Can behind-the-meter structures solve the impending power bottleneck caused by AI data centres?

David Beckstead

Womble Bond Dickinson, Boston

david.beckstead@wbd-us.com

Introduction

The hottest topic in the US power sector for the past year has been the rapid development of new data centre projects, many of which are being built to assist in the development and rollout of artificial intelligence (AI) applications. Many of these AI data centres will consume power at rates best measured in the 100s of megawatts (MW) or gigawatts (GW) with minimal or no downtime. Moreover, given the decarbonisation commitments made by prominent hyperscalers, much of the new power demand appears likely to come from low or zero carbon generation resources.

Indeed, based on a report issued late in 2024 by Lawrence Berkeley National Laboratory, electricity consumption by data centres is expected to increase from 176 terawatt hours (TWh) in 2023 to 325 – 580 TWh by 2028.[1] Measured in capacity, this would amount to approximately 17 – 46 GW of additional firm capacity coming online within the next five years, just to serve this additional load growth. If these forecasts are accurate, data centres will represent between 6.7 and 12 per cent of total electricity consumption by 2028. While these statistics encompass non-AI related functions performed by data centres, such as cloud computing, AI is seen by many utilities as being the key driver to forecasted demand growth.[2]

With the anticipated influx of AI data centres over the years to come, attention has been given to power supply being delivered in a behind-the-meter configuration. At first glance, this structure is attractive because it would permit the AI data centre boom from placing an undue physical burden on the grid, while preventing cost-shifting to the public at large. This brief article will examine development considerations for the buildout of behind-the-meter generating assets which are co-located with AI data centres. It will then examine recent activity at the federal level and provide a brief overview of state-level considerations that developers must take into account. Although behind-the-meter power supply is likely to be an attractive and viable option for many data centre projects, current regulatory frameworks are likely to preclude this structure from being a panacea.

Development considerations

A behind-the-meter structure for the delivery of power supply to a data centre project may involve existing generating resources, in which case the power plant supplying the data centre will be effectively removed from the network supply stack. This raises a number of concerns relating to resource adequacy, likely increases to electricity tariffs for other ratepayers being supplied by the grid, as well as compensation for development and construction costs that were passed on the ratepayers after the power plant achieved commercial operations. While data centres reaching agreements with the owners of existing generating assets may solve a short-term power supply need, in the long-term, expanded generating capacity will be required to serve the needs of all data centre customers.

A greenfield behind-the-meter power supply structure would effectively result in the parallel project developments, with the data centre and dedicated generating assets being developed and constructed more or less simultaneously. Data centre owners will generally have a number of priorities when developing any particular project, but some of the common themes include expedited delivery on construction, security of dedicated power supply, and decarbonised electricity.

Given these priorities, there is no power source which is obviously preferable to other options. Intermittent renewable energy, such as wind and solar, coupled with onsite storage is green and can be deployed relatively quickly; however, given the electricity demands for many data centres in the pipeline, the geographical footprint of the electricity supply may render some of these projects impractical. Up-and-coming power sources, such as many next-generation geothermal technologies, are attractive but not yet competitive on pricing compared with other available resources. Natural gas-fired power is also attractive, and while emissions from these power plants are significantly lower than those of coal-fired plants, without deploying carbon capture and storage (CCS) technology it would inhibit data centre operators’ decarbonisation ambitions.

Nuclear is a technology that is frequently mentioned as a solution to meeting anticipated data r power demands. While nuclear is low carbon and capable of providing firm capacity, it has some drawbacks. First, the timeline needed to develop nuclear projects is considerably longer than that required for many other technologies, mainly due to the licensing regime under the auspices of the Nuclear Regulatory Commission (NRC). Second, the construction and operation costs for nuclear projects tend to be higher than many other resources; the owner of the generating assets will be more likely to pass these costs along to the data centre in the form of higher tariffs.

A new class of generating assets that rely on nuclear technology has emerged, styled as ‘small modular reactors’ or SMRs. The commercial development of SMRs is still in its infancy around the world, but the key question that regulators are grappling with is whether they should be treated as smaller version of traditional nuclear facilities, or if the licensing and regulatory regime should be altered or more narrowly tailored on order to promote a more rapid development of the technology.

Activity at the federal level

Congress has been active over the past few years, adopting a number of laws which should help in unlocking generating assets necessary to power the impending data centre boom.

Inflation Reduction Act of 2022

The Inflation Reduction Act has been widely seen as one of the most significant pieces of legislation adopted during President Biden’s term in office. The law amended the Internal Revenue Code in order to create new and extend existing tax credits for the generation of low- and zero-carbon electricity. Furthermore, the Act introduced the transferability of investment and production tax credits, thus permitting developers to unlock greater streams of capital necessary for the construction of generating assets.

In late January, the Trump Administration issued a sweeping Executive Order seeking to freeze all federal financial assistance, including clean energy incentives under the Inflation Reduction Act. A number of states already have filed legal challenges to the Executive Order, but the debate may complicate or at least delay efforts to access these benefits.

Accelerating Development of Versatile, Advanced Nuclear for Clean Energy Act (‘ADVANCE Act’)

The ADVANCE Act was enacted in July 2024 with strong bipartisan support, and aims (among other objectives) to simplify the licensing procedures for advanced nuclear technologies. The ADVANCE Act requires the NRC to take a number of actions in order to implement its substantive prescriptions, which should be forthcoming in 2025 and 2026. The NRC has begun a process of public hearings to provide updates on the status of its implementation of the ADVANCE Act, which should continue in 2025.

In addition to positive steps being taken by Congress, the judiciary may also play a role in shaping nuclear policy in future. The States of Texas and Utah have joined with a private developer of SMRs to bring a lawsuit against the NRC at the United States District Court of Eastern Texas.[3] The primary claim being asserted by the plaintiffs is that the NRC has adopted an overly broad interpretation of the statutory language setting out its licensing authority over the construction and operation of nuclear facilities. The lawsuit alleges that the NRC’s interpretation does not align with statutory revisions made by Congress to the Atomic Energy Act in 1954, and it therefore seeks declaratory relief that the NRC’s stringent standards are unlawful in as much as they relate to certain small, non-hazardous reactors. If the claim is successful, it could result in a more permissive environment for the development of SMRs, which have great potential to fuel the AI data centre boom.

State-level considerations

Although the federal government has been active in facilitating the buildout of electricity infrastructure, a behind-the-meter power structure must comply with state laws and regulations relating to retail power trading. Local utilities primarily carry out retail sales of electricity in most states. These utilities are typically subject to the state’s public utilities commission. Electric utilities are usually granted a monopoly to operate within a given geographical area, in exchange for a commitment to serve any load within that area. While some states take a permissive approach to distributed generating resources, others permit utilities to jealously guard their monopoly status.

A full review of the legal and regulatory landscape for behind-the-meter retail power sales would not be possible in this brief article. However, there are key issues that must be analysed to determine whether a behind-the-meter structure will be feasible for any particular project. First, does state law permit entities other than public electric utilities to sell power on a retail basis? Second, what is the definition of ‘public utility’ or its equivalent term under state law? And would the mere sale of electricity to a third-party render a generator a public utility within this definition? Third, would the project be located in the franchise area of an investor-owned utility (IOU), an agricultural co-operative, or a municipal utility? State law tends to treat these utilities differently, so understanding the precise location of any particular project is important in understanding the applicable regulatory regime.

State laws and regulations will play an important role in determining whether a behind-the-meter structure will be feasible for any particular AI data centre project. As the AI boom gathers pace and further constraints are felt on the grid due to load growth and the retirement of certain legacy assets, states with more permissible regulatory frameworks may be successful in attracting greater investment for greenfield data centres with behind-the-meter power supply.

Conclusion

Promoting the AI data centre boom has been identified by US lawmakers as being critical to ensuring the country continues to lead in AI. As such, policies are being considered in the context of national security rather than merely on pure economic grounds.

There has been movement at the federal level that should encourage the development of electricity infrastructure resources generally. However, based on the existing legal framework, much of the regulatory reform that would give rise to greater buildout of behind-the-meter power assets would need to occur in states where retail power sales are currently restricted. While the behind-the-meter structure for supplying power to fuel the AI data centre boom may be practical for some projects, existing state laws remain an impediment to full-scale nationwide adoption.