Pennsylvania and Texas Data Center Rules Put AI’s Power Problem on Developers

Pennsylvania and Texas have imposed new data center controls after proposed projects created power demands far beyond ordinary industrial development. The Pennsylvania and Texas data center rules deliver a blunt message to AI infrastructure developers: secure credible power, pay the associated costs, or lose access.

That does not mean every server campus must operate as a completely isolated power island. Pennsylvania requires developers to secure sufficient energy and accept the costs tied to generation and grid infrastructure. Texas now requires projects to document on-site generation, expected demand, public assistance, and measures that reduce dependence on its grid.

The distinction matters. “Bring your own power” sounds simple, but electricity does not behave like equipment delivered to a construction site. New generation requires fuel, transmission, environmental permits, financing, and years of development. A private plant can also create pollution and local costs outside the electric grid.

The policy shift reveals a deeper reversal in the AI infrastructure race. States once competed to attract enormous computing campuses with tax incentives and faster permitting. Now two major development markets are demanding evidence that those campuses will not transfer their energy risks to households and existing businesses.

Developers therefore face a new competition. The winning project will not simply secure the most chips, land, or investment. It must assemble a credible package of power generation, grid access, water management, community support, and financial responsibility.

Pennsylvania Turned Voluntary Standards Into a Permit Gate

Pennsylvania has converted responsible development principles into a practical test for whether a proposed AI campus deserves state review.

Governor Josh Shapiro signed Executive Order 2026-05 on August 18, 2026. The order implements his Responsible Infrastructure Development requirements, known as GRID, and changes how Pennsylvania processes data center permits.

The order followed a surge of proposed projects across the state. Pennsylvania says it knows of at least 100 proposals, although most had not meaningfully entered the permitting process. The state’s Department of Environmental Protection had received applications connected to 20 proposed facilities when the order took effect.

That gap between announcements and permit activity is central to the policy. A speculative proposal can reserve land, seek utility service, and influence regional forecasts without having a committed customer or adequate financing. Enough uncertain proposals can make future electricity demand appear larger and more immediate than it really is.

Shapiro said many projects lacked detailed plans, financing, electricity access, or agreements with technology companies that would occupy the facilities. His administration is responding by making developers establish credibility earlier.

Developers that commit to GRID must notify the state, submit detailed plans for review, and sign an enforceable consent order. That agreement can include penalties if the developer fails to satisfy its commitments.

The executive order also removes data centers from Pennsylvania’s fast-track permitting program. Future projects cannot use that program to accelerate state approvals.

Developers must secure the energy needed for their facilities and pay the costs associated with supplying it. Those costs can include new generation, transmission, distribution, and related infrastructure. The purpose is to stop projects from shifting expenses onto other electricity customers.

The requirements also call for an increasing clean-energy contribution. Pennsylvania has identified options including solar energy, advanced nuclear generation, and battery storage. This provision makes the policy more than a simple instruction to install gas turbines beside a server building.

Local approval now carries more weight. A project must comply with municipal plans and obtain required local approvals before receiving final state permits. Noncompliant developers face an even slower sequence because state environmental review will wait until local requirements are complete.

Pennsylvania agencies under the governor’s authority also cannot sign nondisclosure agreements for these projects. That addresses a recurring complaint from communities that receive little information about a campus’s owner, customer, power source, or environmental footprint.

The administration’s GRID requirements cover community engagement, workforce development, water conservation, and environmental protection. Developers that refuse them can lose favorable permit treatment and eligibility for an existing sales-tax exemption on qualifying equipment.

This is not a complete construction ban. A developer can choose the noncompliant review path, although that route introduces more delay and uncertainty. The mechanism uses permitting order and state discretion to reward projects that accept enforceable obligations.

That structure makes the policy more consequential than a voluntary pledge. It ties AI data center power requirements to approvals that developers need before construction can proceed.

The policy also changes negotiating leverage. Municipal governments can demand clearer answers before a developer assumes that state review will advance in parallel. A local dispute can now become a direct obstacle to the project’s wider permit schedule.

Pennsylvania’s action therefore targets two risks at once. It screens speculative proposals while moving infrastructure costs toward developers with serious customers and viable power plans. That creates a higher entry bar, but it does not guarantee that every accepted power plan will benefit consumers.

Texas Is Auditing a Grid Queue Larger Than Its Power System

Texas is treating the data center pipeline as an information and reliability problem before treating it as economic development.

Governor Greg Abbott ordered a comprehensive audit on August 3, 2026. The Public Utility Commission of Texas and the Electric Reliability Council of Texas, or ERCOT, must review projects before they advance through the interconnection process.

An interconnection process evaluates whether and how a new customer or generator can connect safely to the electricity network. It can identify required substations, transmission upgrades, operational limits, and other costs.

The scale behind the audit is striking. According to the governor’s order, ERCOT was considering more than 474 gigawatts of connection requests. That total exceeded five times the system’s record peak demand, and data centers represented approximately 90 percent of the requested load.

A gigawatt measures one billion watts of electrical power. For perspective, the requested capacity was not a forecast of projects certain to operate simultaneously. It represented a queue containing plans at very different stages of maturity.

That distinction does not make the queue harmless. ERCOT and utilities still need reliable information to plan generation and transmission. Duplicate, incomplete, or speculative requests can distort where infrastructure appears necessary.

Texas now wants each project to disclose annual and peak electricity consumption. Developers must explain whether they will construct or procure on-site generation and identify the technology involved. They must also describe other measures that reduce demand on ERCOT.

Under the Texas audit, projects must disclose ownership and controlling interests. They must identify public incentives, grants, abatements, and other financial assistance received or expected.

Water has become part of the same review. Developers must estimate annual and peak consumption, identify water sources, and describe cooling technology. The state specifically asks whether facilities will use air cooling, closed-loop systems, or another water-efficient design.

The audit also examines noise, lighting, traffic, setbacks, and emergency coordination. These issues can determine whether a technically feasible campus becomes politically acceptable to neighboring communities.

A project that refuses the audit can be denied connection to the Texas grid. Abbott’s directive says the review must finish before a project moves forward, creating what functions as a pause for unaudited proposals.

However, Texas is not rejecting data centers as a category. The state is asking developers to prove that their requests represent real projects with workable infrastructure.

Later commitments from CoreWeave, PowerHouse Data Centers, and Emergent Data Centers illustrate that distinction. The governor’s office said those developers agreed to comply with the standards. It also said another unnamed project ended before construction because it could not comply.

The Texas standards call on projects to pay for electric infrastructure, reuse water, disclose public assistance, and limit neighborhood impacts. They also require plans for on-site generation or other ways to reduce grid dependence.

That language does not necessarily require complete electrical isolation. A facility can use on-site generation while retaining a grid connection for backup, balancing, or part of its normal supply. Conversely, a project with generation beside it can still depend on shared gas pipelines and transmission infrastructure.

The practical question is how ERCOT will distinguish a firm project from a placeholder. Financing, equipment orders, land control, signed customer contracts, and realistic construction schedules can help. Public statements alone provide little certainty.

The Pennsylvania and Texas data center rules share this emphasis on credibility. Both states want developers to establish who controls a project, where its power will come from, and who will absorb the resulting costs.

Pennsylvania and Texas Data Center Rules Reverse the Old Bargain

The central conflict is no longer growth versus regulation; it is developer-funded infrastructure versus costs spread across the public grid.

For years, the standard data center pitch rested on speed. Developers offered investment, construction work, and a larger technology footprint. States and municipalities responded with tax incentives, expedited reviews, and access to infrastructure built for broader public demand.

AI altered the scale of that bargain. Training and operating large models require dense clusters of specialized processors. Those clusters concentrate electricity demand in places where transmission and generation were not designed for sudden industrial loads.

A large project can also arrive faster than a conventional power plant. Server buildings and computing equipment may be developed within a few years. Major generation and transmission projects often face longer planning, permitting, procurement, and construction schedules.

That mismatch creates pressure on utilities. They must consider upgrades before every proposed campus becomes operational. If the project disappears, other customers can remain exposed to infrastructure costs unless contracts allocate those risks carefully.

Pennsylvania’s approach places the obligation at the permit stage. Texas places it within an audit of grid connection requests. Their legal mechanisms differ, but their political message is similar.

Developers must show that growth will not increase household bills or undermine reliability. They must also demonstrate that public incentives are justified by projects capable of reaching construction.

This is the real meaning behind demands that data centers build their own power. The state wants a new source of supply attached financially to the new source of demand. Whether the facilities share physical grid infrastructure is a secondary question.

The approach has support beyond state governments. The White House and 13 governors previously urged PJM Interconnection to arrange long-term power contracts funded by technology companies. PJM operates the regional grid covering Pennsylvania and portions of 12 other states plus Washington, D.C.

Under that proposal, technology companies would bid for contracts supporting new power plants. Data centers without dedicated supply or emergency curtailment commitments would bear the related capacity costs.

PJM has separately considered curtailment, which means reducing a customer’s electricity use during system emergencies. Data centers often possess backup generators, but those systems were generally designed for outages rather than routine grid support.

The regional power proposal reflects a broader change in public policy. Governments still want AI investment, but they increasingly reject arrangements that socialize its infrastructure costs.

Industry groups have said technology companies are willing to pay a fair share and invest in grid modernization. The hard part is defining that share before utilities know which proposed projects will materialize.

A developer-funded power plant offers one answer. A long-term supply contract offers another. Flexible computing loads, batteries, and agreed curtailment can also reduce pressure during the grid’s most constrained hours.

These paths carry different risks. A dedicated plant can improve certainty but lock a campus into fuel and operating costs. A grid contract can fund shared generation but depends on market rules and timely construction. Curtailment can support reliability but conflicts with expectations of uninterrupted computing.

The change also pressures cloud providers and AI companies, even when a separate developer owns the campus. A landlord cannot finance generation without confidence that a creditworthy tenant will pay for long-term capacity.

Power obligations will therefore influence which AI projects survive. Campuses backed by committed customers and strong balance sheets can fund new generation more easily. Speculative developments may struggle before they reach environmental review.

That produces a less visible form of industry consolidation. Large technology companies gain an advantage because they can sign long contracts, finance energy assets, and spread costs across cloud customers. Smaller AI firms become more dependent on infrastructure controlled by those companies.

The old bargain treated electricity as an available input after a site was chosen. The new bargain makes power procurement part of the product. A data center proposal is now also an energy-development proposal.

Building Private Power Does Not Make the Costs Disappear

Moving generation behind the meter can protect the grid, but it can also exchange shared financial risks for local environmental ones.

Behind-the-meter generation sits on the customer’s side of the utility meter and directly supplies a facility. It can reduce the amount of electricity that a data center draws from the wider network.

Developers increasingly pair computing campuses with natural-gas turbines because those plants can provide steady output. Batteries can respond quickly but store electricity rather than create it. Solar and wind generation need storage, flexible demand, or another supply source when weather conditions change.

Recent projects show the scale of this approach. In Abilene, Texas, Crusoe announced two additional AI data center buildings for Microsoft and an attached 900-megawatt power plant. The wider site was expected to reach 10 buildings and 2.1 gigawatts of computing capacity.

The existing OpenAI and Oracle development at the location has a 350-megawatt gas plant. Oracle previously described that facility as backup generation because the campus primarily receives power through the regional grid.

These details illustrate why “off-grid” can become an imprecise label. A campus may own generation while using grid electricity. It may also keep a connection for startup power, maintenance periods, emergencies, or economic dispatch.

The Abilene expansion also exposes the emissions tradeoff. On-site gas can help a project avoid waiting for new transmission. Burning that gas still produces carbon dioxide and local air pollutants.

Private plants require pipelines, water arrangements, emissions permits, and reliable fuel delivery. Their effects remain public even when their electrical output does not enter the shared grid.

Pennsylvania tries to address that problem through environmental standards and a growing clean-energy requirement. Its GRID framework also invokes the state constitution’s environmental protections. Yet implementation will determine how meaningful those commitments become.

One uncertainty is what counts as bringing power. A developer might construct generation on the same property, finance a plant elsewhere, or sign a long-term contract with an existing supplier. Each arrangement has a different effect on total regional supply.

Buying output from an existing plant does not necessarily add electricity. It can simply redirect power from other customers. A policy focused on additional generation must distinguish new supply from the relabeling of existing capacity.

Transmission remains another constraint. A newly built plant can sit far from the data center, but its output still needs a path through the network. Paying for generation without completing transmission upgrades does not guarantee deliverable electricity.

Reliability creates a further complication. An isolated campus must maintain enough generation for equipment failures, fuel interruptions, and maintenance. A grid-connected campus can share reserves with other customers, which can reduce total backup requirements.

Complete isolation is therefore not automatically the most efficient outcome. The stronger principle is cost causation: the customer creating a major new requirement should fund the generation and infrastructure needed to serve it.

There is also a risk that governors overstate what executive action can accomplish. Pennsylvania’s agencies control important permits, but local decisions, courts, legislatures, utilities, and federal regulators retain authority over different parts of development.

Texas faces a data-quality problem. Its 474-gigawatt queue is far larger than plausible near-term demand, indicating that many entries will not become operating facilities. An audit can remove weak proposals, but it cannot predict every customer decision or financing failure.

The states must avoid two opposite errors. Treating every request as firm can lead to unnecessary infrastructure planning. Treating an enormous queue as meaningless can leave the grid unprepared for projects that do proceed.

Developers also need stable rules. Generation equipment has long lead times, and project financing depends on predictable approval processes. Requirements that change after major commitments can raise costs without improving planning.

These concerns do not invalidate AI data center power requirements. They show why enforcement details matter more than slogans. A credible system needs deposits, milestones, withdrawal penalties, transparent demand forecasts, and contracts assigning upgrade costs.

Public reporting will be important. Pennsylvania is building a map of known projects and permit status. Texas is requiring ownership and resource disclosures. Those records can help communities separate active construction from promotional announcements.

The environmental balance deserves equal scrutiny. If data centers build their own power primarily through gas, grid pressure can fall while emissions rise. If clean projects face slower permitting than gas plants, the policy can undermine its stated goals.

The Pennsylvania and Texas data center rules should therefore be judged through measurable outcomes. Consumer bills, verified new generation, local pollution, water withdrawals, and completed projects matter more than the number of applications rejected.

Three Signals Will Show Whether the New Model Works

The next test is whether stronger screening produces credible infrastructure or merely moves projects and pollution elsewhere.

The first signal is the number of proposals that survive each audit and permit gate. Pennsylvania says more than 100 facilities have been proposed, while only a fraction entered formal state review. Texas is examining a queue representing more than 474 gigawatts.

A steep reduction would support the states’ argument that speculative demand has distorted planning. However, the remaining projects must still disclose committed customers, financing, construction milestones, and energy arrangements.

A smaller queue is not automatically a successful queue. Regulators need to show that surviving projects advance, build their promised infrastructure, and meet enforceable deadlines.

The second signal is how much genuinely additional generation reaches operation. Announcements and power-purchase agreements do not supply a server campus until plants, fuel systems, and transmission connections work.

Watch whether developers order equipment, obtain environmental permits, secure fuel, and begin construction. Those steps provide stronger evidence than projected capacity in a press release.

The mix of generation will matter too. New gas plants can deliver steady power sooner than some alternatives, but they create long-term emissions exposure. Clean generation paired with storage or flexible computing offers a different cost and reliability profile.

Pennsylvania’s clean-energy requirement provides a clear policy test. Officials should report how the required share changes over time and whether it produces new supply rather than accounting transfers.

Texas should disclose how much on-site generation audited developers actually complete. It should also show whether those facilities reduce peak grid demand or retain substantial dependence on ERCOT.

The third signal is the effect on customers and communities. Both governors have framed their actions as protections against higher electricity costs, water stress, noise, and weak local control.

That promise requires public measurement. Regulators should track infrastructure charges assigned to developers, changes in peak load, water commitments, emissions, and the use of public incentives.

Electricity prices will not provide a simple verdict. Fuel costs, weather, transmission constraints, retirements, and wider demand growth can move bills independently of data centers. Cost-allocation decisions will reveal more about whether developers paid their share.

Community outcomes also need evidence. A benefits agreement has limited value if residents cannot inspect it or enforce its commitments. Public permit records and reporting requirements make those promises easier to evaluate.

Developers will watch another outcome: approval speed. A clear but demanding process can be easier to finance than an informal process filled with political uncertainty. Projects that meet published standards should receive decisions on predictable schedules.

The approach will strengthen if compliant projects move forward while weak proposals exit early. It will weaken if rules remain vague, enforcement differs by project, or promised generation never enters service.

Other states will follow these results. Pennsylvania operates inside the large PJM market, while Texas relies mainly on its own ERCOT system. Similar policies succeeding in both settings would offer a model that travels across different grid structures.

AI companies and enterprise buyers should care because infrastructure constraints eventually affect computing availability. Delayed campuses can tighten cloud capacity, change deployment locations, and increase the value of efficient models and workloads.

Software teams cannot control a state’s grid, but they can track where providers are building credible capacity. Procurement reviews should now include energy delivery, curtailment exposure, construction status, and the environmental consequences of dedicated generation.

The final question is not whether data centers build their own power in a literal sense. It is whether the companies driving new demand accept the financial, operational, and environmental obligations that come with it.

The Pennsylvania and Texas data center rules have established that expectation. Now developers must prove they can turn power plans into functioning infrastructure without transferring hidden costs to everyone else. Watch the permit records, completed generation, and customer bills. Those outcomes will show whether “bring your own power” became a durable policy or remained a memorable political warning.