The New Roman Empire of AI Data Centers
A year ago, when Sam Altman noted that OpenAI’s “Roman Empire” is the actual Roman Empire, he was not joking. Just as ancient Rome steadily expanded its territory to span three continents and cover nearly one-ninth of the planet’s total circumference, Altman and his cohort are now staking claims to their own massive holdings across the globe—these are not agricultural latifundia, though: they are AI-powered data centers.
Top tech leaders including Altman, Nvidia CEO Jensen Huang, Microsoft CEO Satya Nadella, and Oracle co-founder Larry Ellison are all fully convinced that these new IT-stocked facilities are the future of the U.S.—and possibly global—economy. But data centers themselves are nothing new. In the earliest days of computing, power-guzzling mainframes filled entire climate-controlled rooms, with coaxial cables shuttling data between the central mainframe and individual end-user terminals. Then the late-1990s consumer internet boom kicked off a whole new era of digital infrastructure. Huge purpose-built facilities began cropping up near Washington, DC, stacked with row after row of server racks that stored and processed data for a growing roster of tech companies.
A decade later, “the cloud” emerged as the flexible behind-the-scenes backbone of the modern internet. Storage costs plummeted, and early movers like Amazon turned that shift into a massive, industry-defining business. Giant data centers continued to proliferate, but instead of tech companies relying on a mix of in-house on-premise servers and rented data center rack space, they offloaded nearly all their computing needs to distributed virtualized cloud environments. If that sounds confusing, you are not alone: back in the mid-2010s, one perfectly intelligent relative of mine asked, “What even is the cloud, and why am I paying for 17 different subscriptions to it?”
All the while, tech companies were hoovering up petabytes upon petabytes of data—data that people voluntarily shared online, in enterprise workspaces, and through everyday mobile apps. Firms developed new methods to mine and structure this “Big Data,” and promised it would revolutionize daily life. In many ways, it delivered on that promise. And it was always clear where this trajectory was headed.
Today, the tech industry is in the feverish boom period of generative AI, a technology that demands unprecedented levels of computing power. Big Data is old news; the new race is to build massive, AI-wired data centers at scale. Powering these facilities requires faster, more efficient chips than ever before, and chipmakers like Nvidia and AMD have been falling over themselves to tout their AI-ready offerings. The sector has entered an unprecedented era of massive capital investment in AI infrastructure, a boom large enough to pull U.S. GDP into positive growth territory. These are sprawling, complex deals that often amount to little more than high-powered handshakes at industry cocktail parties, greased by gigawatts of promised energy capacity and unbridled investor exuberance—while the rest of us struggle to keep track of actual binding contracts and total dollars changing hands.
OpenAI, Microsoft, Nvidia, Oracle, and SoftBank have closed some of the biggest deals in this space. Earlier this year, Stargate, an existing supercomputing collaboration between OpenAI and Microsoft, became the foundation for a massive national AI infrastructure project across the U.S. Former President Donald Trump called it the largest AI infrastructure project in history—on-brand for him, but that claim may not actually be hyperbole. Altman, Ellison, and SoftBank CEO Masayoshi Son all joined the initial deal, committing $100 billion upfront, with plans to ramp total investment up to $500 billion over the coming years, with Nvidia GPUs deployed across the network. Then in July, OpenAI and Oracle announced an expanded Stargate partnership—with SoftBank curiously absent from the announcement—touted to have a total planned energy capacity of 4.5 gigawatts and expected to create roughly 100,000 jobs.
Microsoft, Amazon, and Meta have also unveiled plans for multi-billion-dollar data infrastructure projects of their own. At the start of 2025, Microsoft said it was on track to invest “approximately $80 billion to build out AI-enabled data centers to train AI models and deploy AI and cloud-based applications around the world.”
This past September, Nvidia announced it would invest up to $100 billion in OpenAI—on the condition that OpenAI follows through on a deal to use up to 10 gigawatts of Nvidia’s systems for its infrastructure plans. Essentially, OpenAI has to pay Nvidia for hardware to qualify for Nvidia’s investment in OpenAI. The following month, AMD struck a similar deal: it would give OpenAI an equity stake worth up to 10 percent of the chipmaker, if OpenAI purchases and deploys up to 6 gigawatts of AMD GPUs between now and 2030.
It is this circular, self-reinforcing structure of these investments that has left the general public and bearish analysts wondering if we are barreling toward an AI bubble that will soon burst.
Instagram content
What is clear is that the near-term downstream impacts of this wave of data center construction are already very real. AI infrastructure places enormous demands on energy, natural resources, and local labor. As WIRED has reported, by some estimates, global AI energy demand will surpass that of Bitcoin mining by the end of this year. Data center processors run extremely hot and require constant cooling, so big tech companies draw heavily on local municipal water supplies—and they do not always disclose exactly how much water they consume. As a result, local residential wells have run dry in many areas, or test as unsafe for drinking. Residents living near new data center construction sites have also reported rising traffic delays, and in many regions, a sharp increase in car crashes. One area of Richland Parish, Louisiana, home to Meta’s $27 billion Hyperion data center, has seen a 600 percent spike in motor vehicle crashes just this year.
Leading AI proponents argue all these tradeoffs will be worth the long-term payoff. Very few top tech executives will publicly entertain the idea that this boom has overshot demand, either ecologically or economically. “Emphatically … no,” AMD CEO Lisa Su said earlier this month, when asked if AI hype had gotten out of hand. Like other industry leaders, Su cited overwhelming unmet demand for AI to justify these enormous capital outlays.
But demand from whom? That is far harder to pin down than leaders let on. In their view, demand is universal: it comes for all of us, including the 800 million people who use ChatGPT every week. The progression from 1990s purpose-built data centers, to 2000s cloud computing, to today’s AI-optimized data centers is not just a steady linear evolution of technology. It parallels the internet’s own growth: from the small, niche early web, to the mass consumer internet, to today’s AI-native internet. And realistically speaking, there is no turning back now. Generative AI is out of the bottle. On that core point, Altman, Huang, Ellison, Su, and the rest of the tech world’s top leaders are not wrong.
That does not mean they are right about the underlying math, though. Not about their optimistic economic projections, not about their claims for AI-driven productivity growth and its impact on the labor market, not about whether there are enough natural and raw material resources to complete all these planned projects, not about whether there will actually be enough demand to fill all this capacity once it is built, and not about the timing of the sector’s long-term growth. After all: even Rome eventually collapsed.
