Show Up With Electrons
Dividends | May 2026
Somewhere in the South, a company needed industrial sized generators. The lead times its suppliers quoted ran past eighteen months. The company solved the problem the way anyone would. It bought a factory.
The factory was idled, located several states away, and only notable for the fact that it had generators. The company purchased the building, hired trucks, removed the generators, drove them south, and installed them at the new site. Each of those verbs was someone else’s livelihood, and each of the verbs that followed (the wiring, the foundation, the framing, the pipe and duct) was someone else’s again. The factory that was bought now sits stripped of the equipment that made it useful, no plans for what comes next. Somewhere up north there is a lot full of weeds where industrial production used to happen. Somewhere down south there is industrial production happening.
A country that strip-mines one industrial site to feed another is doing something other than reindustrializing. It is improvising under constraint. The dysfunction is real. So is the demand it creates. This is where the investor’s wheels start turning, and somewhere Ice-T starts humming “High Rollers.â€
The numbers describe a grid no longer matched to its load. Order books at the major gas turbine OEMs, GE Vernova, Siemens Energy, and Mitsubishi Power, now stretch to 2029, putting new equipment effectively four years out. Turbine prices have nearly tripled since 2019. Every AI data center, every new gas plant, and every energy security project on the planet is queuing for the same machine. Load forecasts at the major grid operators have been revised upward by multiples over the past two years, after a quarter century in which planners assumed flat demand growth was the steady state. More than two thousand gigawatts of proposed generation are now sitting in interconnection queues nationwide, a backlog that exceeds the country’s existing operating fleet. The grid the country has is not the grid the country is asking for, and the gap is widening faster than the institutions that maintain the grid can close it.
This is also a different problem than it was eighteen months ago. The earlier conversation about industrial power constraints centered on transformers and switchgear, the equipment that connects generation to distribution. Distribution can be debottlenecked; capital and time will eventually catch up. The current binding constraint is generation itself. Generation cannot be solved with capital and time alone, because generation sits downstream of permitting, regulatory cycles, technology choice, and a political process that has not produced a coherent answer in twenty years.
Demand for power, meanwhile, is moving in only one direction, and it is structural rather than cyclical. Follow the labor. There is not enough of it, and what exists is not interested in factory work. The country has not stopped demanding the output, only the people who used to produce it. What fills the gap runs on electricity. The U.S. fertility rate has been below replacement for nearly two decades and now sits near 1.6, the lowest figure on record; the labor pool of 2030 was set by the births of roughly 2008 to 2010, and the math does not change. Layered on top of that demographic floor is a generation of cultural drift. Parents pushed children toward college and white-collar careers, and trades and apprenticeships eroded into curiosities. Even if demographics turned tomorrow, the pipeline that would have produced factory workers is not there to be turned on.
The substitution runs through compute. The country cannot conjure workers it did not produce, and retraining the workforce it has into manufacturing will take a generation. Compute reads the gauges, runs the inspection passes, predicts the maintenance, schedules the machining, and increasingly performs the cognitive work on the floor that fewer human operators are available to do. The substitution is not optional, and it is not temporary.
The optimistic counter, that a technology solution will become efficient enough to outrun its own power demand, is real and worth taking seriously. It is also unable to escape the basic physics. Every plausible solution to the labor question runs through electricity generated and delivered on this continent. The argument can be had about how many electrons, but not about whether or not they are needed.
At present, compute demand is running ahead of efficiency gains, and the gap is widening. Whether the curve bends through better models, smaller models, more efficient chips, more clever scheduling, or some recombination nobody has yet thought of is the central technical question of the moment. Until the curve bends, electrons are the binding input.
The current response is improvisation. Hyperscalers are spending tens of billions a year to outrun the constraint by brute force. They are bringing mothballed gas and coal plants back online to capture capacity that cannot be built fast enough through the queue. They are signing twenty-year contracts for nuclear capacity that has not yet been recommissioned. They are committing to small modular reactors that have not yet been built. Capital this large does not sit still in an inefficient state for long. The question is what bends the curve, and when, and which industries do not survive the wait.
The slow response of the system is the product of multiple factors operating at once, not one of them dispositive. Regulatory cycles run long. Utility capital expenditure behavior has, for thirty years, been calibrated for flat load growth and now confronts the opposite. Equipment supply chains for transformers and turbines and switchgear are concentrated in a handful of suppliers worldwide. Permitting for new generation moves at the speed of a paper pusher killing time in a government office waiting for retirement. Technology choices are unsettled. Together this describes a system under load that was not designed for what is now being asked of it.
Where the answer ultimately comes from is genuinely unknowable. Small modular reactors are a candidate, with both real progress and real timeline risk. Grid-scale storage is a candidate, increasingly economic but constrained by mineral supply and manufacturing capacity. Demand response and AI-driven grid optimization are candidates, with measurable but bounded gains. Distributed generation, on-site cogeneration, microgrids, and several technologies that are not yet on the radar are also candidates. We do not know which one, or ones, will win. Somewhere, almost certainly, some clever kid in a basement has already worked it out, and we will look back in five years and wonder how it was not obvious at the time. Five years is also about how long it takes for a working answer to walk through the regulatory, capital, and adoption gauntlets that stand between a solved problem and a deployed solution. Something will work.
The unknowability is also not the relevant question for the investor. The constrained system has already announced where the demand is, through the behavior of those willing to pay for it. Hyperscalers buying mothballed gas plants for the queue position. Hyperscalers signing nuclear restart deals for dispatchable baseload. Manufacturers buying factories for their generators. Operators paying premium for any path to power that does not run through the queue. These are the actions of a system willing to pay almost any price for an input it cannot get fast enough. The investor is not the cause of the dysfunction, but the owner of the businesses that meet the demand the dysfunction creates.
The investor question, then, is not which technology wins. It is what every plausible answer still requires. Every plausible answer still requires equipment that moves and transforms electrons. Every plausible answer still requires installation, service, and maintenance at industrial scale, along with fabrication and contracting. Every plausible answer still requires the industrial businesses that surround and support the buildout, regardless of whether the generation comes from gas, nuclear, solar plus storage, or something not yet on a slide deck.
That is the picks-and-shovels position at lower middle market scale. Generation itself is a different game played by different capital, with dollar amounts, technology risk, and time horizons that are the wrong fit for LMM investors. What LMM capital can own is the layer that serves, supplies, and surrounds the buildout, organized loosely into three categories. The equipment layer includes metal fabrication for transformer enclosures, substation components, and turbine housings, alongside valves, gauges, and process equipment for thermal, nuclear, and gas generation. The construction and contracting layer includes site work for new substations, transmission corridors, and data center campuses, and MEP contracting at industrial and commercial scale, where the demand for mechanical, electrical, and plumbing services is structural rather than cyclical. And as the data center clusters anchor new economic gravity in towns that did not have it before, the follow-on infrastructure layer rolls outward: the roads, the water, the housing, the schools, the commercial development that always follows industrial concentration. The buildout does not stop at the fence line. It creates a long tail of demand for businesses sized exactly for LMM ownership.
The cycle ahead will not be decided by who has the cheapest labor or the cleverest tariff arbitrage. It will be decided by who shows up with electrons. The companies that secure capacity early will be priced as scarce. The companies that cannot will be priced as exposed. Manufacturers in regions where load is rationed, or where industrial connections are deferred for years, are not in the same business as manufacturers who have power to sell back to the grid. Both will appear on the same line of the same NAICS code. The market is already learning to price them differently.
The weeds and the lost production are real. So are the paychecks. The deal that cannibalized one factory generated work for everyone who stood up the new site, and for the infrastructure around it. The companies that make new equipment like that are filling order books to 2029, and they are too large for LMM capital to participate in. But the generators in this story did not arrive at their new home unaided. Someone wired them in. Someone trucked them down. Someone built the road to the new site. Someone poured the foundation, and then fabricated the steel, and then framed the structure, and then ran the pipe, and ran the duct, and finished the floor. Once the lights came on, someone built the houses for the people who would work there, and the schools for their children, and laid the water and sewer that runs underneath all of it. Multiply this by every site, every plant, every queue project that finds its way to the grid. None of that requires picking the winner in the technology race. All of it requires capital deployed at LMM scale into the businesses that supply, install, fabricate, and service the layer on which the buildout depends on.
As the weeds grow around the factory up north, the generators that left are running somewhere else, doing the work the country needs done. The people who carried, wired, paved, framed, plumbed, and serviced what came next are getting paid. So are the people doing the same work everywhere else the buildout is happening. The investors worth watching are in that group, where the capital is the right size and the demand is the most durable. The picks and the shovels are already at work.
Robert Williams is the founder of 2nd Line Equity, a private equity firm focused on lower middle market industrial and field services businesses in the Southeastern United States.
Sources and Further Reading
Interconnection queue data: Lawrence Berkeley National Laboratory, “Queued Up†annual report on transmission interconnection requests.
U.S. installed generation capacity: U.S. Energy Information Administration (EIA).
U.S. fertility rate: Centers for Disease Control and Prevention, National Center for Health Statistics, Births: Final Data annual reports.
Gas turbine order books and pricing: industry coverage of OEM disclosures from GE Vernova, Siemens Energy, and Mitsubishi Power, 2024 to 2025.
Load growth forecasts: North American Electric Reliability Corporation (NERC) Long-Term Reliability Assessment; regional grid operator load forecasts.
Hyperscaler power agreements: public announcements including the Constellation and Microsoft Three Mile Island restart (September 2024), the Amazon and Talen Energy data center co-location agreement, the Amazon and X-Energy small modular reactor commitment, and the Google and Kairos Power small modular reactor agreement (October 2024).