Across the defense and infrastructure conversations I keep having lately, one question comes up more than any other: is the data center boom actually building the future, or is it building the next generation of stranded assets? It's a fair question, because right now the United States has roughly 4,000 data centers running and another 3,000 under construction or planned — and almost none of that build-out is being planned against the possibility that the underlying compute technology itself is about to change shape, not just scale up.
The Water Story Is Smaller Than It Looks
Start with the controversy that gets the headlines: water. It's real, but it's smaller than the debate suggests. U.S. data centers directly used around 17 billion gallons of water in 2023 for cooling — about 0.3% of the country's water supply. For comparison, California's almond industry alone uses somewhere between 70 and 90 times that volume every year. Most of the eye-catching “data centers are draining our water” figures actually bundle in the indirect water footprint of the electricity that powers them — closer to 211 billion gallons — which is a power problem wearing a water costume. Some individual facilities do draw up to 5 million gallons a day for cooling, and that's a legitimate local concern in drought-stressed regions, but treating it as the central issue misses where the real pressure is building.
The Real Bottleneck Is the Grid
The actual bottleneck is the grid. Data centers consumed about 1.5% of global electricity in 2024; Berkeley Lab projects U.S. data centers alone could hit somewhere between 9.5% and 15.3% of American electricity by 2030. For two decades, from roughly 2005 to 2025, U.S. electricity demand was essentially flat, and utilities built permitting rules, retirement schedules, and pricing structures around that assumption — including retiring coal and nuclear capacity that isn't quick to replace. Data centers didn't create that structural fragility; they exposed it, arriving right as demand finally started climbing again. That's why 70% of Americans now say they oppose new data center construction near them — not because of almonds and swimming pools, but because they're watching their own electricity rates move for the first time in a generation.
The Part Nobody's Planning For
Here's where it gets genuinely uncomfortable, and where the Defence-and-Space angle actually bites: none of this build-out assumes the compute itself might shrink. Researchers at CU Boulder and Sandia National Labs recently demonstrated a microchip-scale optical phase modulator — a component central to controlling qubits — that is almost 100 times smaller than a human hair and uses roughly 80 times less microwave power than the commercial equivalent. It's manufactured on standard CMOS lines, the same fabrication process behind every smartphone chip, which means it can be mass-produced rather than hand-assembled on an optics bench. One of the researchers put it bluntly: you're not going to build a quantum computer out of a hundred thousand bulk optical modulators sitting in a warehouse. The whole point of the breakthrough is fitting the control hardware for a million-qubit system onto something the size of ordinary electronics — not a room full of racks, not a hall full of water-cooled GPUs, but hardware that increasingly looks like it belongs on a bench, not in a football-field-sized campus.
That's the white-elephant question I want to put in front of you directly. Today's hyperscale data centers are being financed, permitted, and built as 15-to-20-year assets on the assumption that dense compute will always mean acres of water-cooled racks pulling gigawatts off the grid. If a meaningful share of tomorrow's highest-value compute — the workloads that actually justify defense, intelligence, and space applications — ends up running on quantum or quantum-hybrid hardware that fits in a rack, or eventually a chassis, drawing a fraction of the power and none of the water, then some portion of what's being poured into the ground right now isn't infrastructure. It's a bet that the compute paradigm holds still for two more decades, at exactly the moment the physics says it might not.
I'm not arguing the classical AI buildout is a mistake — the demand for today's large models is real, and the grid strain is real regardless of what quantum computing eventually does. But planning only for more of the same, when the leading edge of quantum hardware is explicitly racing toward room-to-chip miniaturization, is planning with one eye closed.
What I'd Ask You
If a quantum-accelerated chip the size of a thumbnail can eventually do the job that today needs a data hall pulling five million gallons of water and enough electricity to strain a regional grid, are the hyperscale campuses breaking ground in your backyard right now durable infrastructure — or future white elephants? And if you're the one approving the capital for the next campus, is anyone in the room asking that question, or just the one about how fast it can get built?
#DataCenters #QuantumComputing #AIInfrastructure #DefenseTech #SpaceTech #EnergyPolicy #Sustainability #AI
