In December of last year, Senator Bernie Sanders called for a national moratorium on the construction of data centers “that are powering the unregulated sprint to develop & deploy AI.” Sanders’ stand follows bipartisan backlash resisting proposals to construct data centers in communities across the country. Recent reporting from Data Center Watch shows an estimated 125% surge in opposition to data centers between March 2025 and June 2025, preventing or delaying $98 billion in data center projects.
As the race for artificial intelligence (AI) has accelerated, communities – particularly those in rural and disadvantaged areas – find themselves at the center of the technology’s future. On the one hand, data centers potentially offer an influx of capital investment, demand for labor and additional tax revenue to help revitalize local and regional economies.
However, data centers also come with distinct costs, namely rising water and energy costs, increased strain on utilities infrastructure and environmental degradation and pollution that can have significant ramifications for public health. With capital expenditures on data centers expected to continue to increase this year after unprecedented levels of investment in 2025, it is critical that community-based organizations, local governments and residents understand the tradeoffs of data centers as well as the levers at their disposal to ensure construction happens (or doesn’t) on their own terms.
In the following article, we will provide an overview of what data centers are, explain what’s driving the rush to build them and analyze what geographical areas tend to be most vulnerable to the rapid expansion of data centers. We will also discuss strategies and policy levers community leaders have to ensure data center projects lead to equitable community development.
What is a data center?
Data centers are physical buildings housing file servers and networking equipment that can store, process and analyze various data sources including text, images, code and other information. Data centers house the infrastructure needed to run the computer algorithms behind programs like ChatGPT. While more recent advances have been driven by AI, data centers have been an important part of the modern digital economy for many years, helping to support internet connectivity and large software systems often referred to as the “cloud”.
Most data centers house servers (connected computers) and storage systems capable of vast numbers of calculations and computing applications. There are several general categories of data centers to be aware of:
- Micro data center: The smallest recognized data center is generally used by single companies or for remote offices. Micro data centers usually exhibit a capacity of 10 server racks or less, which equates to a total capacity of approximately 140 servers. Micro data centers typically occupy less than 5,000 square feet of space and draw on no more than 100 – 150kW of energy.
- Small data center: Small data centers typically require between 5,000 – 20,000 square feet of space and may host anywhere from 500 to 2,000 servers, drawing on around 1 – 5 MW of energy.
- Average data center: The average onsite data center typically has between 2,000 and 5,000 servers. Its square footage could vary from between 20,000 square feet and 100,000 square feet with an energy draw of around 100 MW.
- Hyperscale data centers: The International Data Corporation (IDC)’s definition of a true hyperscale data center states that the facility should contain at least 5,000 servers, occupy at least 10,000 square feet of physical space, and draw on over 100 MW of energy.
The recent drive in construction has primarily focused on hyperscale data centers because they have enough capacity to run energy-intensive AI services and training processes. As such, our focus in the remainder of the article will be on hyperscale data centers.
What’s driving the rush for hyperscale data center construction?
In November 2022, ChatGPT took the world by storm, followed closely by a slew of similar products all driven by a new form of AI called large language models (LLMs). With remarkable speed and novelty, ChatGPT and its peers (i.e., Anthropic’s Claude, Google’s Gemini, etc.) are able to handle more complex inquiries from users and respond with personalized answers in a matter of seconds. Over the last couple of years, the models driving these tools have become more efficient and accurate while also expanding to new features like image and video generation. The resulting scramble to innovate and create the dominant AI product is expected to involve trillions of dollars of global investment.
As tech companies compete for users and investors, their executives are not shy about their ambitions to reshape society via the new technology. From curing cancer to solving climate change, it would appear as though AI is the silver bullet to all the world’s problems according to the executives leading the charge. The most ambitious objective is the pursuit of artificial general intelligence (AGI), or superintelligence, models that would either rival or exceed human capabilities. While progress toward a technology of this caliber remains unrealized and overhyped for the time being, the AI arms race has exploded in earnest.
To improve and service current AI models, tech companies have three key constraints: data, compute and energy. Data is what allows software engineers to improve AI models through a process called training. Training AI involves feeding data into a model to “teach” it to respond to different stimuli. The actual mechanics of this process are less important than the main takeaway: tech companies need a solid infrastructure and ample energy as well as the data to continue improving AI.
Data
In the digital age, data is abundant and widely available to Big Tech companies. From ordering a product on Amazon to liking a post on Instagram, all of our interactions online contribute to our digital footprints. For behemoths like Microsoft, Amazon and Google – the companies that are largely driving the investment in AI – data is easy to come by for two main reasons. First, they own many of the most dominant apps we all use on a daily basis. Second, any data these companies want but do not have direct access to can often be purchased.
An entire sector of so-called “data-brokers” have created markets for the buying and selling of data. Due to the lax digital privacy laws in the United States, the companies who develop and own the websites we visit, the apps we scroll on and the streaming services we watch not only collect our information but often have a right to sell that data to other firms. Furthermore, using published works to train AI models has revealed grey areas in copyright law. All this amounts to an abundance of data that tech companies can use to train and improve their AI models.
Compute
However, training AI models requires more than just data. It also requires the computational infrastructure needed to process it, often referred to as compute. Data centers house that equipment as well as the systems needed to ensure they run safely and properly. AI models also need compute power to maintain their current services. Every query into ChatGPT requires the digital infrastructure in a data center to execute the request of the user.
As companies push to increase the number of users, their computational demands have increased exponentially in order to fulfill all of the search queries and other tasks AI models have become tasked with. As the race to innovate hastens, Big Tech companies’ quest for AI innovation and expansion hinges on their ability to construct data centers. As a result, they have invested heavily to build large, high capacity hyperscale data centers across the country as quickly as possible.
Energy
The final piece of the puzzle is obtaining the energy required to run a data center. Because they are basically warehouses filled with computers that are running all day every day, data centers consume vast amounts of energy. Federal researchers project American data centers could double, or even triple, their electricity usage by the end of this decade, going from roughly 4.4% of national electricity sales to upwards of 9-12%.
By 2030, the US may consume more electricity for data processing than it does to manufacture all its steel, aluminum, cement and other energy-intensive goods combined, demonstrating the sheer amount of energy needed to meet the demand. The cost of electricity and the current state of grid infrastructure has become one of the top considerations for tech companies when deciding where to build a data center. Communities who will be impacted by data center construction are similarly on alert for how data centers will impact their electricity bills as well as their pollution-related health and environmental impacts.
Location, Location, Location: Where Are Data Centers Built and Who Is Most Affected?
Where data centers are built matters. Hyperscale data centers house the compute power needed to build the future of AI, but their presence does not come without consequences. Community advocates and residents resisting data center construction in their backyards raise a number of issues with the potential and realized effects data centers can have, including rising electricity bills, environmental pollution and large tax incentives for firms without an economic benefit to community members. This raises important questions about what areas of the country are most vulnerable to data center construction proposals. Based on data from the FracTracker Alliance and the American Community Survey, we found that there are certain areas of the country where data center construction tends to occur.
Figure 1 shows the states where small and medium-sized data centers tend to be concentrated in. Virginia, Texas, Georgia and Pennsylvania are among the states with the greatest number of data centers currently operating. Similarly, Figure 2 shows that these four states are also among the leading locations for hyperscale data center operations. However, Texas and Georgia appear to be leading in hyperscale data center development, surpassing Virginia.
The hyperscale data center construction trends bear both similarities and differences to data center construction more broadly. States like Texas and Georgia are represented at the top in both categories, whereas Michigan has a greater focus on hyperscale data centers. This could represent the changing policy environment for data centers more broadly. Some states are making an intentional effort to attract hyperscale data centers and compete with states that have historically been friendly to data center development.
It is important to note that the publicly-available, crowdsourced data from FracTracker Alliance is incomplete and under-representative in several areas of the country compared to other maps based on private data, such as California and various states throughout the midwest. Companies go to extensive lengths to mask the locations of data centers and the names of the operators, and the availability of comprehensive data locating data center builds remains a key limitation of efforts to analyze construction and operation trends.
What We Know About Data Centers and Community Benefits
When data centers are proposed, companies often tout their community benefits: increased property taxes for local schools, short and long-term job growth, improvements to local infrastructure and economic activity that supports small local businesses. However, it is important for community members to remain vigilant when faced with such promises. The impacts of the most recent push for hyperscale data centers have not always lived up to the promises made before construction was approved. These failures to deliver for residents are especially clear when accounting for costs to taxpayers in the form of tax breaks and energy bills.
Data centers are often pitched as boons to local labor markets, creating well-paying long-term jobs in high-demand industries such as security, operations and information technology (IT). They also provide short to medium-term job opportunities for construction workers, electricians, plumbers and other skilled tradespeople who are needed to build them. However, the long-term benefits beyond the construction phase are less clear.
Generally, as more data centers have gone online, there has been a corresponding increase in people working in them. At a national level, employment in data centers increased more than 60% from 2016 to 2023, growing from 306,000 to 501,000 jobs. A 2017 report sponsored by the US Chamber of Commerce Technology Engagement Center estimated that a typical data center employs around 1,700 local workers during the construction phase but only about 150 in the years following. Even so, data centers may have downstream effects supporting jobs at other firms. The Data Center Coalition, the trade association representing the data center industry, commissioned a report showing that from 2017 to 2021, each direct job in the data center industry supports more than 6 jobs elsewhere in the US economy.
However, these estimates involve both the economic impact of income earned in the data center industry as well as spending across the entire supply chain. This means that those 6 added jobs are probably not in the area where the data center was built and, therefore, do not provide a direct community benefit. Moreover, the jobs that data centers do create locally are typically low-wage, term-limited, non-technical positions, such as security, maintenance and janitorial roles. These roles are often filled by contractors rather than full-time employees, meaning they lack union protections, health benefits and job security.
The disparity between short and long-term job growth as well as the quality of long-term employment opportunities highlights the issue with characterizing data centers as beneficial for workers. Unlike manufacturing or distribution facilities of comparable size that produce or store goods and require human labor to function, most of the “work” in a data center is being done by the computer systems and servers housed within them. In this sense, data centers are more akin to traditional infrastructure projects like highways or bridges that have large upfront costs and provide value through their presence rather than the jobs they create in the long run.
Although data centers do not create much job opportunity for local communities, the handful of long-term jobs they do offer could make a big difference in small towns. In these cases, the question becomes whether or not the benefits of those local jobs outweigh the costs of making the project happen in the first place. Increasingly, state and local governments are favoring subsidies in the form of tax breaks or exemptions to encourage AI companies to build in their communities.
This approach is misguided and provides limited return on investment for taxpayers as evidenced by reporting from ProPublica on the effect of a tax break for data centers in the state of Washington. Trends in Alabama show that the amount of taxpayer funding per local job created can range from about $205,000 to over $1 million with little if any oversight or enforcement mechanisms to ensure that the tech companies deliver on promised benefits. Similarly, a 2016 analysis from Good Jobs First found that data center subsidies cost governments nationwide about $2 million per job created.
More recently, researchers have questioned whether tax breaks even change firms’ choices about where to build. A study from Rice Business and Harvard Business School professors found that other factors such as construction costs, network connectivity and access to reliable, affordable electricity are more important to companies building hyperscale centers. So, not only are tax breaks a bad use of public funds because they produce poor return on investment, but they may not even have their intended effect.
Further complicating the cost-benefit analysis, energy rates for consumers in the areas surrounding data centers often see increases in monthly energy costs. Utility commissions are typically organized such that the costs of repairing and creating new power lines and power plants is socialized across the entire user base.
Because data centers require large amounts of additional grid infrastructure to receive the electricity they need to keep their systems running, all consumers in the market see an increase in their bills to pay for those upgrades. Additionally, due to the energy demands of large data centers rivaling those of entire cities, the sheer volume of demand without a corresponding increase in energy supply unsurprisingly leads to higher utility prices.
Data centers can in fact create jobs for a local economy, but the impact is often small and transitive with most being temporary positions for constructing the facility. Even so, many communities are investing taxpayer money in subsidies to attract data center development. Not only is it unclear whether such public investments provide a positive return on investment for taxpayers, but they also do not appear to impact firms’ decisions about where to build hyperscale data centers.
Community Strategies for Taking on Data Center Development
Communities who could benefit from what data centers have to offer must ensure they protect themselves and their neighbors before approving construction plans. Data centers can be a net benefit to communities, but only if local leaders and community members organize to ensure that costs are minimized while retaining maximum benefit. Doing so requires coordination and an understanding of the tools residents have to gain leverage on data center developers who are backed by some of the most wealthy and powerful companies in the world. Local leaders can take a proactive approach to planning data center development by addressing sustainability, zoning and infrastructure concerns to ensure data centers are designed to operate in the best interests of communities.
Local governments have several policy options and tools to shape data center construction and operations. The first is through zoning and permitting reforms. Older zoning codes do not explicitly address data centers so they often fall into grey areas between different zoning classifications. The expectations associated with development and overall length and speed of the permitting process depend on how a data center project is classified.
Amending zoning codes can help to alleviate uncertainty during the classification process by explicitly defining data center parameters and making clear demands in terms of what the community expects as far as where and how data centers are built. This can include but is not limited to noise, aesthetics and environmental impacts. In addition to zoning reforms, some local governments have passed other local ordinances to increase data center transparency around resource usage, ensuring accountability for delivering on promised benefits and advancing other place-based priorities for communities.
Perhaps most importantly, local governments must exercise caution in offering local tax incentives. As described earlier, unconditional tax subsidies for data centers are poor public investments. Instead of seeing data centers as default positives, local leaders need to make a more calculated analysis of the impact on the local infrastructure, utility rates and capacity and the equitable distribution of economic benefits.
One of the most promising approaches is to mold data centers into opportunities to enhance local infrastructure, expand digital access and build renewable energy capacity. When tax breaks are offered, they should be targeted to the construction phase and tied to clear community demands. These can include requiring local hiring and contracting firms to ensure data centers pay local property taxes as well as requiring data center companies to cover the costs of building or expanding utility infrastructure.
State and local leaders should not shy away from simply pressing pause on any new data centers until they have taken the necessary time to weigh all of these considerations, listen to their communities, and clearly define how data centers will be zones. Recently, both state and local governments are adopting time-limited moratoriums or outright bans on data center facilities. Multiple councils across Georgia, Illinois, Kentucky, Kansas, Ohio, Tennessee, Missouri, Idaho and Pennsylvania have all introduced moratoria of different restrictions against data center development or denied ordinance changes that would allow them to be built.
Conclusion
Data centers are the key to the future of AI. Regardless of whatever beliefs one may have about AI, most of the decisions about how this technology is created and to what ends are being made by a few executives at Big Tech companies and their shareholders. Data center construction sites may be one of the only opportunities for the rest of society to leverage our perspectives and ideas to our overall benefit.
Simon Wang is the Economic Mobility Specialist with NCRC’s Economic Mobility team.
Joseph Dean is the Racial Economic Research Specialist with NCRC’s Research team.
Photo credit: Google DeepMind via Pexels.
