6

Societal Considerations of Data Center Expansion

Data center expansion can affect economies and communities through impacts on energy consumption, the environment, and workforces. Prashant Shenoy, University of Massachusetts Amherst, moderated a panel that explored impacts of the expansion of artificial intelligence (AI) data centers for local communities, society more broadly, and the global environment, as well as policies and regulations to address these impacts. The panelists were Kelly Sanders, University of Southern California; Nate Benforado; Southern Environmental Law Center; Tim Cywinski, Sierra Club; and Varun Rai, University of Texas at Austin.

PANELIST REMARKS

Approaching Artificial Intelligence Concerns as a Catalyst for Progress

Sanders posited that the need to address issues around the expansion of AI data centers could be leveraged as a catalyst to spur progress toward decarbonization and other goals as the United States charts the future of its electric power grid. The grid is the backbone of the physical economy, while AI is becoming the backbone of the digital economy, and both are critical to modern life. Where they intersect, issues of national security, global competition, geopolitics, and decarbonization arise. While the growth of AI and its related energy demands has potential negative consequences, Sanders highlighted how the current moment also offers unique opportunities to create policy solutions with broader impacts.

Echoing a point made by other speakers throughout the workshop, Sanders noted that data center energy demands are important and growing but still are anticipated to represent only a relatively small portion of the rapidly increasing demands on the U.S. grid driven by vehicle and building electrification, hydrogen production, and other factors over the next few decades. To meet future demands, raise grid capacity, and achieve decarbonization goals, she underscored the need for technology companies, utilities, policy makers, and other stakeholders to create equitable, effective policies that protect ratepayers and minimize health and environmental hazards. However, it takes more than good ideas to create policy, Sanders said, noting that policies are connected with time and space. To make the most of this moment in which there is a great deal of attention focused on AI and its future, she suggested finding ways to connect policies around AI and technology with other goals for the grid more broadly. “We could look at this issue like the sky is falling, but I really challenge us to say, we’re going to meet the moment and we’re going to pull through decarbonization and all of these other societal priorities as we think about this issue,” Sanders said.

Increasing Transparency in Decision-Making

Benforado highlighted the importance of transparency in informing smart decisions around data center expansions and their interactions with the electric power grid. Given the potentially large impacts that data centers have on land, water, and energy needs both locally and globally, he emphasized that local utilities and officials need more information about these impacts to make intelligent decisions and plan for the future amid unprecedented growth and uncertainty.

Data center operators negotiate in private with localities and utilities for zoning approval and service agreements. Utilities today are using those agreements to amend their plans and justify new fossil fuel–based infrastructure, which generates profits but deprioritizes climate goals. In Virginia, two main utilities have what is essentially a monopoly in their service areas, and regulatory oversight merely requires reliable power at a fair price. In addition to posing a setback for decarbonization goals, Benforado noted that this situation could result in an overbuilt grid if planned data centers do not materialize or AI fails to become commercially viable, leaving local customers footing large bills.

Benforado posited that it is possible to create a more sustainable path that supports AI data center growth without sacrificing decarbonization goals. To achieve this, he underscored the need for more transparent negotiations; more coordination at the local, state, and regional levels; and more collaboration among utilities, regulators, and other stakeholders to

make informed decisions that mitigate uncertainties, maintain reliability, and prioritize clean energy strategies such as storage, offshore wind, and grid-enhancing technologies.

The Urgency of Action to Protect Public Interests

Cywinski outlined an urgent need for strong policies to combat AI data centers’ power demands, which he said unfairly threaten climate goals and raise consumer electricity rates. As examples of the negative impacts AI data centers can have on local communities, he said that water-hungry AI campuses are being built in rural areas that are already struggling with drought, while new utility infrastructure is worsening air quality in economically disadvantaged areas. In addition to these environmental impacts, he said that data center expansion is exacerbating rate increases, creating serious economic strains, especially in places like Virginia where 29 percent of residents already report forgoing basic necessities to pay their electric bills.

Cywinski said that the United States can be an AI leader while minimizing harm to the public and the environment. To do so, however, will require overcoming unfair structures that support data center growth at the expense of protecting communities and the environment, policy lags that allow for growth to continue unchecked while protective policies are being formulated, and a dearth of transparency into data centers’ needs and operations. While there are promising technological solutions, he said that time is running out to craft effective policy and protect public interests.

Opportunities and Challenges in Transforming the Electric Industry

Rai highlighted key opportunities and challenges AI data centers bring during a pivotal moment for the electric power industry. He said that AI can enable enormous change, not just in the electricity market but across the entire economy and society, but the challenges of meeting the demands of AI data centers are markedly different from those associated with the types of load increases that drove grid expansions in the past.

Rai posited that these large and variable loads present opportunities to transform the electric industry, utility business models, and regulations, as well as an opportunity to push technology companies toward electrification. For example, innovations in grid technology and mechanical and electrical engineering will be needed to enable data centers to dynamically interact with the grid. In addition, he suggested that radical changes will be needed to update the utility sector’s approaches to regulation, infrastructure investments, cost recovery, rate design, load

integration, and more. He added that AI can also drive behind-the-meter changes to enable better grid integration.

There are also many challenges, many of which arise from uncertainties with regard to the scale, timeline, and location of data centers, as well as uncertainties about when and how AI will be commercially viable. To plan amid these uncertainties, he underscored the need for a multi-stakeholder collaboration that encourages experimentation, innovation, flexibility, and diversity. Another challenge is the lack of adequate renewable energy generation and storage, which has led to a need for more natural gas generation in order to meet growing demands. However, he posited that even that challenge can be seen as an opportunity to clean up the natural gas production process and minimize associated emissions.

PANEL DISCUSSION

In an open discussion, panelists delved deeper into data center impacts, strategies for driving change through policy, AI challenges and opportunities, and increasing electrification.

Data Center Impacts

Andrew Grimshaw, Lancium, expressed his concern about Virginia’s approach to data centers and the impacts they are having in the state. He stated that it is important to separate AI’s impacts on society from data centers’ impacts, and he posited that data centers should be built where there is more access to renewable power and storage and more land to build these resources out—places like rural Texas, not Virginia. Cywinski agreed, noting that he is also opposed to what he characterized as “reckless” data center development in places like Virginia. Despite the economic and environmental drawbacks for communities, projects are still being approved and serviced through increases in fossil fuel generation. He argued that this situation is both environmentally irresponsible and unfair to Virginia’s taxpayers, but it is the easiest solution for utilities and promises them large profits.

To address this disconnect, Cywinski suggested that AI companies should pay more of the energy and infrastructure costs associated with data center expansion. He noted that technology companies have over-promised and under-delivered before, and he said that it is unfair for local communities to pay for new utility infrastructure that brings a plethora of problems, including excess construction noise, more transmission lines, increased bills, and decreased air and water quality. “We’re subsidizing them through taxpayer money. We are giving them extremely luxurious tax benefits. And then we have to pay for the new power plants through

our monthly electric bills,” he stated. “Is this the price of progress in technology … that a group of people in one state or a collection of states have to pay for the benefit for the rest of us?”

Jonathan Herz, Environmental and Energy Study Institute, wondered if economic analyses of data center builds included cogeneration potential or external costs like energy. Benforado replied that their economic analyses are not made public; data centers file needs and projections with utility commissions, but the process moves too fast, and the terms are subject to very little oversight. In addition, he said that data centers and utilities are so fragmented that it is challenging to conduct holistic evaluations to understand and optimize or mitigate their cumulative impacts on the grid, communities, and the world.

Rai highlighted the value of more widely communicating the environmental impacts of AI data centers to increase public awareness of these issues. He also suggested that data centers could create a model of shared responsibility and accountability through increased transparency, data sharing, community engagement, and commitments to self-regulated decarbonization goals. Simply modeling data center demand does not always capture the full picture; instead, he posited that researchers should more carefully examine these demands to better coordinate the dynamic and growing drivers of AI across sectors. He reiterated that in the short-term, AI data center growth will require increased natural gas production, which is unfortunate but does create opportunities to experiment with new approaches.

Benjamin Lee, University of Pennsylvania, asked if carbon-free voluntary pledges had any impact on decreasing data center emissions or advancing clean energy generation. Sanders replied that emissions are challenging to evaluate and there is a lack of standardization around how to define “100 percent clean energy.” However, she said that many large companies do appear to be concerned about their emissions profiles and suggested that more rigorous definitions and standards could help to advance meaningful progress. Rai added that hourly matching can advance clean energy generation and there has been progress on this front, but the growth is simply happening too quickly for renewable generation to keep up. Companies may make carbon-free pledges or pay more for power purchase agreements, but if they do not directly influence actual energy generation, utilities may often wind up using fossil fuel–based generation to meet the demand anyway. As a result, Rai said, “the rest of the system gets browner while the commitments get greener.”

Building on this point, Cywinski said that in an ideal world, data center growth would pause until the benefits and harms could be more accurately weighed at the local and global levels to inform responsible decision-making. Unfortunately, that is unrealistic—AI is not slowing

down, pointing to an urgent need for safeguards to encourage responsible growth. Creating such safeguards, however, will require massive political will, something that in turn requires public education, influence, and representation to create a better-informed citizenry that can move policy forward, he said.

Driving Change with Policy

Panelists discussed how policies on the part of governments, utilities, and technology companies can help to address concerns and drive change. Andrew Chien, University of Chicago, noted that data centers are at the intersection of a patchwork of municipal, regional, and state regulations and policies that weigh benefits and trade-offs differently. Cywinski agreed, stating that the lack of uniformity among regulations and safeguards enables large companies to influence local decisions and obscure potentially negative impacts. Greater harmonization around regulations and policies could force them to at least consider, if not prioritize, energy efficiency and public health issues, but right now policy lags behind the industry and technological advances, adding to the urgency of the problem.

Benforado suggested that it would be very powerful if data centers united around decarbonization goals and advocated for better regulation at the state, local, and utility-commission levels. Cywinski agreed, noting that in his experience, industry interests can move quickly to influence and accomplish policy goals. He suggested that the technology industry should use its influence to uphold public climate commitments, advocate for clean energy policies, and oppose increased fossil fuel production as a way to meet data centers’ energy demands. Sanders also pointed out that some localities are relaxing air and water quality restrictions to attract data centers—for example, allowing large deployments of diesel-powered backup generators—and stressed the need to advocate against decisions like these that allow data center expansion to worsen pollution in local communities.

Rai agreed that policy goals are important but noted that policies often change when administrations change. However, even though a new administration could deprioritize clean energy policy or innovative financial incentives, he posited that it is still possible to better align the current patchwork of state and federal regulations through consumer and activist-led engagement focused around increasing data sharing and transparency and holding utilities and technology companies accountable for their commitments and actions. Benforado shared that there are bright spots; for example, some states are passing new regulations or even banning data centers, and others are experimenting with de-risking

mechanisms, such as preferential systems or interruptible contracts, to protect existing customers. He also said that innovative pricing structures that incentivize data centers to choose between renewable or fossil fuel–based energy or implement demand response mechanisms during peak load times could have a large impact.

Chien noted that another challenge is that stakeholders have to make important decisions amid great uncertainty. One way to tip the scales in favor of clean energy generation over fossil fuels—in line with public goals and, in many cases, legal mandates—is to alter cost recovery rules to drive decisions in the direction of decarbonization. Sanders agreed that policy tools such as financial incentives, loan guarantees, and other de-risking mechanisms have successfully increased the appeal of emerging technologies. She added that policies that make incumbent strategies less attractive are also needed to encourage utilities to consider alternatives to building more fossil fuel–based infrastructure. Utilities are used to “perverse economics” that essentially incentivize the building of capital-intensive projects, which can come at the expense of more clean and economical options like energy efficiency, demand-side management, and clean energy sources. Cywinski agreed, noting that creating these mechanisms can be easier than passing policies, especially when what is popular and what is possible are not in alignment. In addition, it can be especially challenging to make gains on policy when industry has a strong influence over legislation, which he said is often the case with utilities.

Artificial Intelligence Challenges and Opportunities

Shenoy asked the panelists to elaborate on the challenges and opportunities they see with AI. Sanders agreed that AI presents opportunities to transform society and drive clean energy, but these benefits could also come at great costs. Utilities are arguing that they can only meet today’s increased demand with an increase in natural gas investment, forgoing decarbonization strategies. She said that utility customers should be a part of these conversations. There are already efforts from large load customers interested in having more opportunities to buy clean energy to band together and pressure utilities to choose more clean energy pathways going forward. However, there is also the risk that buyers can unite and lobby for their interests through processes that are opaque to the public. “We have this catalyzing moment,” she stated. “Do we use this to move fast toward a decarbonized future … or do we use this moment as an opportunity to backtrack into natural gas?”

Rai agreed that AI has tremendous potential value, and meeting its energy demands—whether through renewable energy or fossil fuel generation—will have local and global impacts. He said that community

perspectives and increased data sharing should be integrated into early-stage discussions to improve transparency, create a more holistic picture of the impacts, and avoid regrettable decisions. If AI is wildly successful, the public would benefit from reduced energy prices and returns on public subsidies, and AI companies would avoid being blamed for high energy prices and poor grid reliability.

Bill Billman, Department of State, expressed concern that unprecedented, large-scale infrastructure projects are being proposed to meet what is a highly uncertain and ambiguous forecast of AI profitability. He also noted that the demand is relatively small compared to other energy demands, suggesting a need to take a step back and reflect on how complex and urgent the problem truly is. Benforado appreciated this perspective and said that the uncertainty makes it extremely challenging for grid and utility operators to plan and prepare. In his view, AI can improve the grid, accelerate decarbonization, and create revenue in the communities that host data centers. However, he said that those communities should maintain authority over these decisions, ensure public concerns are fully addressed, and investigate data centers’ cumulative impacts, especially where they are clustered. Chien pointed out that while generative AI and ChatGPT are recent developments that have garnered a lot of attention, a broader suite of other AI technologies have become pervasive in many applications and have been successful for nearly a decade. Some projections may overstate future growth, but he said there is ample reason to believe that AI demand is not simply going to fizzle out.

Spurring Grid Modernization

Shenoy asked panelists to elaborate on opportunities for AI data centers to contribute to the overall modernization of the grid, in light of broader electrification trends. Sanders reiterated her view that the current attention on AI and data centers brings a powerful opportunity to drive progress in grid improvements more broadly and outlined several examples of policy ideas in this space. First, she said that policies are needed to encourage better leveraging of grid-enhancing innovations that enable capabilities such as increased situational awareness, efficiency standards for transmission lines, or reconductoring via advanced materials. Second, she said that the deployment of stable and dispatchable clean energy resources, nuclear power, and enhanced geothermal technologies can be accelerated with innovative financing structures such as clean tariffs, direct policies, or buyer’s consortia. Third, intelligent demand-side management will require improved data sharing, privacy protections, and cybersecurity standards. Finally, she suggested fast-tracking creative interconnection and permitting plans that increase grid capacity, such as

siting transmission lines along the federal highway system, employing AI to accelerate environmental permitting, and repurposing retired or underutilized connections.

All of these suggestions, Sanders continued, require regulations, structural reforms, and multi-stakeholder conversations to better align utilities’ business models to reward responsible decarbonization strategies that also maximize community benefits. In addition, she emphasized the need for data center owners to become active partners in improving grid storage and resilience to extreme events. Herz pointed out that nuclear power will not progress until there is a safe system for controlling and disposing of waste.