We are all talking about data centers, but nobody really knows what they are. We hear about them consuming large amounts of energy, driving up electricity bills, and somehow being the engines of the AI revolution. But what is exactly going on inside these large facilities? And why are tech companies such as META, Microsoft, Open AI competing to build structures that could cover Manhattan, consuming enough electricity to power millions of American homes?
Let’s break it down with the most ambitious data center projects ever planned and the staggering numbers behind them.
What actually is a data center?
A data center is a very simple infrastructure: a facility with thousands computer servers, storage systems, and networking equipment organized in racks and rows. But modern hyperscale data centers are essentially large supercomputers, industrial facilities where hundreds of thousands of processors work in a synchronized way to train AI models, run cloud services, and power the digital infrastructure which we are all increasingly depending on.
The key components: Servers containing CPUs to handle computational tasks, with GPUs accounting for roughly 60% of total electricity consumption in AI-focused facilities. Cooling systems prevent the equipment from overheating, consuming 7% to 30% of total power. Some say that number can be as high as 45%. Power infrastructure includes uninterruptible supplies and backup generators ensuring continuous operation because when you’re training a billion-dollar AI model, power interruptions cannot happen.
The age of the gigawatt cluster
Colossus: Speed at an unprecedented scale
When Elon Musk’s xAI announced plans to build the world’s largest AI training cluster in Memphis, Tennessee, most experts predicted it would take 18 to 24 months. Instead, xAI accomplished it in 122 days. That is how Elon Musk gets stuff done.
Colossus 1 launched in September 2024 with 100,000 Nvidia H100 GPUs, then doubled to 200,000 GPUs in just 92 additional days—built from scratch in a refurbished Electrolux factory and operational since September 2024. But that was just the beginning.
Colossus 2, announced in March 2025, aims for over 1 gigawatt of power capacity and 1 million GPUs. For perspective, 1 gigawatt is enough electricity to power approximately 700,000 homes. When fully operational, it would become the world’s largest data center by far. The company is seeking $12 billion in debt financing, with the first batch of 550,000 chips scheduled to come online in early 2026.
OpenAI’s stargate: The $500 billion moonshot
If Colossus is ambitious, OpenAI’s Stargate project is unheard of. Announced at the White House in January 2025, Stargate is a joint venture between OpenAI, SoftBank, Oracle, and MGX planning to invest $500 billion over four years to build AI infrastructure exclusively for OpenAI.
The scale is breathtaking: 10 gigawatts of total capacity, enough to power roughly 7 million homes. The flagship Abilene, in West Texas campus came online in September 2025 with potential to scale past 1 gigawatt. By October 2025, Stargate had expanded to eight sites across Texas, New Mexico, Ohio, Wisconsin, and Michigan with nearly 7 gigawatts of planned capacity and more than $400 billion of committed investment, putting the project ahead of schedule.
Virginia: Data Center Alley
While mega-projects capture headlines, Northern Virginia remains the undisputed capital of the data center world. Often called “Data Center Alley,” the region hosts over 300 data centers with nearly 4,000 megawatts of commissioned power, representing approximately 15% of global hyperscale capacity.
Amazon Web Services, Microsoft Azure, Google Cloud, and Meta all maintain massive infrastructure in Northern Virginia. Virginia’s data center power demand is projected to reach 12.1 gigawatts by the end of 2025, up from 9.3 GW in 2024. When you stream Netflix, ask ChatGPT to find a discount for your favorite brand, or scroll Instagram, it is very likely that your data passed through Northern Virginia.
The hyperscaler arms race
Meta is perhaps the most aggressive of any hyperscaler, announcing plans to invest over $600 billion in U.S. infrastructure by 2028—including $60-65 billion in 2025 alone, that is more than double its 2024 spending.
Key projects include Prometheus in New Albany, Ohio (1-gigawatt AI supercluster expected online in 2026) and Hyperion in Richland Parish, Louisiana, a planned 5-gigawatt data center with a footprint large enough to cover most of Manhattan. The Louisiana facility will bring 2 GW online by 2030, scaling to 5 GW in subsequent years, enough to power about 3.5 million homes. By the end of 2025 or early 2026, Meta will operate more than 1.3 million GPUs, the largest fleet of AI processors in the world.
Microsoft invested $3.3 billion in a 315-acre Wisconsin campus, while Google invested $9 billion in AI-ready facilities across Virginia in 2025, securing a place among the top three hyperscalers globally. The three major cloud providers, AWS, Microsoft Azure, and Google Cloud, combined account for 58% of all global hyperscale capacity.
The energy crisis nobody saw coming
Here’s the bad news: data centers consumed approximately 415 terawatt-hours (TWh) in 2024, about 1.5% of global electricity use. This figure is projected to reach 945 TWh by 2030, more than doubling in six years.
In the United States, data centers used 176 TWh in 2023, representing 4.4% of U.S. electricity consumption. By 2028, projections suggest this could reach 6.7% to 12% of total U.S. demand. In Ireland, data centers already consume over 20% of the country’s total electricity.
AI’s role is critical: AI-optimized servers consumed 93 TWh in 2025 (21% of data center power). By 2030, they’re expected to consume 432 TWh, accounting for 44% of all data center electricity.
The rapid expansion has triggered environmental concerns around water consumption, emissions, and grid strain. Data center construction timelines (months) vastly outpace grid expansion timelines (years), creating local infrastructure crises.
The industry is responding: AWS aims for 100% renewable energy by 2025, Meta targets net-zero emissions by 2030, and operators are exploring clean on-site power alternatives including green hydrogen, geothermal, and small modular nuclear reactors.
The bottom line
Data centers are no longer invisible infrastructure. They’ve become the physical image of the AI wave ie, massive industrial facilities consuming huge amounts of electricity to train models that can write code, generate images, and answer questions in any language.
The numbers are almost incomprehensible: millions of GPUs, hundreds of billions in capital expenditure, gigawatts of power consumption. But they represent something fundamental: the recognition that artificial intelligence requires physical infrastructure at a scale humanity has rarely attempted.
When Elon Musk built Colossus 1 in 122 days, he demonstrated that the constraints we thought were immutable were actually just conventions waiting to be challenged. When Meta announced a 5-gigawatt facility and OpenAI committed $500 billion to Stargate, these weren’t hyperbolic announcements—they were statements about the computational requirements of superintelligence.
These are not data centers in any traditional sense. They’re the “cathedrals” of the information age, monuments to humanity’s belief that intelligence can be engineered, scaled, and deployed to transform civilization. The future doesn’t wait for permission. It demands electricity.