Demand resulting from competition to establish market share in AI and related new industries and policy imperatives is clashing with supply friction caused by regulatory and industrial bottlenecks. Existing legal and business challenges inherent to developing energy infrastructure remain: regulatory complexity and differences among states and US regions regarding permission to generate and deliver energy; a need for return of and on significant capital outlays; and support for customers’ and governments’ energy transition objectives.
The market is reaching for tools old and new to reconcile these concerns and create reliable commercial and financing arrangements that accommodate the various transactional constituents.
Background
In the US power market a few years ago, significant load growth was limited to areas experiencing population and industrial expansion, the latter driven in part by policies such as energy transition incentives in the Infrastructure Investment and Jobs Act and the Inflation Reduction Act. Much of the investment in infrastructure to meet this load dovetailed with the buildout of renewable energy and storage systems and reductions in capacity of older fossil-based stations for which necessary retrofits and other upgrades were not feasible, for various reasons. Reduced availability of dispatchable resources raised fewer concerns than today, under the expectation that new energy transition technologies to augment the renewables buildout would be deployed in time to meet US macroeconomic needs.
Generating facilities co-located with contracted industrial and other power purchasers have not been rare in the US market. Such facilities have typically connected both to their customers and the larger electric grid. This has enabled both generators and users to rely upon pricing and other agreed terms while also benefiting from the availability of wholesale power markets and derisking bilateral exposure.
Historically these facilities were installed near users with a need for superior service to that provided by the electric grid, including firmer power and delivery of other commodities such as steam. Examples have included refineries, factories, hospitals and universities. Structuring these projects has always involved consideration of the relevant state utility laws and the limited permissibility of direct sales of electricity not involving a regulated utility.
Markets have developed structures over time to accommodate these restrictions on a limited basis. Data centres and similar users, such as crypto miners, have more recently joined this cohort of users, having evolved from earlier power contracting arrangements that were frequently financially settled, did not require delivery of or proximity to generated electrons and enabled users to access useful environmental attributes.
Both generator and load interconnection procedures have been available throughout the US to enable both parties in these relationships to connect to the grid and deliver, or receive, power for which transmission, distribution and other services are provided by utilities and other third parties. Depending upon their location in the US, generators often ran headlong into interconnection queue processes that were clogged with an excess of projects and imposed years of development delays, combined with studies, collateral posting sand notification to the public of projects.
Load interconnection procedures generally imposed less onerous requirements on power users in past years. Special tariff schedules have been made available by utilities to qualifying customers, with standard grid service often available without significant delay in most locations. Importantly, this type of grid service has not historically provided its users with high reliability guarantees, which is part and parcel of the legal environment in which utilities typically operate and their customarily limited exposure to negligence claims.
Current LLM and load growth era
The arrival of large language models and the AI era, together with other sources of load growth including reindustrialisation and electrification, has made the need for change obvious to most. The US is in something of a Sputnik moment, with high stakes for long-term geopolitical influence tied to the buildout of data centre infrastructure and advancement of AI in the US economy and among partner nations. The possibility of immense gains is driving competition among industrial heavyweights within the US toward broadly aligned objectives. Commensurately, US policymakers, businesses and financiers have put their efforts behind rapid capacity expansion in generation and transmission and related industrial supply chains.
A number of changes in the market for data centre power are resulting from this evolving picture. First, capital is moving geographically, with data centres being developed in newer US markets as long as the key ingredients of land, power, water and a supportive regulatory and community environment are – or are reasonably likely to become – available. States with the greatest historical lead in these projects, including Virginia and Texas, have experienced new political friction with increased public awareness of project NIMBY effects and effects on local power and water availability and cost.
Utilities, their regulators and their customer bases are more sensitive to the possibility of large load users and developers not paying their fair share and passing on the costs of transmission and generation buildout to ratepayers. The aperture for project developers has thus greatly expanded, with states and other participants competing to attract game-changing investments.
Second, urgency caused by users’ need to increase computing power as soon as possible is driving shorter development timeframes and affecting the way participants in all parts of these transactions view their risks. Fully islanded, co-located power projects have become more theoretically attractive, on the assumption that cutting out interconnection waits solves a major development challenge.
Supply chains are being asked to adapt quickly, having adjusted to baseline order book expectations that no longer align with current demand. Large data centre projects are pulling in any kind of generating asset they can find, and the cohort of OEMs competing in the market has expanded. Of particular value are reservations for large gas turbines, with new requests typically being assigned wait times of several years.
Development contracting has become in significant part an exercise in preserving and managing these reservations and assigning responsibility among power and data centre developers for payment and termination liabilities. In tandem, ubiquitous preference for low emissions power technology has given way to a textured and not outspoken degree of variance among users, with some but not all willing to assign lower priority to this technology for specific projects or more broadly. There remains a sizeable market for transition technologies, though, and purveyors of these new technologies continue to see significant scaling opportunity presented by the premium that these users are in many cases willing to pay for energy with premium characteristics.
Third, the energy regulatory background is evolving to accommodate the sharp increase in energy demand due to data centres. Over the last decade, the grid has experienced significant delays in generator interconnections caused by the influx of small renewable generators and increased electricity demand. As a result of these delays, the Department of Energy and Federal Energy Regulatory Commission have been forced to reevaluate the rules around co-locating and quickly interconnecting generation and load. In October 2025,Department of Energy Secretary Chris Wright instructed FERC to consider and act on a proposed rulemaking designed to expeditiously interconnect large loads, with particular attention on data centres. FERC recently announced that it will be acting on the proposed rule at the end of June 2026.
Soon after the Department of Energy published its proposed rulemaking, FERC took a large step toward clarifying its position on large load co-location when it issued an order requiring PJM Interconnection LLC to facilitate co-located load development. Specifically, FERC ordered PJM to define “co-located load” to make clear that the co-located generator is physically connected to and synchronised with the transmission system, not isolated. FERC emphasised that states would retain exclusive jurisdiction over retail power transactions while it would oversee generator interconnections involving FERC-jurisdictional facilities, including for generators attached to co-located large loads.
Utilities throughout the country have also taken unilateral action to revise their rules to address large loads. For example, Southwest Power Pool Inc, PJM, and Midcontinent Independent System Operator Inc recently enacted, and FERC approved, one-time expedited generation interconnection processes meant to allow qualifying utilities to bypass the traditional and clogged interconnection queue and bring new, large generators online quickly to reduce system pressure.
FERC has also recently accepted several pathbreaking innovations from SPP. The combined High Impact Large Load and High Impact Large Load Generation Assessment study processes pair large loads to one or more new generating resources in an expedited interconnection process designed to bring load and supporting generation onto the grid simultaneously. SPP’s conditional large load filing, which builds off of those processes and will expedite the interconnection of conditional large loads – ie, large loads that agree to receive conditional, rather than firm, electric service – is still pending. Most recently, FERC accepted SPP’s consolidated planning process that combines SPP’s generation interconnection process and transmission planning process into a single, integrated workflow.
FERC commissioner David Rosner called the consolidated planning process a “revolution” for its ability to “allow SPP to more proactively and holistically identify transmission facilities that address the needs of both load and interconnection customers, provide greater upfront cost certainty to interconnection customers, and reduce interconnection study timelines.”
State and local policies, and utility requirements, are also changing. Utilities may be more sensitive to the possibility of being excluded from data centre service than was historically the case with large industrial loads. This is unsurprising, given the larger demand of these facilities – in some cases more than 1GW individually – and the possibility that overbroad exclusions from public utility regulation could be utilised to prevent utilities from fully sharing in this generational growth opportunity.
Class-specific utility tariffs are being designed to allocate price and stranded energy infrastructure asset risks to data centre developers rather than utilities, including through mechanisms such as demand ratchets, collateral postings and minimum tenors. Data centre moratoria and other restrictions are being proposed by some governmental authorities in response to public dissatisfaction with the NIMBY effects of these projects, adding uncertainty to development processes.
PPAs and similar contracts
Current negotiations for data centre power contracts have differed significantly compared with those for other large industrial loads and with the data centres of a few years ago. Parties rightly focus on several of the same key issues that appear across both physical PPAs and financially settled PPAs – on-time commissioning, predictable pricing, operations period performance guarantees, termination remedies that incentivise appropriate conduct, placement of regulatory risk on the power provider and the availability and usability of environmental attributes. However, there are key distinctions.
First, it is challenging to predict how much of the forecasted data centre buildout, and accompanying power buildout, will move from the planning stage through to completion. Deal documents in this market tend to include hedging mechanisms allowing the parties to use offramps that are accompanied by make-whole payments. These offramps may appear as general termination for convenience rights or may be tied to identifiable project risks – the receipt of a permit or the achievement of certainty on a change in state or local permitting requirements, for example.
Remedies in these contexts may be intertwined with third-party relationships, especially involving power equipment reservations. If turbines or other equipment are reserved in the name of one of the parties, assignments of purchase and related rights may be implicated, as may the payment of breakage and other costs that are not able to be cancelled. The parties also may negotiate for mitigation efforts that include reuse of reserved power equipment. Exclusivity and other restrictions also tend to accompany these offramps, to deter their use and ensure that the parties can each recover potential lost opportunities.
Second, data centre operators are likely to owe vast liquidated damages to their users for late commissioning or for failures to satisfy agreed service levels, which can be unforgiving and exceed 99.9%, whether measured annually, monthly or more frequently. The liquidated damages sizing at this downstream phase of the data centre use chain is generally accepted to be multiple times larger than the sizing between the power facility and the data centre in the PPA.
Power providers are not well suited to take a 1:1 or close to that ratio passthrough of this kind of liability and ought not be expected to do so, due to the fundamentally different risk-to-reward ratio of their businesses and smaller capital expenditure. Nonetheless, data centre owners are right to insist that power providers be incentivised to perform within an appropriate risk band. Novel insurance products are evolving to help fill this gap, but these are early days. The ability of power providers to consider data centres’ backup power(e.g., diesel units and on-site battery energy storage systems) or temporary power units in assessing their own compliance liabilities is sometimes a part of these negotiations.
A variation on this issue involves the construction schedule. Time lost by data centres without power when users are ready to pay for compute is crucial. Commissioning ramp schedules tend to be more granular than in other power purchase agreements, with phased commissioning permitted so that data centre users may obtain power and begin operations at the earliest possible date even as other data halls continue to be fitted for use. Data centre owners tend to place emphasis on early notification and required mitigation plans when phases of a project are likelier to fall behind schedule.
Third, in fully islanded projects, neither the data centre nor the power project may be viable without the other. Grid interconnections may be useful to derisk the exposure these assets have to one another, including as a means to reduce liquidated damages and allow the parties to mitigate obligations with third party purchases and sales, but may not be practical given the changing environment.
Each party will thus tend to focus on the possibility of total loss due to the other party’s failure to comply. From the power provider’s perspective, this argues in favour of reimbursement of unpaid capital costs. From the data centre owner’s perspective, this argues in favour of ensuring rights to take ownership of the power project, including through contractual purchase options and mechanisms to protect against bankruptcy and similar risks (eg, security interests and integrated prime and sublease arrangements).
Fourth, the power generated in these deals is a premium product – ie, rather than standard grid power available at a standard grid price, it is power using urgently provided equipment delivered at large scale with custom specifications. The price may be further increased by high performance emissions technology or shared infrastructure buildouts. Capital expenditures required to deliver on these projects may be significantly higher than in other deals.
Elevated pricing, complete co-dependency among counterparties and credit risk create a feedback loop. Each party will need to carefully conduct diligence of its counterparty’s wherewithal to fund power payments or liquidated damages, respectively. The problem may be exacerbated in situations where third parties are involved, such as utilities who are placed in a sleeve position for regulatory reasons or adjoining complementary projects, such as for carbon capture and sequestration. Parties may find it appropriate to use lockbox structures or multiparty agreements in these situations.
Finally, financing structures for these projects are evolving before our eyes. Well-established bankability characteristics for power projects looked to senior project finance, tax equity and transferability and back-leverage debt structures premised upon the standalone financial profile of the power facility. Today, new players are providing financing for these projects in less tested ways. Data centre owners are reportedly providing loans or upfront payments to power providers. Financiers are beginning to consider lending to combined project vehicles, including data centre-plus-power or power-plus-carbon-capture, rather than to each individual project (although the latter remains more prevalent and is likely to continue to be). Capital markets are increasingly providing avenues for large debt financings.
Conclusion
We are likely in the early stages of a boom in energy infrastructure investment in the US, with new assets being deployed and put to specific uses. In the long term, as projects are built and energy demand associated with data centres, reindustrialisation, electrification and other drivers becomes clearer, many of these assets will probably be put to use for a broader variety of purposes than are currently targeted.
This should result in a more robust energy system, including improved transmission and distribution infrastructure and more finely tuned federal and state regulations. The pace of change described in this article may slow, and associated pressure on energy infrastructure deal negotiations may abate. Would-be participants in this market are likely to confront these evolving issues for some time to come, however, and preparedness will be a differentiator.
Originally published on May 6, 2026 online with Project Finance International (PFI). Reproduced with permission. Further duplication without permission is prohibited. All rights reserved.
