The patchwork of carbon pricing regimes across the United States and globally has become one of the most consequential – and least discussed – variables in AI data center siting decisions. But the compliance cost differential is only part of the picture. The federal Clean Air Act’s permitting framework, which is triggered primarily by nitrogen oxide emissions rather than carbon dioxide for most projects, creates a parallel set of regulatory considerations that directly affect project timelines, technology selection, and deal structuring. This alert explains both frameworks, identifies key areas of legal uncertainty, and outlines the structuring steps that developers, lenders, and investors should consider.
Introduction
The rapid buildout of AI data center infrastructure has surfaced a regulatory dynamic that is quietly reshaping how developers, investors, and lenders evaluate project sites. A large-scale natural gas power fleet supporting a data center campus – say, 700 megawatts of on-site generation – faces dramatically different compliance costs depending on where it is built. In Texas, which imposes no carbon price, the carbon compliance cost is zero. In California, where the Cap-and-Invest program prices carbon at roughly $29 per ton today and rising, the same fleet would face tens of millions of dollars in annual compliance costs. In the European Union, where Emissions Trading System (ETS) allowances currently trade at approximately $78–79 per ton and are projected to exceed $150 per ton by 2030, the annual compliance burden can reach well into nine figures. Over the 20-year useful life of a generation fleet, these differentials compound into sums that can rival or exceed the capital cost of the generation assets themselves.
That concentration of cost goes a long way toward explaining why hyperscale AI data center investment has gravitated toward Texas and a handful of other states with favorable regulatory frameworks. But the compliance cost differential, striking as it is, tells only part of the story. For the clients we advise – developers selecting sites, general counsel negotiating construction contracts, lenders underwriting project finance, investors evaluating risk – the more important question is: What do we need to structure, permit, and document to capture this advantage, and what legal risks should we be aware of along the way?
This alert addresses those questions.
The Federal Permitting Framework: Why the Conversation Starts with Nitrogen Oxides, Not Carbon
Most observers of the data center space assume that carbon pricing is the primary regulatory variable in siting decisions. It is an understandable assumption, given the attention that carbon policy receives in the press. But it is incomplete – and, for the majority of U.S. projects, it puts the emphasis in the wrong place.
The federal Clean Air Act (CAA) establishes a permitting framework for large industrial sources of air pollution. Two programs are particularly relevant here: Prevention of Significant Deterioration (PSD), which applies to the construction of new major sources of pollution in areas that meet federal air quality standards, and Title V, which imposes ongoing operating permit requirements on major sources. Both programs are triggered when a facility’s emissions of certain regulated pollutants – known as “criteria pollutants” – exceed specified thresholds.
For on-site power generation at data centers, the criteria pollutant that matters most is nitrogen oxides, commonly abbreviated as NOx. NOx is a byproduct of fossil fuel combustion and a precursor to smog and particulate matter. The federal PSD thresholds for NOx are 100 tons per year for combustion turbines and stationary engines (which are specifically listed source categories under the statute) and 250 tons per year for unlisted sources.
Greenhouse gas emissions – including carbon dioxide – are regulated under a separate and narrower framework. The so-called GHG Tailoring Rule sets a Title V threshold at 100,000 tons of carbon dioxide equivalent per year, a figure so high that most data center power configurations do not approach it on greenhouse gas emissions alone. More importantly, the Supreme Court’s 2014 decision in Utility Air Regulatory Group v. EPA established that greenhouse gas Best Available Control Technology (BACT) requirements apply only to facilities that already qualify as major sources under criteria pollutants – the “anyway source” doctrine. In practical terms, this means that if a data center’s on-site generation stays below the NOx major source threshold, there is no federal obligation to undergo greenhouse gas BACT review.
The implication for project planning is significant. Technology selection for on-site power generation is not merely an engineering decision; it is a permitting strategy decision. The choice among generation technologies – reciprocating engines, simple-cycle combustion turbines, combined-cycle gas turbines, solid oxide fuel cells – determines whether a facility’s NOx emissions will exceed the major source threshold and, consequently, whether it will be subject to full PSD review (a process that typically adds 18 to 36 months to the project timeline) or can proceed through a faster, more streamlined permitting pathway.
The numbers illustrate the point. An uncontrolled simple-cycle combustion turbine operating at 100 megawatts of capacity can produce 219 to 438 tons of NOx per year – well above both PSD thresholds. The same turbine, equipped with selective catalytic reduction (SCR), a widely used emissions control technology, drops to 22 to 88 tons per year. A solid oxide fuel cell, which generates electricity through an electrochemical process rather than combustion, produces roughly 1.3 tons of NOx per year at the same capacity – effectively invisible to the PSD framework at any realistic campus scale.
An additional layer of complexity arises from the source aggregation rules. Under the Environmental Protection Agency’s (EPA) adjacency standard, multiple emission units located on contiguous property, under common control, and sharing the same industrial classification code are treated as a single stationary source for purposes of the threshold calculation. For a data center campus with dozens or hundreds of individual generators, the entire fleet’s NOx output is summed. This aggregation calculation – how the campus is designed, how ownership is structured, how phases are sequenced – is a critical variable that must be analyzed before engineering, procurement, and construction contracts are executed.
In December 2025, the EPA launched a dedicated Clean Air Act Resources for Data Centers webpage, that consolidates guidance on emissions thresholds, aggregation rules, and permitting pathways. A companion report from the Congressional Research Service documented the source aggregation framework in detail. Together, these resources confirm that the federal government recognizes data centers as a distinct and significant category of air emissions source – and that the permitting framework will be applied with increasing specificity. Two additional dimensions of this framework warrant particular attention: the treatment of facility modifications and expansions, and the distinction among state-level permitting pathways.
A related but distinct issue arises when a facility that was originally permitted below the major source threshold later expands or modifies its operations in a way that increases emissions. The Clean Air Act’s New Source Review modification rules apply when a physical or operational change at an existing major source results in a significant net emissions increase. For a data center campus that is designed for phased buildout over several years, the modification analysis must be considered at the outset – not only for the initial phase, but for the cumulative impact of subsequent phases. A campus that is permitted as a minor source in Phase 1 but crosses the major source threshold in Phase 2 may trigger full PSD review for the expansion, with significant implications for the project timeline and for contractual commitments made on the basis of the original permitting assumptions. A phased buildout over several years may also trigger the Clean Air Act’s anti-circumvention principle that prevents regulated entities from structuring operations in a way that may cause EPA to view the buildout as an attempt to avoid PSD permitting requirements.
It is also important to distinguish among the permitting pathways available in key jurisdictions. In Texas, TCEQ administers several distinct pathways for air quality permits: emissions registration (available for facilities that meet certain low-emission criteria), standard permits (pre-established permits for facilities that meet prescribed conditions), and case-by-case New Source Review. These pathways involve different timelines, different levels of agency review, and different operational constraints. The choice among them is a function of the generation technology selected, the campus configuration, and the emissions profile – and it should be made in consultation with environmental counsel before engineering and procurement decisions are finalized.
For practitioners, the state-level overlay adds further complexity. States like Colorado, where air quality regulators have been among the most active in the country on emissions from energy operations, impose their own standards that may be more stringent than federal requirements. In nonattainment areas – regions that do not meet federal air quality standards for one or more criteria pollutants – NOx compliance costs alone can add $25 million to $50 million annually. The interplay between federal PSD requirements and state-level standards creates a layered regulatory environment that demands jurisdiction-specific analysis.
The Domestic Carbon Pricing Landscape
Layered on top of the federal permitting framework is a patchwork of state and regional carbon pricing regimes that creates a compliance cost spectrum ranging from zero to nearly $30 per ton within the United States.
Texas imposes no carbon price – no cap-and-trade program, no carbon tax, no emissions fee. The TCEQ holds greenhouse gas permitting authority and applies it only to facilities that are already major sources under criteria pollutants. The TCEQ standard permit pathway, which has been used by several of the largest data center generation deployments in the state, offers a substantially faster route to operation than full New Source Review. For developers who prioritize speed to power, this pathway represents a material competitive advantage.
California operates the Cap-and-Invest program (renamed under 2025 legislation from Cap-and-Trade and extended through 2045), which covers on-site fossil fuel combustion above 25,000 metric tons of carbon dioxide equivalent per year. The current allowance price is approximately $29 per ton, with a statutory price floor that escalates at 5% plus the consumer price index annually. Data centers are not classified as trade-exposed industries and receive no free allocation of allowances – they pay for every ton emitted. For a 500-megawatt combined-cycle gas facility operating at 85% capacity factor, the annual compliance cost at current prices is roughly $43.5 million. At projected escalation rates, that figure approaches $120 million per year by 2040. Over a 20-year project life, the present-value differential between a California siting decision and a Texas siting decision can run into the billions of dollars.
The Regional Greenhouse Gas Initiative (RGGI) covers ten states in the Northeast at a price of $22 to $27 per ton. RGGI applies to fossil-fuel-fired electric power generators with a capacity of 25 megawatts or above. For data center developers, RGGI presents what may be the most interesting open legal question in this space: Does on-site generation that is consumed entirely by the data center – never exported to the grid – qualify as an “electric power plant” under RGGI’s definitions? The question is genuinely unresolved and is state-specific within the RGGI region. A facility generating exclusively for its own consumption has a reasonable legal argument that it falls outside RGGI’s coverage. Developers are currently taking different positions on this question across RGGI jurisdictions, and the answer carries significant financial consequences. For a multi-hundred-megawatt behind-the-meter deployment, the difference between covered and exempt is tens of millions of dollars annually. Any developer treating this question as clearly settled – in either direction – is accepting legal risk that should be identified and evaluated.
We raise this question not to advocate for a particular position but to identify a genuine area of legal uncertainty that developers and their counsel should evaluate on a project-specific and state-specific basis. The resolution of this question will likely vary across RGGI member states and may evolve as regulators respond to the growth of behind-the-meter data center generation.
Washington State operates the Climate Commitment Act at approximately $26 per ton, with the same 25,000-ton applicability threshold as California. Washington’s legislature has been actively considering provisions that would specifically target data center fossil fuel generation, making it the jurisdiction with the highest near-term political risk for behind-the-meter gas deployment in the Pacific Northwest.
Taken together, this patchwork of state and regional programs produces compliance cost differentials that are large enough to reshape project economics. A brief example illustrates the magnitude of the siting decision. Consider a 500-megawatt combined-cycle natural gas facility operating at 85% capacity factor, producing approximately 1.5 million metric tons of carbon dioxide per year. In Texas, the annual carbon compliance cost is zero. In California, at current Cap-and-Invest prices of approximately $29 per ton, the annual cost is roughly $43.5 million. In the European Union, at current ETS prices of approximately $78 per ton, the annual cost is approximately $117 million. Over 20 years, even before accounting for projected price escalation, the cumulative differential between the Texas and California scenarios alone exceeds $800 million in nominal terms. These figures are necessarily approximate – they will vary with actual capacity factors, emission rates, allowance prices, and applicable exemptions – but they convey the order of magnitude that makes regulatory cost modeling a first-order siting variable.
Structuring the Deal to Match the Regulatory Reality
The regulatory landscape creates the opportunity. The deal documentation – the contracts, the permits, the financing agreements – determines whether a particular client captures that opportunity or is exposed to risks that erode it.
Several structuring considerations deserve attention.
Siting criteria should formally incorporate regulatory cost modeling. The 20-year net present value of the carbon and NOx compliance cost differential across candidate sites can exceed $1 billion for large-scale deployments. That figure belongs in the siting analysis alongside land cost, power availability, and fiber connectivity – not in a footnote to the environmental section of the feasibility study.
The NOx aggregation analysis should be completed before construction contracts are executed. Ownership structures, campus layout, phasing strategies, and adjacency determinations all affect whether separate buildings or project phases will be aggregated into a single stationary source for purposes of the emissions threshold calculation. A post-signing determination that previously separate phases aggregate into a single source can fundamentally alter the permitting pathway, adding 18 months or more to the timeline and triggering cascading contractual defaults. This analysis requires counsel experienced in Clean Air Act source determinations who understand how the EPA and state agencies evaluate these questions in practice.
Construction contracts should allocate permitting risk with specificity. The EPC contract should distinguish between permitting delays caused by regulatory determinations (such as a change in the aggregation analysis or an unexpected nonattainment area designation) and delays caused by contractor performance. Business interruption and delay-in-start-up insurance should be structured to match these contractual allocations – a point that requires coordination among deal counsel, environmental counsel, and the insurance placement team.
Lender representations and warranties should address the permitting pathway. Project finance lenders are increasingly requiring borrowers to represent that the facility will not trigger major source thresholds and to covenant that operations will maintain the emissions profile assumed in the permitting strategy. Borrower’s counsel should ensure that the technology specifications, operational parameters, and aggregation analysis support those representations before they are made.
Escalation and optionality provisions warrant careful attention. Carbon prices in high-cost jurisdictions are projected to rise substantially over the next decade. Contracts with 20-year terms should include escalation mechanisms that reflect this trajectory, and the project design should preserve flexibility for future hybrid configurations – incorporating renewables, battery storage, or alternative generation technologies – that can reduce carbon exposure as compliance costs increase.
A related consideration involves the interaction between carbon compliance costs and federal tax incentives. The Inflation Reduction Act’s clean electricity production credit (Section 45Y) and clean electricity investment credit (Section 48E) provide substantial incentives for qualifying generation technologies, including certain renewable and zero-emission sources. For a developer evaluating the total cost of power across jurisdictions, the analysis is not limited to the carbon compliance differential; it also includes the tax credit value of alternative generation technologies that may reduce or eliminate carbon exposure. The interplay among carbon pricing, tax incentives, and generation technology selection is complex, and the optimal configuration will depend on the specific project economics, the applicable jurisdiction, and the developer’s tax position. These are determinations that require coordination among energy counsel, tax counsel, and the project’s financial advisors.
Community engagement should be incorporated into the legal and permitting strategy from the outset. Building at a massive scale – whether in rural Texas counties, suburban corridors in Colorado, or fast-growing communities anywhere in the country – brings economic benefits, but it also invites scrutiny on water use, noise, emissions, and land use. The environmental enforcement and community engagement landscape across multiple states has taught us that regulators and communities alike are paying closer attention to cumulative impacts than ever before. Proactive engagement, transparent environmental data, and well-structured community benefit agreements can meaningfully reduce both litigation risk and permitting timelines. Developers who treat community relations as an afterthought are the ones most likely to encounter delays that cost more than the engagement would have.
Conclusion
The multi-gigawatt commitments flowing into Texas, the rapid deployment of behind-the-meter generation fleets, and the broader migration of capital toward jurisdictions with favorable regulatory frameworks reflect a rational response to a complex regulatory landscape. The opportunity is real – but so is the legal and regulatory complexity involved in capturing it.
The clients who are navigating this landscape most effectively are the ones who bring energy and environmental counsel into the siting conversation at the beginning of the process – not after the letter of intent is signed and the permitting questions have already become expensive surprises.
This alert is intended to provide a general overview of the regulatory and structuring considerations relevant to AI data center siting and on-site power generation. It does not constitute legal advice, and the appropriate approach will depend on the specific facts, jurisdiction, and circumstances applicable to each project.
R.J. Colwell is a senior associate at Davis Graham & Stubbs LLP who advises data center developers, power generation companies, and their investors and lenders on the regulatory, contractual, and permitting dimensions of AI infrastructure projects. R.J. can be reached at rj.colwell@davisgraham.com. Randy Dann is a partner at Davis Graham & Stubbs LLP whose practice focuses on Clean Air Act compliance, air quality enforcement, and environmental regulatory matters for energy and natural resources clients. He serves as vice chair of the American Bar Association’s Air Quality Committee. Randy can be reached at randy.dann@davisgraham.com.