Nuclear option: The only logical answer to AI energy crisis

.

On quiet, residential streets in Sterling, Virginia, residents often wake to the roar of diesel backup generators and smog that leaves their throats raw by morning — the price of AI data centers built without proper zoning oversight.

This is the leading edge of a much bigger problem. Virginia alone hosts roughly a quarter of the nation’s data centers, backed by some 10,500 diesel generators, many the size of semitrucks. Virginia Commonwealth University researchers who examined emissions permits for more than 130 Northern Virginia data centers found that running those generators less than an hour a week creates a public health burden equivalent to five large natural gas plants operating continuously — at least three premature deaths a year and 14,000 annual asthma episodes. And the AI buildout has barely begun.

The tech industry’s appetite for power is outpacing the grid. Energy Secretary Chris Wright has already authorized grid operators seven times in a single year to fire up diesel backups just to relieve strain. The choice we’re being offered is stark: computers go dark, or residents choke on pollution.

There’s a better option, a technological fix, hiding in plain sight: small modular nuclear reactors.

Why renewables can’t do the job

Data centers need power that is continuous, dense, and unfailingly reliable — they can’t adjust demand to match supply. Solar produces nothing at night, wind is intermittent, and batteries capable of bridging extended gaps are expensive, resource-intensive, and themselves generate toxic waste. Natural gas offers reliable baseload power but comes with emissions, and, as Virginia shows, diesel backup generators impose real health and quality of life costs on nearby communities.

Nuclear is the only source that’s simultaneously carbon-minimal, land-efficient, and capable of delivering steady power on demand. Massachusetts Institute of Technology nuclear engineering professor Jacopo Buongiorno’s life-cycle analyses, which account for construction, mining, operation, and waste, found that lifetime greenhouse emissions from nuclear are a fraction of those from coal, gas, and even solar, while requiring vastly less land and raw material than wind or solar for equivalent output. Nuclear waste volume is a tiny fraction of what solar and wind infrastructure generates. A lifetime of nuclear-powered electricity for one person produces about half an ounce of waste.

The safety record holds up, too: Even counting Chernobyl, coal is responsible for roughly 2,000 to 3,000 times more deaths than nuclear, oil 400 times more. We’ve tolerated those costs for generations while treating nuclear as uniquely dangerous — a judgment shaped by politics and propaganda rather than data.

Right-sized for the AI era

Traditional nuclear plants do have real drawbacks: costs approaching $10 billion and decade-plus construction timelines. Small modular reactors cut both roughly in half. Microreactors, the smallest class, generating 1 to 20 megawatts, can power hundreds to tens of thousands of homes from a small footprint, need refueling only once every 5-10 years, and can be built on assembly lines that standardize licensing and installation. Many are air-cooled, designed for rapid shutdown, and some can be delivered by truck or train.

More than 70 SMR projects are underway in the United States, backed by private capital and driven partly by tech-sector demand, and several companies have already been cleared to go critical. For a data center operator, placing a microreactor on-site means clean, continuous power without grid dependency or the legal exposure diesel generators now invite — building on the Navy’s 67-year record of incident-free shipboard reactors.

A political problem, not a technical one

If SMRs are so promising, why are Virginia suburbs still choking on diesel smoke? The obstacle isn’t engineering — it’s decades of regulatory inertia and a propaganda campaign that conflated civilian nuclear power with weapons and disaster. Serviceable plants have been shuttered under activist pressure — until recently, only four new U.S. reactors had been approved in 50 years.

Every year of delay is a policy choice, not a physical necessity — and, ironically, it’s the same progressive constituencies most vocal about climate change who have most fiercely blocked the one reliable, low-carbon power source capable of meeting AI’s scale.

AMERICA DOESN’T HAVE A POWER SHORTAGE. IT HAS A DELIVERY PROBLEM

Congress and the Nuclear Regulatory Commission should accelerate SMR licensing. States facing acute data-center power demand should prioritize SMR permitting with the same urgency they’ve shown in approving diesel generators. Data center operators, who have both the capital and the self-interest, should invest aggressively in SMR deployment. Delaware regulators recently rejected a data center’s permit over pollution from 516 proposed diesel generators. The right fix isn’t blocking data centers, it’s powering them differently.

Neighbors of data centers deserve a night’s sleep without diesel fumes. Northern Virginia’s children deserve air they can breathe safely. The nuclear option shouldn’t be a last resort — it should have been the first.

Henry I. Miller, a physician and molecular biologist, is the Glenn Swogger distinguished fellow at the Science Literacy Project. Andrew I. Fillat spent his career in technology venture capital and is a co-inventor of relational databases. They were undergraduates together at MIT and have no economic stake in their recommendations.

Related Content