Powering AI data centers:

In September 2024, government officials and private industry leaders announced that Three Mile Island, the Pennsylvania power plant that was the site of the nation’s most severe commercial nuclear accident, will be reopened to provide electricity aimed at offsetting the use by artificial intelligence (AI) and cloud computing data centers.

Industry leaders and Pennsylvania elected officials stressed that the energy produced under the agreement would be “carbon-free,” while also touting associated economic benefits, such as the creation of jobs and additional state and federal tax revenues. Under the 20-year agreement, the plant would be scheduled to reopen in 2028.

The initial positive reaction from the majority of political leadership points to a growing bipartisan consensus that the US must develop energy infrastructure for AI domestically in order to establish itself as a leader in the global adoption of AI.

A fundamental AI policy challenge

The rise of AI has brought with it a unique set of challenges for policymakers and system designers, from concerns about privacy, reliability, and bias, to questions about how to protect intellectual property, national security, and elections.

Add to this a new challenge demanding the urgent attention of both the public and private sectors: meeting the electricity needs of energy-intensive AI data centers.

Electrical demand for these data centers is expected to grow 160 percent over 2023 levels by 2030, according to a recent estimate by Goldman Sachs Research.

A related challenge will be finding ways to satisfy that demand while minimizing carbon emissions through the use of clean energy-generating sources – though how much of a priority carbon neutrality will be in the future could well hinge in part on the outcome of next month’s elections.

Data centers have been described as the backbone of the tech industry. Some of the larger campus facilities can consume at least a gigawatt (GW, one billion watts) of power – enough to power 750,000 homes, or as much as a major city. Leading AI companies anticipate the need for its data centers to have power requirements as much as 5 GW.

The Goldman Sachs study estimates that about 47 GW of incremental power generation capacity will be required to support US data center power demand growth cumulatively through 2030, with about 60 percent powered by natural gas and 40 percent by renewable sources, which calls for around $50 billion in capital investment in US power-generation capacity cumulatively through 2030.

Current power-generating capacity is not nearly sufficient to meet the increasing needs of the tech industry. In some of the major regions where data centers proliferate, tech companies are being told by power suppliers that they are unable to provide electric service for new data centers. Other utilities are temporarily rationing power until new transmission lines and other infrastructure are completed. This has in part led to a flight to more rural US locations that have available power.

New White House Task Force launched

On September 12, 2024, faced with the growing disparity between power supply and demand, the White House convened representatives from hyperscalers (tech companies that provide cloud computing services to businesses and organizations and have their own AI programs), major AI companies, data center operators, and utility companies to discuss ways to further US leadership in AI while meeting clean energy, permitting, and workforce objectives.

Following the meeting, the Biden Administration announced a coordinated effort across government to speed up the construction of new AI data centers that are “in line with economic, national security, and environmental goals.”

To meet these goals, the Administration is launching a new interagency Task Force on AI Datacenter Infrastructure (Task Force) to be led by the National Economic Council, National Security Council, and White House Deputy Chief of Staff’s office. The Task Force will work with AI infrastructure leaders to secure the necessary resourcing; streamline organizational flowcharts; identify how to use existing authorities, versus areas where new legislation is needed; and “properly prioritize AI data center development to reflect the importance of these projects to American national security and economic interests.”

In addition, the Biden Administration will scale up technical assistance to federal, state, and local authorities handling data center permitting. The Permitting Council has been tasked to work with AI data center developers to set comprehensive timelines and allocate funds to agencies that accelerate evaluations for FAST-41 covered clean energy projects that support data centers. The Permitting Council is the federal agency charged with improving the transparency, predictability, and outcomes of environmental review and authorization process for certain large-scale critical infrastructure projects. FAST-41 refers to the title of the 2015 Fixing America’s Surface Transportation Act (FAST Act) that established procedures and funding authorities to improve the federal environmental review and authorization process for covered infrastructure projects.

As part of the initiative, the Department of Energy (DOE) is creating an AI data center engagement team to leverage programs to support AI data center development, including loans, grants, tax credits, and technical assistance. In particular, the DOE will continue to share resources on repurposing closed coal sites with data center developers.

Congressional proposals

The AI/energy nexus has also attracted the attention of Congress.

As part of a flurry of legislation moved in September 2024 by the House Science Committee, a measure titled the Department of Energy Artificial Intelligence Act of 2024 (HR 9671) was advanced. Under the legislation, the Energy Department would “carry out a cross-cutting research and development program to advance artificial intelligence tools, systems, capabilities, and workforce needs and develop artificial intelligence capabilities for the purposes of advancing the missions of the Department.”

The legislation includes a provision ensuring energy security for data centers and computing resources, and another regarding permitting reform. It would authorize annual appropriations of $300 million per year from 2025–30.

With Congress adjourned until after the election, the bill could be considered by the full House of Representatives in November or December 2024, or it could potentially be attached to another piece of legislation.

A similar bill in the Senate, titled the Department of Energy AI Act (S 4664), is scheduled for markup in November 2024. That measure, which would require the Secretary of Energy to establish a program to promote the use of AI, includes a data center provision.

Policy forecast

In a rare show of bipartisan agreement, policymakers from both sides of the aisle are committed to ensuring that the US remains the world’s largest market for data centers. At the moment, however, a considerable degree of planning, construction, and financing are needed to generate and deliver the energy to achieve this goal.

Significant new policies may be expected over the coming years, including investing in operational efficiency through new grants, tax incentives, and regulations.

As noted above, the Biden Administration has prioritized the use of non-fossil fuel energy-producing technologies to power the new generation of data centers as part of an overall push toward green energy conversion. November’s election could determine whether that continues to be a major focus of US energy policy.

These policies are in their nascent stages but are moving forward. As legislation and the regulatory process begin, companies in the data center and AI space would be well served having their competitive advantages reflected in new government incentives, and they are encouraged to position themselves accordingly in anticipation of any expected policy changes and new government opportunities.

For any further information on these developments, please contact the authors.