A man in Manassas, Virginia opened his January bill: $281. The month before, it was near $100, same 40-year-old house. The cause sat down the road: Northern Virginia is the world’s data center capital. Machines training AI now consume power like a midsized city.
That bill sits at the small end of a $2 trillion problem. Here is who pays it, and who is racing to fix it.
Podcast – AI Data Centers Are Coming for Your Neighborhood, and Your Wallet
One AI Query vs. Your Toaster
The Numbers That Made Me Check Twice
The International Energy Agency projects data center demand will more than double by 2030. It will reach roughly 945 terawatt hours, more than Japan uses today.
In the U.S., data centers drive nearly half of all electricity growth through 2030. America will spend more electricity processing data than on all heavy manufacturing.
I checked these numbers twice, assuming I had misread a decimal somewhere. I had not; the figures were correct as reported.
One AI data center needs 300 to 500 megawatts of power. Over 3,000 data centers already operate in the U.S., with nearly 1,500 more planned.
Why Big Tech Went Nuclear
In September 2024, Microsoft signed a 20-year deal to restart Three Mile Island Unit One. Constellation renamed it the Crane Clean Energy Center, investing $1.6 billion. Microsoft buys all 835 megawatts, with first power expected in 2027.
Every hyperscaler followed the same nuclear playbook. Meta committed more than 6 gigawatts across several nuclear partners. Amazon backed X-energy and locked in 1.92 gigawatts from Susquehanna. Google signed the first corporate deal to buy small-reactor power.
Together, these companies signed 13 nuclear deals worth 9.8 gigawatts combined. That adds up to more than $50 billion, aimed at powering AI.
Restarting old plants skips a decade; new reactors promise three to five year builds.
The U.S. interconnection queue, the line to plug in new projects, tops 2,600 gigawatts. The average wait runs five years, and roughly 80% of projects withdraw.
The richest companies in history built their own power plants instead of waiting. The people solving AI’s energy problem are the people who created it.
The Second Story Showing Up at Your House
Regular households are feeling this squeeze too. PJM’s monitor found data centers drove 70% of Illinois’s cost spike last year. ComEd customers face bill increases of 12% or more in 2026.
Community pushback followed fast, spreading beyond Illinois. Seven in ten Americans oppose new data centers in their own neighborhoods. This resistance blocked or delayed 75 projects worth $130 billion in Q1.
I want to be honest: I run my business on these AI models. Every query adds to that demand curve, so I’m no bystander.
Does this buildout end with cheap energy for everyone, or with families subsidizing server farms? Both outcomes remain entirely possible right now.
The Quantum Question, Answered Honestly
Some readers ask if quantum computing solves this, since the space attracts heavy hype.
D-Wave’s quantum computer simulated a magnetic material in twenty minutes. The classical equivalent needs a million years, burning more electricity than the world uses yearly.
That comparison holds up, but so does its limitation.
A quantum computer cannot run ChatGPT or train language models. These machines work as hybrid accelerators beside classical supercomputers. Analysts agree quantum won’t displace AI’s data centers this decade.
What quantum can do is narrower: brutal workloads in simulation and materials science. Then it cuts their energy cost by orders of magnitude. I might be wrong; this field has missed deadlines before.
The Recap
Near term, restarted nuclear plants carry most of the load. The wealthiest companies on Earth are paying, because the grid stalled. Long term, quantum efficiency, fusion, and smarter AI may take over.
We are watching big tech become big energy in real time. In ten years, that sentence will sound obvious.
Intelligence abundance requires energy abundance, full stop. There is no machine-powered future that skips the power bill.
