The Turbine.
An interactive novel about the machine that built the modern world, and the bottleneck inside the AI buildout.
The question.
There is a question that nobody at General Electric, Siemens Energy, or Mitsubishi Heavy Industries asks out loud. Not because they don’t think about it. Because the answer is uncomfortable. What would it take to be able to design a turbine once, and have it work?
A turbine is, by quiet consensus, the most strategically important industrial device on earth. It converts about 80 percent of the world’s electricity. It propels the airliners and the warships. It backstops the grids that data centers and hospitals and cities rely on every minute of every day. And it is built, today, by a closed cartel of three companies that have been refining the same physics for a century.
Inside those three companies sits something close to twenty billion dollars of accumulated R&D hours. Most of it spent on iteration. Test rigs, blade redesigns, casting reworks, recalibration of cooling channels by tenths of a millimeter. The output is extraordinary. The process is not.
Dylan Morris, who writes a weekly Substack from El Segundo called Turbine Tuesdays, put the question this way: “The question these companies fail to ask is what would it take to be able to design once.”
This novel is the unpacking of that sentence. Why the question matters. Why nobody has answered it. And why a team in a low-slung industrial building two miles from the LAX runway is the most interesting attempt anyone has made in forty years.
“The question these companies fail to ask: what would it take to be able to design once? ”
First principles.
Before we can talk about the bottleneck, we need to talk about the machine. And before we can talk about the machine, we need to talk about energy itself, because the machine is, fundamentally, a conversion device.
Every turbine on earth runs the same ladder. Thermal energy at the top. Electrical energy at the bottom. The rungs are fixed by physics. The engineering is the climb.
At the top of the ladder is heat. You produce it by burning something, splitting something, concentrating sunlight, or routing geothermal flow. The thermal source defines the upper bound of your efficiency through the Carnot inequality. There is no negotiating with Carnot.
Then the fluid expands. The expansion does work on a set of carefully shaped blades. The blades spin a shaft. The shaft drives a generator. Each step has a physical bound. Each step also has a margin of engineering loss that good design can compress.
1 minus T-cold over T-hot. The hard ceiling.
Profile, secondary, tip leakage. 88 to 94 percent stage efficiency.
Bearing losses, disc friction. About 1 to 2 percent off.
Copper, iron, stray. 98 to 99 percent in modern generators.
“The Carnot bound says the best you can ever do is one minus T-cold over T-hot. Every honest turbine engineer carries that number in their head.”
Anatomy of the machine.
On the floor of GE Vernova’s Greenville, South Carolina factory, a single H-class gas turbine takes up the length of a basketball court. It weighs about 880,000 pounds. It contains approximately 35,000 components, several thousand of which are individually serialized parts that must be tracked from supplier to install over a lead time that now reliably exceeds five years.
Air enters through the inlet at the cool end. It is compressed across seventeen stages, going from atmospheric to roughly 22 bar, heating along the way through pure work input. It enters the combustor, where natural gas is sprayed in and burned at a turbine inlet temperature now reaching 1600 degrees Celsius in the newest H and J class machines.
The combustion products expand through four turbine stages, each made of single-crystal nickel superalloy blades coated with a thermal barrier and pierced with film-cooling holes that bleed compressor air through the blade’s leading edge. The shaft spins at 3,600 RPM. It drives a generator. The exhaust, still hot, can be routed into a heat-recovery steam generator to drive a second cycle.
Click any zone below to see what lives inside.
Four stages of single-crystal nickel-superalloy blades. Each blade is a serialized, traceable part. Thermal barrier coating plus film cooling. The bottleneck inside the bottleneck.
The two cycles.
There are only two thermodynamic cycles that matter at the gigawatt scale. Brayton and Rankine. Everything else is a footnote.
Brayton is the cycle of the gas turbine. Air in, fuel burned, hot gas expanded, exhaust out. It runs open. There is no phase change. Modern industrial Brayton machines push turbine inlet temperatures to 1600 Celsius, achieving 42 percent thermal efficiency simple-cycle.
Rankine is the cycle of steam. Water is pumped, heated until it boils, expanded through a steam turbine, condensed back to liquid. The phase change is what gives Rankine its compactness and its limits. Modern supercritical units run at 600 Celsius and 250 bar, hitting 47 percent on coal, 33 percent on nuclear with its lower temperature constraint.
The trick that wins is to run both. Burn the gas in a Brayton cycle. Catch the still-hot exhaust in a heat recovery steam generator. Use that steam to drive a bottoming Rankine cycle. The combined arrangement reaches 63 to 64 percent. It is the most efficient practical heat engine ever built.
Adjust the slider below to see how cycle, temperature, and efficiency move together.
The bill of materials.
An H-class gas turbine sells for about $240 million bare. Add the balance of plant, the heat recovery steam generator, the steam turbine, the cooling system, the electrical interconnect, and the bill walks to between $440 and $640 million per train. Multiply by two or three for a typical combined-cycle station.
But the cost is not evenly distributed. A quarter of the cost lives inside the hot section, where blades are cast as single crystals from nickel superalloy ingots and finished with thermal barrier coatings to tolerances that demand atomic-scale process control. The hot section also has the longest lead time. Thirty-six months, on a good day.
The chart below shows where the money goes and how long each piece takes to arrive.
“There is something like twenty billion dollars of R&D hours sitting on the balance sheets of the world leading turbine developers. Millions of engineer-hours, most of which wasted through these iteration loops.”
The value chain.
The turbine value chain has five tiers. Raw materials at the bottom. Components above. The OEM frame builders in the middle. The engineering, procurement, and construction firms that assemble plants. And, at the top, the operating and aftermarket services business that keeps the turbines running for thirty years.
Dollar share moves down the stack. Margin share moves up. The largest dollar pool is plant construction, the lowest-margin step. The smallest dollar pool is aftermarket service, the highest-margin step. The OEMs sit in the middle and keep an outsize fraction of total industry profit because they control both the frame and the aftermarket flow.
Anyone trying to enter this industry has to choose where to play.
The bottleneck.
In May 2025, every single one of the three companies that manufactures the world’s heavy-duty gas turbines simultaneously ran out of factory slots through 2030. GE Vernova’s backlog passed 80 gigawatts. Siemens Energy and Mitsubishi were not far behind. Pricing power flipped to the seller, and the customer line started to include not just utilities but hyperscalers paying cash for capacity that used to be allocated by relationship.
The Big Three together control roughly 77 percent of the heavy-duty market. The remaining quarter is split among Ansaldo Energia, Doosan, and a long tail of smaller players. Concentration at this level is what makes pricing decisive. There is no second bid.
Three things converged. The U.S. coal fleet retired faster than gas capacity was added, shrinking the reserve margin. Renewables grew, but the firming requirement grew with them. Then AI training and inference loads added roughly 60 percent of new data center demand inside a single eighteen-month window, with hyperscaler capex commitments now exceeding $400 billion annually.
Crusoe’s recent order, 29 LM2500XPRESS units for OpenAI’s Stargate site, is the new template. Modular aero-derivative gas turbines, ordered in industrial quantities, delivered directly to a hyperscaler site without the involvement of a regulated utility. This is not how the U.S. electricity market was designed to work.
- GE Vernova · 32%
- Siemens Energy · 27%
- Mitsubishi · 18%
- Ansaldo / others · 23%
Design once.
Now you design the turbine. Pick a customer. Pick a working fluid. Set the temperature, the pressure ratio, the mass flow. Then read the verdict.
The constraints are real. The math below is calibrated to the published efficiencies of the modern fleet, the lead times that have prevailed since 2022, and the customer requirements that map to the four real archetypes buying today.
Pick a customer.
Each archetype maps to a real buyer in the market today.
The El Segundo bet.
Two miles south of the runway at LAX, in a low-slung industrial building that used to house a defense subcontractor, a small team is doing something that has not been attempted in the United States in forty years. They are trying to design and build heavy-duty turbines from a clean sheet.
The company is called Stone Power. The founder is Dylan Morris, who studied at SMU, worked in turbomachinery, and publishes Turbine Tuesdays, a weekly Substack that has become the closest thing the industry has to an open-source briefing. The pitch is three things at once. Simplified gas turbines that drop into hyperscaler sites today. Closed-cycle turbines that will pair with advanced nuclear reactors tomorrow. And an AI-accelerated engineering loop, paired with American-scale manufacturing, that compresses the iteration cycle the incumbents are trapped inside.
The location is not incidental. El Segundo is now the densest concentration of American hard-tech operators in the country. SpaceX is across the runway. Anduril is two miles east. Hadrian, Castelion, K2 Space, and Apex are within a 15-minute drive. The talent that built the SpaceX cadence is now spreading laterally into turbines, hypersonics, manufacturing, satellites, and defense electronics.
Stone Power is not alone. NET Power is commercializing a supercritical CO2 oxy-combustion cycle with built-in carbon capture. Arbor Energy is building a biomass-fueled turbine that runs net-negative. The advanced nuclear vendors, X-energy and Kairos and Aalo and a half-dozen others, need turbine partners. The category around it is what David Sacks and Marc Andreessen have called American Dynamism. It is the bet that the next decade of industrial value will be created by founders who can do physical things at velocity.
The reader.
Imagine you are an architect-turned-operator, sitting in a Los Angeles coffee shop in May 2026. You have read this far. The question now is what you do.
A short, opinionated reading list. The eight items below are the ones that, taken together, will let you hold a serious conversation about turbines, the AI buildout, and American industrial capacity.
What happens next.
The AI buildout is now publicly committed. The hyperscaler capex curve has cleared $400 billion annually. The power requirement that follows is not a forecast, it is an arithmetic. Either the United States rebuilds the turbine industry and the surrounding civil capacity to install it, or it imports the answer from somewhere else.
Stone Power, Arbor Energy, NET Power, the advanced nuclear vendors, and the El Segundo manufacturing cluster are the visible bets. They are not the only ones. They are the ones whose architecture is legible today.
The next 36 months will determine the cycle. Either the new entrants reach commercial demonstration and the cost curve breaks open, or the incumbents extend their pricing power for another decade. Both paths produce a working grid. Only one of them produces an American industrial base on the other side of it.
The right time to learn turbines was twenty years ago. The second best time is now.
Colophon.
Built with curiosity and Claude Code in May 2026 by David T Phung. The architect’s instinct, the operator’s discipline.
Faith framing borrowed from Bezalel of Exodus 31. The Spirit-filled craftsman, the working hands.
- Stone Power. stonepower.us
- Dylan Morris, Turbine Tuesdays. dylanmorri.substack.com
- Visual Field Manual (PDF) available on request.