratoriums keep appearing, and 2026 has already produced dozens of them, both sides of this fight appear likely to keep hiring.

States in this story

Massachusetts (2)Michigan (2)
A Data Center Digest Original Story · 3 months ago