Why Are They Building All These Data Centers?

A Brief History of the AI Boom

Sam Law

Introduction

We are living through an artificial intelligence (AI) data center boom. Across the United States, the largest tech corporations, fossil fuel companies, the biggest banks on Wall Street, and ordinary people's retirement funds are pouring money into data center construction at a pace with few precedents in American history. The four biggest tech companies alone are on track to spend roughly $750 billion on the AI infrastructure and data center buildout in 2026—up nearly 80 percent from the record they set the year before.

Relative to the size of the whole economy, the spending spree has already exceeded the late ‘90s dot-com bubble and is closing in on the railroad booms of the nineteenth century, making it the single largest investment in infrastructure in our lifetimes. By one widely cited calculation from the Harvard economist Jason Furman, the investment categories tied to the AI buildout accounted for 92 percent of all US economic growth in the first half of 2025. Take out spending related to data centers and the economy grew only one tenth of one percent, or basically not at all. This is why claims of economic growth feel so distant from most Americans' everyday reality, where prices are rising faster than our wages. That reality is hidden in economic reports and politicians’ speeches that point to the economic growth driven by the AI data center boom, growth which only benefits big banks, tech companies, and a handful of private contractors.

Emblematic of the scale of this boom is Texas, which, thanks to rampant corruption, fossil fuel infrastructure, and lax regulation, is now the largest market for new proposed data centers. As of this summer, developers have asked to connect more than 438 gigawatts of new large-scale projects—almost all data centers—to the Texas grid. For perspective: the entire state has never drawn more than about 90 gigawatts, even on the most brutal summer day, the record for power demand across the entire United States set this July was 759 gigawatts.

They are proposing to build enough energy infrastructure to power five Texases. That's equivalent to 2.5 times the total electricity used by every household in America. It's unlikely that all of this will get built, but even just a fraction of what's proposed means industrializing rural landscapes, building a massive amount of new power plants that will lock in decades of greenhouse gas emissions and pollution, and condemning future generations to long term consequences in the form of drought and catastrophic climate change.

They Are Buidling Data Centers For AI, Not You

Even if you don't live in Texas, chances are they are planning a data center near you. That should worry you: data centers devour land, water, and power, and they arrive with higher electricity bills, health hazards, and a corrosive secrecy that eats away local democratic control over our communities' futures. Facing this tsunami of development, many of us are left wondering: why now? What is driving the boom behind these data centers? Who are they for? And why has our entire economy become singularly dedicated to the buildout of this infrastructure, regardless of the needs of our communities?

There is a simple answer, and a more complicated one. The simple answer is what everyone already knows: it's for artificial intelligence. Building and running new artificial intelligence requires an almost unthinkable amount of computing power, and more water and electricity than ever before. When communities push back, data center developers often make people feel like guilty hypocrites. They tell people, "You use data centers every day: the movies you stream, the map apps you use, your email, your family photos—all of that lives in a data center. You wouldn't want to give that up, would you?" These data center developers are lying to you. ¹

We do not have to accept the dystopian future the data center developers, blinded by their fantasies of profit and power, are building for us. Instead, we can fight for a future we want.

The data centers they are building don't have anything to do with the cloud-based services you're accustomed to using on the internet. The internet as we know it runs just fine on data centers that already exist, which are focused on storage and serving data and are much smaller and less energy intensive. What they are building is a new form of data center specialized in the complex computation needed to train and run modern AI. Now for the slightly more complicated question: why do they need data centers for AI, and why are the wealthiest banks and corporations pouring so much money into them right now? To answer that, we need to understand more about the type of AI they are building.

From Academic Research to Business Plan

The new artificial intelligence technology behind ChatGPT, Gemini, and Claude works by finding patterns and making predictions. These systems are fed almost unimaginable quantities of text and images, a large fraction of everything ever posted to the internet, and when this data is analyzed using complex mathematical modeling, the underlying statistical relations can be learned. This produces the "large language models" that themselves contain all these relations: which words tend to follow which and what a good answer looks like.

The basic ideas underlying this technology are relatively old. Neural networks—predictive programs loosely inspired by the brain—date back to the 1950s. The method for training them, called back-propagation, was developed in the 1980s. For decades, however, this technology was more of an academic curiosity, and leading researchers were skeptical it would actually result in useful forms of artificial intelligence. Then, within a few years, two breakthroughs transformed the technology into the cutting edge of AI research and the object of the largest corporate spending spree in modern history.

The Chips Inside the Data Centers

The basic ideas underlying this technology are relatively old. Neural networks—predictive programs loosely inspired by the brain—date back to the 1950s. The method for training them, called back-propagation, was developed in the 1980s. For decades, however, this technology was more of an academic curiosity, and leading researchers were skeptical it would actually result in useful forms of artificial intelligence. Then, within a few years, two breakthroughs transformed the technology into the cutting edge of AI research and the object of the largest corporate spending spree in modern history.

The first breakthrough came in 2017 when researchers at Google published a paper called "Attention Is All You Need." Earlier AI machine learning read text similar to how you’re reading this sentence right now, in order, each step waiting on the one before. This was incredibly slow. Google's invention—called the transformer—lets a machine absorb an entire passage at once and find, for every word, which other words it should "pay attention" to. Training could now be split up into millions of small pieces and run simultaneously, across thousands of different computer chips at once. This allowed the process to be rapidly scaled.

Those chips are worth a brief detour, because they are what fill these AI data centers. Training and running the AI models comes down to arithmetic: multiplying enormous grids of numbers together in an operation called matrix multiplication. AI requires billions upon billions of these operations. In the early 2010s, some researchers realized that the computer chips optimized for rendering graphics on video games, called GPUs, were freakishly better at these AI-specific calculations than the regular chips used for most computation.

Today, this has transformed the video game chip company NVIDIA into the most valuable company on earth, with an estimated value of five trillion dollars. TSMC, the Taiwanese company that makes these AI chips, is now one of the most strategically important companies in geopolitics. As factories that once made chips for consumer electronic devices are shifting their focus to the more profitable production of chips for AI data centers, the cost of laptops and cellphones is spiking dramatically. These specialized chips make it clear what new data centers are for. A warehouse filled with tens of thousands of specialized GPUs, consuming the same amount of power as a medium-sized city and astronomical amounts of water to prevent the chips from melting, has only one purpose. These are industrial facilities dedicated to the mathematical operations that power AI, not to stream your movies, run the websites you use everyday, or store your family photos.

The Race to Superintelligence

The second breakthrough that led to the modern AI boom was a paper by researchers at OpenAI, the company that would later create ChatGPT. Published in January 2020, "Scaling Laws for Neural Language Models" showed that AI improves predictably as you increase the size of the model, the amount of data it is fed, and the computing power used to train it. The gains in intelligence seemed to follow a smooth curve you could plot in advance, in effect turning a massive scientific problem into a business plan. Hiring the smartest people was no longer the best way to develop AI. Whoever bought the most chips, built the most data centers, and burned the most electricity could have the most powerful AI.

For those developing artificial intelligence, this powerful AI is quite literally the gold at the end of the rainbow. So-called “Artificial General Intelligence” is an AI so powerful it can perform any intellectual task the same or better than any human, whether that is mundane office work, writing an essay like this one, programming computers, or undertaking research in math or science. As Anthropic’s Dario Amadeo put it, AGI would be like having his own “country of geniuses in a data center.”

The first company to develop such an AI or have the data centers to run one would hypothetically be able to fulfill a long running science fiction fantasy: to replace human ingenuity and labor with machines.2 Doing so would make whoever controlled this AI unimaginably wealthy and powerful, a fantasy that, as we see, has driven the world’s most wealthy corporations into a spending frenzy that has no recent historical precedent. Since OpenAI published the scaling laws that said all you need to make a stronger AI is more compute, the race to build the most powerful AI became a race to build the most data centers.

The players in this race are the companies that already run the internet’s infrastructure—Amazon, Microsoft, Google, and Meta, known in the industry as "hyperscalers" for the sheer scale of their data center empires—and well-funded new AI labs like Anthropic and OpenAI that can raise enough money to rent or build their own. The costs involved in developing and training AI exploded, roughly tracking the industry’s growing need for the computation AI data centers perform. In 2020, training GPT-3 cost a few million dollars; today a single training run costs hundreds of millions, and industry leaders like Dario Amodei at Anthropic say multi-billion-dollar runs are next. That means entire buildings, often more than a square mile in size, buzzing for months as their chips perform trillions of calculations, guzzling astronomical sums of water and power.

Training costs aren't the only thing spiraling out of control. The newest models "think" before they answer, burning far more computation on every question a user asks. So the more the world uses AI, the more data centers it takes. And since each company is rushing to be first to develop the powerful AI, the frantic scramble has taken on a name of its own: the AI race.

The Gold Rush

The AI race is a modern-day gold rush, a veritable frenzy of people trying to make as much money as quickly as they can. Just like Levi Strauss made money during the original gold rush by selling shovels and blue jeans, today's speculators are cashing in by trying to build data centers. These aren't just the big tech companies. Fossil fuel companies, facing the rise of renewables and a transition to clean energy, see in the massive power needs of data centers a way to keep the American economy hooked on dirty fuels, in particular fracked gas.

Local land speculators, developers, and real estate magnates make use of widespread political corruption to buy rural land cheap, quickly get environmental and water permits without public review, and then sell this package of land and data center permits to the highest bidder. And big banks and investment firms are pouring money into data center projects using complex forms of financing that have many experts warning of a 2008-style economic crash.

Pulling the Emergency Brake

Driven by hype and sci-fi promises, the data center construction boom has all the signs of being a massive bubble. The continued construction of data centers requires endless rounds of fundraising, and AI companies are relying on ever more sketchy forms of financing. Eventually the bills won't be able to be paid, and this boom will come crashing down, leaving everyday people holding the bill—stuck with new dirty fossil fuel infrastructure, generational loss of our water, deep impacts on community health, higher energy prices, and more. This is despite none of us agreeing that this is the future we wanted. Indeed, some of the most profitable uses of AI seem to be directly opposed to our own interests, whether that is new forms of AI-powered mass surveillance like Flock cameras or the autonomous drones waging unpopular wars.

The AI race is like a runaway train, an out of control frenzy driven by people deluded by their own sci-fi dreams and fantasies of immense wealth. They don't care about what will get destroyed on the way. It is up to us to pull the emergency brake—to demand a vision of the future and technological progress that puts human needs over the greed of the wealthy. Fortunately, opportunities to pull the brakes exist everywhere a data center project is being developed.

Data centers are not necessary, they are not critical infrastructure, and the future they promise is a dystopia we do not want. Across the country, a data center rebellion is already underway. From the Appalachian foothills of Pennsylvania to the Texas Hill Country, from the farm townships of the Upper Midwest to California’s Imperial Valley, communities are getting organized and fighting back. These local fights playing out in rowdy town halls, crowded public hearings, and packed meetings in church basements and VFW halls show that we can win. Proposed data centers have been defeated, and more than five hundred communities have banned or enacted moratoria against new data center construction.

When our communities organize to defend the places we live, we are more powerful than the fossil fuel companies, tech billionaires, and land speculators trying to cash in on this gold rush. We do not have to accept the dystopian future that data center developers, blinded by their fantasies of profit and power, are building for us. Instead, we can fight for a future we want. Our future can be one where our land, water, and the health of future generations is not auctioned off behind closed doors to the highest bidder, a future in which our collective life is organized around the conditions that allow people, communities, and the living world to flourish.