The Four-Hundred-Megawatt Ghost
A story about a company that had the answer in its own catalog and kept reaching past it
Herbert Roberts, P.E.
Part one of two.
Ten years ago I worked at an energy production manufacturer, and we kept trying to build a machine nobody wanted.
We had two good gas turbines. The mid unit was the 200-to-300-megawatt class — a regional machine. You didn’t sell one; you sold three or four into a region, and the fleet gave you something no single unit could: when one came down for service, the others carried the load. The customer never went dark. Redundancy wasn’t a feature we bolted on. It came free with the arithmetic of buying more than one.
The large unit was the 500-plus-megawatt class, aimed at the handful of places on earth big enough to swallow it whole. Mexico City. New York. Tokyo. A magnificent machine with a small address book, because there are only so many cities that fit that number.
And then there was the ghost. Somewhere between those two classes sat a 400-megawatt hole, and we could not stop trying to fill it with a single machine. The plan that kept coming back was to take the mid unit and push three-dimensional aerodynamics into the compressor — squeeze more efficiency out of it and lift one unit to 400 megawatts. We ran the iterations. Three different design efforts, as I remember it. We got right up to the edge of a prototype. And every time, marketing came back with the same sentence: nobody wants to buy it.
Here is why they didn’t, and it is the whole lesson in one configuration. The regional sites that wanted roughly 400 megawatts already knew how they wanted to buy it: two of the mid units feeding one steam turbine — a 2-on-1 combined-cycle block. That arrangement won on both axes at once. It kept the redundancy, because a site running two machines never loses everything when one comes down for service. And it captured the co-generation, because the exhaust heat off two gas turbines drove the steam unit for output and efficiency the single machine couldn’t touch. The customer had already engineered the better answer out of the fleet parts. Our monolithic 400 offered the same nameplate and forfeited both advantages — no redundancy, no co-gen fit, one throat for the whole site to choke on. The ghost wasn’t losing to nothing. It was losing to a smarter version of itself that the customer had already drawn up.
And it had nowhere else to go, because the large buyers scaled the same way. Nobody who bought the large unit bought just one either — they paired two on a shared steam plant, and when a site needed more, it added a third large unit and a second co-gen train. And the reason they stayed on the same machine wasn’t sentiment — it was interchangeable parts. A site standardized on one unit runs a single spare-parts inventory that fits every machine on the property: one set of blades, one set of hot-section components, one stock of consumables, any part swappable into any unit on any shift. That interchangeability is the whole economic point of a fleet — it collapses the inventory, shortens the outage, and lets the crews master one machine cold instead of splitting their attention across two. Drop a 400 onto that site and you break exactly that. Now you’re carrying a second, dissimilar parts inventory for a single orphan unit, a second set of procedures, a second thing that can strand you waiting on a part that fits nothing else on site — all to get less than you’d get by simply adding another large unit the crews already knew cold. So the ghost died from both directions at once. Paired and co-genned, a 400-class block overshot what a regional site could ever absorb. And on a large site it was the dissimilar orphan nobody wanted, losing to the obvious move of just buying more of what already worked. The mid unit scaled by adding mid units. The large unit scaled by adding large units. And the same parts logic that kept the large buyers from stepping down to the 400 kept the mid buyers from stepping up to it — or to a large unit at all. A mid-unit site was already sitting on years of spare-parts inventory, trained crews, and procedures built entirely around the mid machine. Stepping up meant walking away from all of it and standing up a second, dissimilar inventory from zero. So the mid buyer’s next purchase was almost always another mid unit, because the sunk investment in interchangeable parts made staying on the rung cheaper than climbing to a new one. The parts base wasn’t just an operating convenience — it was a switching cost, and it locked every buyer onto their own ladder in both directions. The 400 fit neither ladder — which is the entire reason it never found a customer, and the exact reason we kept mistaking that for a design problem we could iterate our way out of. It was never a design problem. It was a machine with no rung.
So we shut it down. There’d be a lull — maybe a year — and then, like clockwork, a new plan for the same ghost. The four-hundred-megawatt system that never had a customer, resurrected on a schedule.
I’ve thought about that cycle for a decade. It isn’t a story about a bad turbine. It’s a story about a company that had the answer in its own catalog and kept reaching past it.
The mistake had a shape
The mistake was that we were mesmerized by the big sale.
A large bespoke installation is a lump of revenue every ten years, hostage to the customer’s capital cycle, custom enough that the service business behind it never standardizes. The large unit was the purest expression of that instinct — build the biggest possible thing, sell it to the biggest possible customer, book the biggest possible number. And the addressable market was a dozen skylines.
The mid unit was quietly telling us the opposite. Its economics didn’t live in the sale; they lived in the fleet — identical units, identical parts, identical service intervals, spread across a region with redundancy built into the count. That’s not a razor. That’s a razor-and-blades business, and the blades are an annuity.
Here is the part that should have been unforgivable: the best razor-and-blades operation on the planet was sitting one division over. The company’s aviation division had turned the aftermarket into an art form — the engine nearly given away, the money made over decades of service agreements on a standardized installed fleet of thousands of near-identical units. Same parent company. Same finance culture. Same playbook available for the cost of a phone call down the hall. And the power side kept swinging for the monument.
The four-hundred-megawatt ghost was that instinct made visible. It wasn’t a market need. It was an institution that could only imagine selling a bigger thing, when the thing it already had was telling it to sell a system.
But the customers were already there
The ghost was the supply-side half of the failure — we couldn’t stop building a thing nobody wanted. The demand-side half was worse, because it wasn’t a misjudgment. It was a refusal.
I had family in a horse community, and they showed me the answer years before I had a name for it. A group of horse people networked with each other, pooled their money, and bought a large patch of land together. Then they built the commons. A shared area for working manure. Common pasture for the horses. And because some of the animals were trained as jumpers, shared grounds laid out for jump training. They didn’t build a subdivision. They built a community organized entirely around a shared resource, because the shared version was better and cheaper than every family going it alone.
I looked at that and saw a power customer.
Put a small generating unit — call it 75 to 150 megawatts — at the center of eight or nine of those large farms and you’ve got a buyer a gigawatt salesman can’t see, because the buyer isn’t a city. It’s a community. And once I started looking, the pattern was everywhere the map got thin. Fly-in aviation communities, where people already share a runway and a hangar culture. Housing developments going up across the West, where power was never as dense as it is on the East Coast and the grid was never a given. Every one of them a natural buyer for a mid-size, networked, easy-to-maintain machine — and every one of them beneath the notice of a team that measured opportunity in gigawatts.
That was the real prize, and it was the exact inverse of the ghost. Build a network of 75- and 150-megawatt units — simple, small, easy to service — and you don’t book one heroic sale. You put forty or fifty units into the field and you sell the service contract behind every one of them. Recurring money, for decades, on a fleet you can actually maintain — instead of one giant machine, a single onetime contract, and the headache of networking all that power to wherever it had to go.
I brought it up. More than once. The sales-and-service office sat in a different city from the design-and-manufacturing center — and it answered the same three words every time: big money, big money, big money. The sub-two-hundred-megawatt market wasn’t big money, so it wasn’t a market. We ceded it — handed whole segments to competitors we simply chose not to compete with, because serving those customers didn’t feel large enough to be worth serving.
And here is the part that should have made the sale to any executive who could see past the current quarter, because the small end doesn’t just scale — it scales in a way the big machines never could. A small-unit buyer grows with the community it serves: add a unit when the farms multiply, add another when the airpark fills in, add a third when the development builds out the next phase. The fleet spreads across wider and wider regions, and at some point the spread itself becomes the product. Because once you have small units scattered across a set of unassociated communities, you can stand up a shared spare-parts pool — one contrived inventory serving all of them, so no single community has to carry a full parts base of its own. Each community draws on the pool; the pool is cheaper for everyone than a shelf of spares nobody uses most of the year. That is a new service size and a new kind of contract: service and spares as the recurring product, sold across a distributed fleet of small buyers who individually could never justify it and collectively can’t live without it. New contracts, renewing forever, on a base that only grows.
It was the best business in the building, and it was invisible to the people who ran the building — because they were quarter-focused executives hunting immediate, huge wins. A pool of small recurring service contracts doesn’t book like a nine-figure turbine sale. It doesn’t land in one quarter, doesn’t photograph well in a board deck, doesn’t move the number this period. It just compounds, quietly, for twenty years. And a leadership measured on the immediate huge win is structurally blind to the compounding small one — not because the executives were foolish, but because the instrument they were paid against couldn’t see recurring, distributed revenue at all. Same mispricing as the ghost, one level up: the organization optimized for the size of the sale and expensed the durability of the business.
And the pooled-spares model hides one more advantage the quarterly view could never price: whoever controls the spares controls the upgrade path. When you own the parts pool, you don’t just service the fleet — you improve it, one component at a time, through the channel you already run. A better blade, a more efficient hot-section, a new control module enters the pool, and it propagates across the whole installed base at the normal cadence of scheduled maintenance. No forklift retrofit, no heroic capital campaign, no ripping out a plant. The upgrade rides the service contract you were fulfilling anyway. The aftermarket stops being a cost center and becomes the distribution network for every technology gain you make.
Which makes the small distributed fleet the best new-technology launch site in the entire business — for exactly the reason it was easy to dismiss. On a single large machine, trying something new is terrifying: one unit is the site, so a first-of-its-kind failure takes the whole plant down and every megawatt with it. On a fleet of small units backing each other up, the outage risk of any one machine is a rounding error. You can field a new blade in one unit, watch it for a season while the rest of the fleet carries the load undisturbed, and only feed it into the shared pool once it’s proven. The redundancy that made the fleet reliable also makes it fearless — it turns every site into a low-stakes test bed, and the pool into the mechanism that scales the winners. Small, distributed, and pooled isn’t just the resilient architecture or the durable business. It’s the fastest, safest way to advance the technology itself — which is the one thing an engineering company is supposedly for, and the one thing the immediate-huge-win instrument was structurally built to miss.
That is the finding the four-hundred-megawatt ghost was hiding. We built a phantom product that had no customer, and in the same years we ignored the real customers who had no product. One instrument, pointed both directions.
The instrument
Run every one of these decisions through a single forensic control. Call it the cutoff line: did you optimize for one number, and quietly pay the bill on every other axis? Hold the record against it and the same two readings come back.
Sale versus service. The monolith is a lumpy sale every decade. The fleet is an annuity by design. Mispriced: recurring revenue — the exact business the aviation division was printing next door.
Monument versus fleet. A bespoke gigawatt machine is a one-off; a serial-produced fleet gets cheaper with every copy. Mispriced: the manufacturing learning curve — the tenth unit costing half the first.
The verdict
Two divisions of one company. One instrument. One division built a fleet and got an annuity. The other chased the four-hundred-megawatt ghost, ignored the communities already showing us how to build a commons, and got a lull and a funeral, on repeat.
The ghost is still with us. It’s every argument that the answer must be bigger when the answer was always more — more identical units, teamed, serviced, and spread among the people who were always ready to buy them. We built a phantom with no customer and turned our backs on the customers with no product.
And the deepest cut isn’t that we misjudged a market. It’s that the small, distributed, pooled fleet was the better business on every axis the quarter couldn’t see — durable revenue, a captive upgrade channel, the safest test bed an engineering company could ask for. The instrument the division leaders were paid against couldn’t read any of it. They optimized for the size of the sale and expensed the durability of the business.
But blaming the division CEOs stops one level short of the finding. They weren’t blind — they were aimed. A division CEO chases the carrot the parent CEO hangs, and when the only carrot is quarterly dollars, every rational division leader does exactly what ours did: reach for the lumpy immediate sale that lands this period and ignore the compounding, distributed business that pays out over twenty years someone else will be sitting in the chair to collect. They weren’t failing at their jobs. They were succeeding at the job they were actually given, which was to move a single number by the end of the quarter. The ghost, the ceded market, the pooled-service business we never built — none of it was a failure of engineering or even of judgment. It was a failure of the metric at the top. Make quarterly dollars the only thing you measure, and you will reliably, quarter after quarter, kill the best long-term business in the building and never see the body. That is the instrument turned on the people holding it: optimize your leaders for one number, and they will pay the bill on every axis that number can’t see — including the survival of the company they run.
And notice what the wrong metric kills, because it is never just one thing. A patient business plan and a patient technology fail the identical test: both pay out past the horizon the number can see, so both look like pure cost today. Quarterly dollars can’t tell the difference between a slow-growth vision and a slow-maturing innovation — it wipes out both, indiscriminately, for the same reason. The pooled-service community fleet that would have compounded for twenty years, and the new blade or new cycle or new material that needs a decade of development before it earns a dollar, are the same kind of casualty. Anything whose return lands beyond this period reads as a line item to be cut. So the company keeps only what pays immediately — which is precisely the work that required no vision to begin with. Run an engineering company on the wrong metric long enough and it stops being an engineering company. It becomes a sales desk with a famous name, quietly euthanizing its own future on both fronts at once: the slow business and the slow invention, buried in the same unmarked grave, by the same number.
You can watch it happen to a whole division. The same company had a nuclear business — reactors and fuel, headquartered in a coastal Carolina town — and over these years it simply went quiet. It is worth being precise about why, because the easy version is wrong and the true version is worse. The metric didn’t kill it; the domestic market had already gone silent on its own, the new-reactor order book emptied out years earlier by cost and public fear. What the metric did was guarantee nothing would ever wake it back up. A dormant division under a quarterly-dollar regime gets managed for cash, not invested for a comeback — and a comeback was the one thing on offer, because reviving that business would have taken exactly the work the number can’t see. I’ll come back to what that revival would have required. For now, hold the shape: the division wasn’t killed. It was left in amber. Same grave as the ghost, quieter burial.
Part two takes this same instrument out of the gas-turbine yard and runs it across the whole energy transition — redundancy, peaking, the petroleum barrel, and the nuclear fuel cycle — to a place where the mistake we made with the ghost is about to be made again, or finally corrected.
Herbert Roberts, P.E. is a licensed professional engineer with 32 years in aviation research and development across two companies, and has spent eight years analyzing accidents for attorneys under his P.E. license.