Big Tech is betting on Pennsylvania. Can its grid keep up?

Pennsylvania has become the front line of the Big Tech industry’s artificial intelligence (AI) data center boom. State and federal officials initially championed the multibillion-dollar investment as an economic boost; however, that enthusiasm has since collided with mounting regulatory and local opposition.

The Keystone State as a Magnet for AI

So what makes Pennsylvania ground zero for this data center surge? The answer is simple: The state is the second-largest energy producer in the nation, with vast natural gas reserves, available rural land and abundant water resources. A June 2026 analysis from Buchanan Ingersoll & Rooney described Pennsylvania as the “industrial backbone powering AI and cloud growth across the PJM region,” citing the state’s reliable, low-cost power, ongoing grid modernization and manufacturing as key advantages for development.

The Pittsburgh Technology Council reached a similar conclusion in a March 2026 report, finding Pennsylvania uniquely positioned to benefit from the AI data center buildout by leveraging what researchers call a “three-legged stool” of opportunity, consisting of advanced manufacturing, energy generation and distribution and data center development itself. Other states typically compete in only one of those areas; however, Pennsylvania’s asset, the report concluded, is its ability to excel in all three simultaneously, having already established itself as a leader in two.

Manufacturing is where that strength is most visible. Pennsylvania produces many of the physical components that power the cloud, including electrical transformers, switchgear, grid-scale batteries and structural steel. In 2024 alone, the state exported more than $14 billion in data center-related manufacturing and raw materials to the 13 states within the PJM interconnection, supporting nearly 2,000 jobs that are expected to grow by 4,500 more by 2036. Grid reliability is the second piece of that equation. Pennsylvania remains the nation’s top electricity exporter, sending roughly a third of its total generation to other states in 2024, while also maintaining the fourth-lowest commercial electricity rates among PJM states. Data center development, meanwhile, may hold the steepest growth curve. Capacity across the state is projected to surge from 182 megawatts to more than 7,300 megawatts by 2036, fueling an estimated $12 billion in annual economic output and 19,400 jobs across construction, operations and the broader economy. 

Randy Vulakovich of Buchanan Ingersoll & Rooney, a lead sponsor of the Pittsburgh study, said the state’s upper hand lies in its competitiveness. “When people talk about where America’s AI economy lives, they’ll say Pennsylvania because we built the entire infrastructure of the future,” he said. Yet building “the infrastructure of the future” comes with a cost, and communities near proposed and existing data center sites are asking who bears it. Residents in townships across the region have voiced concerns over rising electricity costs, noise, strained water supply and the loss of rural land to industrial-scale construction, arguing that the same resources touted as economic strengths are being consumed faster than local regulations can keep pace with. 

Power, Pollution and the Loophole Economy

Thomas Schuster, clean energy program director for the Sierra Club’s Pennsylvania Chapter, has tracked energy policy across the state for more than a decade. He said the pace of data center development varies, with some projects moving toward construction faster than others. Not all proposed projects will likely be built, but even a partial wave of simultaneous openings would place significant demands on the region’s power and water grids.

Since data centers operate continuously, power is a primary concern. Developers are now installing dozens or even hundreds of backup generators per site, far exceeding what a typical commercial operation would use. “The cumulative capacity is equal to what the data centers consume, which can be equal to a large power plant in and of itself,” Schuster said. Building a traditional power plant requires years of permitting review. However, installing banks of diesel or gas generators does not face the same regulatory timeline, giving developers an incentive to scale up power capacity faster than conventional infrastructure would allow.

That speed comes with notable emissions trade-offs. Diesel backup generators are categorized into tiers based on their pollution controls, with Tier 1 models largely obsolete and Tier 4 models the newest and cleanest, reducing nitrogen oxide and particulate matter emissions by up to 90% compared with earlier tiers. The Tier 4 requirement, though, applies only to generators in regular use. Generators classified as emergency-only are exempt, under the assumption that their limited runtime makes pollution emissions trivial. That exemption permits unlimited use during actual emergencies, such as outages, fires or floods, plus up to 100 hours for testing and maintenance, some of which counts as nonemergency use without affecting the generator’s status. As a result, Tier 2 units remain the industry baseline among developers, Schuster said, since Tier 4 requirements do not apply to emergency-rated engines.

That regulatory gap exists partly because federal oversight of diesel generators is split across three separate standards under Title 40 of the Code of Federal Regulations, depending on a generator’s age and type. Generators built before 2007 are held to Subpart ZZZZ, the emission standards for hazardous air pollutants for combustion engines. Newer stationary models fall under Subpart IIII, the standard that produced Tier 4 for nonemergency generators, while portable, trailer-mounted units are governed by Part 1039, a separate nonroad engine standard.

Developers describe these generators as emergency equipment that runs infrequently and carries limited impact, but Schuster disputed that characterization. Generators are tested monthly for about an hour, he said, and it remains unclear whether multiple units are tested simultaneously at a given site. He said grid emergencies, which are declared when the reserve power margin drops below a set threshold, are becoming more frequent as data center demand increases.

According to Schuster, regulations have not kept pace with the growing overlap between emergency grid declarations and generator use. Permitted pollution limits are sometimes suspended during declared emergencies, which tend to be the same periods when backup generators are most likely to run. Individual generators are usually classified as “minor sources,” exempt from stricter review, though some facilities operate under synthetic minor permits that cap generator runtime on paper, a limit that Schuster said can be exceeded during an actual emergency. He said generator banks should be regulated collectively, much like a power plant, with a minimum Tier 4 standard and a full review triggered under the Clean Air Act based on a site’s total capacity rather than individual units.

Battery storage is an alternative to relying heavily on backup generators. Schuster said batteries that charge during low electricity demand and discharge during peak hours can eliminate the need for diesel and gas backup, while also generating income for developers who sell stored power back to the grid during periods of high demand. However, a June 2026 Data Center Knowledge report found that adoption of battery storage remains limited, largely because of cost, performance issues and integration complexity involved in deploying battery systems at scale.

Water is the second resource these facilities consume on a mass scale. Data center cooling methods keep servers from overheating and are generally categorized into air cooling, liquid cooling and localized or hybrid systems, with the choice depending on a facility’s density, hardware and climate conditions. Systems that rely more heavily on water tend to consume less electricity, and the reverse holds true, so each method carries its own environmental cost. Water-cooled systems discharge heated water back into a water source at a scale that can affect aquatic ecosystems, Schuster said, which is why large power plants typically use cooling towers that recirculate water rather than releasing it downstream. Developers cite maximum withdrawal estimates that apply only during the hottest days of the year, but Schuster said those are also the periods when river levels are the lowest and least able to absorb the impact.

As regulators weigh how to respond, some engineers are approaching the buildout differently, designing systems meant to reduce Big Tech’s strain before reaching the grid.

Building Smarter, Not Just Bigger

Dr. Wangda Zuo, professor of architectural engineering at Penn State University, focuses his research on building open-source modeling tools for data centers, chiller plants and net-zero energy communities, part of a broader effort he describes as applying physical AI to make buildings and communities more sustainable and resilient. Zuo’s research group, working through a partnership with Lawrence Berkeley National Laboratory and Schneider Electric, developed two separate sets of open-source software, one focused on data centers and the other on cooling facilities. The first simulates airflow inside the data hall to optimize how cold air is distributed and hot air is exhausted, helping facilities avoid the wasted energy of overcooling and the dangerous hotspots that form when equipment receives insufficient airflow. The second targets the cooling facility itself, including the cooling towers, chillers and pumps that operate largely out of sight. 

That second tool, Zuo said, has since been adopted internally by several large manufacturers for their own research and development and was used in pilot projects at facilities in Miami and Massachusetts, where his team created calibrated “digital twins” of each site to test fixes before recommending changes to operators. The projects resulted in a 53% reduction in cooling energy in Miami and a 74% reduction in Massachusetts.

Zuo said water use should be broken down into two distinct terms. On-site, or direct, water is what a facility uses locally, mostly through evaporation to keep servers from overheating. Off-site, or indirect, water is the far larger volume consumed upstream to generate the electricity the center draws, since most power plants also rely on water for cooling or steam. Citing a Lawrence Berkeley National Lab study, Zuo said direct, on-site water use by data centers nationally accounts for roughly half a percent of U.S. industrial water use, while indirect water tied to electricity generation runs about 12 times higher, or roughly 6%. 

As demand grows, operators have begun adjusting how they source both power and water. Zuo said many operators now source cooling water from wastewater treatment plants rather than potable supply, so the water that evaporates was never headed to residential taps to begin with. Others are moving to closed-loop “dry coolers,” which reject heat with fans instead of evaporation to avoid water consumption entirely, a trade Zuo said comes at a cost in cooling efficiency. A related trend popular in Pennsylvania is on-site natural gas turbines that generate power without any water use, similar in principle to a jet engine, though again at lower efficiency than a conventional plant. 

Zuo pointed to a factor driving developers to Pennsylvania specifically. As one of the country's largest net exporters of electricity, the state has an opportunity to generate revenue and create jobs by selling power in-state at higher margins rather than exporting it at lower prices. However, he said that supply does not automatically match the demand. “The challenge is that you have generation, but if you just see how much data centers want, or at least the plan to be built, it is more than the capacity we have right now. If all of them are going to be built, then we need to build more power plants.” 

Zuo’s account fits a broader pattern documented nationally. A December 2025 overview of Berkeley Lab’s data center research found that U.S. data center electricity use nearly tripled between 2016 and 2023, with the lab’s researchers projecting the sector could account for as much as 12% of national electricity consumption by 2028. The lab’s own diagnostic and retrofit programs stand out as a leading example: Targeted upgrades in airflow management and cooling controls at facilities it assessed produced an estimated eight percent reduction in cooling energy use, saving more than one million gallons of water annually. Even with these efficiency gains, Zuo cautions that engineering fixes alone cannot resolve the deeper dissonance between how fast data centers are being proposed and how long it takes to build the infrastructure meant to support them. 

New Rules, Old Grid

On Aug. 18, Gov. Josh Shapiro signed Executive Order 2026-05, aimed at protecting Pennsylvania consumers from data center impacts. The order directs the Department of Environmental Protection to review permit applications only from developers who have made a legally binding commitment to meet the Governor’s Responsible Infrastructure Development, or GRID, Requirements, which set “strict standards for energy affordability, environmental protection, workforce and economic development, transparency, and community engagement.” The order also removes all AI data center proposals from the Fast Track permitting process and prohibits developers from using nondisclosure agreements with the communities they are building near. Under GRID, these companies must cover the full cost of new infrastructure that their projects require and must report their energy and water use to the state annually. 

Most development projects tracked in the state have yet to come online. As of August 2026, the U.S. Data Center Map lists at least 133 data center facilities in Pennsylvania, 93 of which are operational. But those facilities account for only a fraction of the 17,138 megawatts of total capacity the map tracks statewide, with the remaining still under construction, planned or paused.  

Almost all grid-connected data centers in Pennsylvania draw power from the PJM Interconnection grid, the regional operator serving 13 states and Washington, D.C. In its 2027/2028 Base Residual Auction, held in December 2025, PJM secured power largely from fossil fuel plants, with 43% from natural gas and 20% from coal, while wind and solar together made up just 3%. It was the first auction in which PJM fell short of its own reliability target, missing it by more than 6,600 megawatts. PJM’s own analysis pointed to the same driver behind that deficit, citing continued growth in data center demand that has kept exceeding the grid operator’s development plans. In other words, as more data centers come online, they will likely rely on a power grid that still primarily uses gas and coal, rather than the clean energy Pennsylvania has promised. How Pennsylvania resolves the gap between what it has promised and what its grid can deliver will shape the next phase of the state’s AI buildout. 

Pennsylvania remains an attractive destination for the Big Tech industry, and Gov. Shapiro’s stricter GRID Requirements show the state beginning to respond. But whether that response can outpace a buildout worth billions in projected economic output remains an open question. For now, the state must confront whether it can regulate its power and water resources responsibly before rising electricity costs and a strained grid become the new normal.