A 120-foot mock reactor in the Virginia hills sat silent for a decade, and two veterans schooled in rocket deadlines revived it to chase the data center power crisis bearing down on American grids
Eight stories of steel, pipe and vessel rise inside a facility in the Virginia hills, running on ordinary electricity, not uranium.
Then the money ran out.
The whole machine went silent on a Lynchburg campus for roughly ten idle years.
Readers tracking the fuel side of this revival can follow how advanced fuel platforms are entering federal programs to supply the next wave of reactors.
Now two engineers schooled under rocket deadline pressure have decided that a $400 million ghost is exactly the head start they need.
What does a dormant nuclear test rig have to do with the power crisis bearing down on American data centers?
How a machine built to teach turned into a machine built to sell
The Lynchburg structure was conceived as an electrically powered replica of a nuclear plant, a device that generates no fission heat of its own but instead mimics the full plumbing and instrumentation of a reactor so engineers can safely monitor and modify the systems around it. The underlying design, known as mPower, emerged during a period of high optimism about small modular reactors, when the argument was that factory built units could replace massive custom construction and bring nuclear power within reach of smaller grids.
That optimism hit a wall. As the energy economy shifted, the project collapsed and the 120-foot machine was left for dead, drawing no power, hosting no engineers, going nowhere. What kept the machine valuable even in silence was the engineering mass it represented.
Every pipe, sensor and valve had been specified, procured, installed and documented to nuclear grade standards, a level of material and regulatory work that cannot be replicated cheaply or quickly. Walking away from it was not throwing away a prototype; it was walking away from a small city’s worth of verified hardware.
Where two rocket engineers saw a shortcut nobody else had taken
With the AI buildout exploding across the country, a pair of engineering veterans saw in the dormant Lynchburg rig a solution to one of the country’s most pressing challenges: powering the proliferation of data centers. Their startup is built around a blunt premise: the hardest part of small modular reactor development is not the physics, it is the years of iterative hardware testing that must happen before any regulator will certify a design.
The naval manufacturer that owns the mPower rig brings a supply chain, a workforce cleared to handle nuclear components, and decades of quality records that a pure startup cannot conjure from scratch.
Ben Kellie, the co-founder who came to nuclear from rocket engineering, put the logic plainly while standing inside the Lynchburg facility. “We’re standing on the shoulders of giants here, right?” he said. “We’re starting with something that had $400 million in investment put into it, and significant time and effort. That’s a leg up.“
What the record shows about the design itself
The mPower concept was designed as a pressurized water reactor, the most common type in commercial service today, but scaled to roughly 195 MW per unit, small enough to be factory built and transported by rail. The founders argue that nuclear engineers need to harness technology already developed rather than inventing a new coolant or fuel form from scratch, a deliberate hedge against the licensing risk that has slowed more exotic designs.
The Lynchburg test rig was built to replicate full scale reactor conditions without radioactive material, so every safety system and control logic sequence could be exercised at operating temperature and pressure. Engineers could deliberately trip alarms, block valves and simulate loss of coolant scenarios in ways never permitted inside an operating plant.
The naval manufacturer’s involvement also brings a credential that matters specifically for data center customers. Submarine reactors have operated at sea for decades without a commercial scale accident, a safety record that carries real weight in boardrooms explaining an on-site power plant to insurers.
The catch that the decade of silence left behind
A machine that sat idle for ten years is not a machine that can simply be switched back on. Valves corrode, seals shrink, and control software written against hardware generations that no longer exist becomes a project in its own right. The recommissioning work takes money and calendar time that corporate imaginations compress but engineering reality does not.
The startup has not yet filed a design certification application, and the gap between a convincing test rig and a licensed commercial unit is measured in years and hundreds of millions of dollars. There is also the question of whether data center operators will wait, since hyperscale power agreements are being signed right now, many pointing to gas turbines and battery storage because those technologies deliver electrons inside a planning cycle that nuclear still cannot match. For a closer look at how another small reactor cleared its own milestone, see how regulators cleared a Michigan site that itself sat idle for years before finding a second purpose.
What an eight-story ghost reactor says about the next decade of nuclear
The Lynchburg story is unusual not because a startup is chasing data center power, but because it found its test asset already built, already documented and already inside the supply chain of a company that has been making reactor cores for the Navy since the Eisenhower administration. That combination almost never exists in the small reactor world, where founders typically begin with a simulation and spend a decade reaching a first physical demonstration.
Stranded nuclear infrastructure from the last wave of optimism is quietly becoming a resource. Decommissioned plants have already provided licensed sites and trained workforces to new owners, as the Michigan example shows, and an idle test rig in Virginia follows the same logic with a different kind of asset.
Whether the mPower rig can become a certified commercial product before the data center window closes remains genuinely open. The machine is real, the prior investment is real, and the demand signal from hyperscale computing is real. Industry analysts estimate that data center construction alone could require more than 300 GW of new power capacity in the United States by 2035, according to the startup’s licensing announcement. Readers tracking the fuel side of this revival can follow how advanced fuel platforms are entering federal programs to supply the next wave of reactors.
Hugo is an engineer with strong technical expertise. Multilingual from an early age, his writing combines technical clarity with a strong interest in science and energy.
