Data Creep: The Geography of Artificial Intelligence

Since OpenAI’s ChatGPT launched in November 2022, the tech industry has bet big on artificial intelligence. In 2023 alone, U.S. investment in AI totaled $67 billion, with Amazon, Google, Microsoft, Nvidia, and Salesforce leading the charge in expenditure. For daily internet users, the integration of generative-AI tools like ChatGPT and Google’s Gemini into our digital lives has been inescapable, as have been headlines about the technology’s paradigm-shifting potential. However, the normalization of large language model (LLM) programs, as mediated by our omnipresent screens, obscures a rapidly changing physical reality.

The earliest digital computers of the 1940s occupied entire rooms—the entirety of their hardware in full sight of the user—but as they evolved into a mass consumer product, the relationship between our technology and the infrastructure powering it has become increasingly abstract. Devices, now shrunk to the size of our palms, provide instant access to a global network, allowing us to find and disseminate information without confronting the vast expanse of servers and underground fiber optic cables that facilitates such immediate connection. The resource-intensive nature of AI, which vastly exceeds that of previous generations of computing, threatens to upend that out-of-sight, out-of-mind dynamic by driving an unprecedented demand for the data centers that make our “cloud”-based world possible.

The United States boasts the largest number of data centers in the world, with nearly 5,000 scattered across the country. Though gargantuan in size, with sites spanning hundreds of acres, the facilities are typically tucked away at the outskirts of rural or suburban communities. The highest concentration can be found in Virginia, particularly in the northern part of the state. Loudoun County, 50 miles west of Washington, D.C., is home to “Data Center Alley,” a sprawling complex of nearly 200 data centers—75 of which have been built in the last five years. Surrounded by metal fences and tightly monitored by 24/7 video surveillance and security guards, the collection of low-lying windowless structures are not much to look at, but safeguard the thousands of servers and web of cables through which an estimated 70 percent of the world’s internet traffic passes every day. The entirety of Data Center Alley currently comprises 30 million square feet of operational space, with an estimated 5 million more in development.

Similarly vast, architecturally nondescript sites, varying in scale and density, can be found fragmenting farmlands and quiet suburbs across America, particularly in Texas, California, and parts of the Midwest, where energy is cheap, land is plentiful, and tax incentives won by tech lobbyists abound. ChatGPT itself was born in the cornfields surrounding West Des Moines, Iowa, after Microsoft partnered with OpenAI in 2019 to develop “the world’s largest supercomputer.” Currently, Microsoft operates millions of square feet of data centers spread across three campuses in West Des Moines. That’s still not enough. It’s building two new 245,000-square-foot structures on its 124-acre “Ginger” campus, already home to three data centers, with plans for a sixth in progress.

Data centers could consume 8 percent of U.S. energy by 2030, a dramatic increase from today’s 3 percent, spurred largely by power-hungry AI technologies.

According to a recent report by real-estate investment firm JLL, the past two years have seen an “insatiable demand” for data-center construction, with development increasing more than sevenfold over the past two years. This boom is driven by the strenuous energy demands of AI-based computing. A recent Goldman Sachs report calculated that a single ChatGPT query requires nearly 10 times as much electricity to process as a traditional Google search. The report also predicted that data centers could consume 8 percent of U.S. energy by 2030, a dramatic increase from today’s 3 percent, spurred largely by the expansion of power-hungry AI technologies. The push to integrate generative-AI tools into more aspects of our digital lives has already seen data centers quietly driving a significant spike in U.S. energy consumption, contributing to the first surge in national energy demand in nearly two decades.

An appetite for more robust energy sources and user proximity—as well as a newly available supply of downtown real estate in the post-Covid era—is already driving the once rural typology into the heart of urban centers. In downtown Los Angeles, for instance, developers recently completed One Wilshire, a disused midcentury office building transformed into one of the most powerful data centers on the West Coast, processing online traffic for multiple internet companies between Asia and the United States.

Such facilities promise a windfall in tax revenue, providing a powerful incentive for local officials to greenlight development. But the colossally scaled sites have profound impacts on their surrounding communities. Data centers divert large volumes of water to cool equipment—between 1 and 5 million gallons a day. This, coupled with strain on local electrical grids, can drive up a household’s utility bill by hundreds of dollars annually. Noise pollution is also a significant concern. A 2023 Business Insider article detailed the stress put on Loudoun County residents, who complained that a recently completed data center’s industrial cooling fans emitted a constant mechanical whir, comparing the sound to “a propeller” and a “loud drone hovering above, 24/7,” severely impacting quality of life for residents up to five miles from the site.

Companies leading the data-center construction boom reiterate their overall sustainability efforts and pledge to enhance energy efficiency in new facilities. But a recent analysis by The Guardian revealed a pattern of “creative accounting” by major tech companies that drastically downplayed their environmental impact. Using location-based data, the publication found that real emissions from data centers owned by Google, Microsoft, Apple, and Meta were more than seven times higher than officially reported.

Microsoft cofounder Bill Gates, a leading investor in both climate solutions and AI, has said the technology’s potential innovations and advancements will mitigate, and eventually offset, its environmental toll. This tired claim about the cost of progress has floated the industry through decades of increasing deregulation, subsidization, and unfettered expansion. But it’s getting harder to defend. The hyperconsumptive nature of AI has left tech companies struggling to obfuscate their environmental footprint, and its accompanying physical expansion is inviting greater public scrutiny, with some local communities pushing back against data-center development. In the past year, residents in towns like Peculiar, Missouri, and Chesterton, Indiana, have successfully opposed new construction, citing concerns about noise pollution, strain on local infrastructure, and environmental impacts already seen in places like Loudoun County.

It’s unlikely that such grassroots movements can threaten, or even slow down, the tech industry’s big plans for infrastructural expansion. But the increasing encroachment of data centers on daily life, and their heightened visibility in the built environment, may provoke a reconsideration of the foundational myth that drives technological progress at any cost.