Power Generation: A Practical Reference

PERC is a monocrystalline p-type silicon cell with a dielectric stack (usually aluminum oxide capped with silicon nitride) deposited on the rear, where the aluminum rear contact reaches the wafer only through laser-opened line openings. The dielectric does two jobs at once: it passivates the rear surface so carriers stop recombining there, and it reflects long-wavelength light back into the wafer for a second pass. Together those are worth 1–2 points of efficiency over the aluminum back-surface-field cell PERC replaced, taking production cells to about 23% and modules to 21–22%. The upgrade was cheap because it bolted onto an existing Al-BSF line, and between roughly 2016 and 2021 it took over the entire industry. It has since lost that position to TOPCon just as fast, falling to something like 1–2% of 2025 module shipments. What survives is an installed base above a terawatt and the price everyone still quotes as the floor.

The strength is that everything about PERC is known: a decade of volume field data, a fully depreciated global tool base, and module prices of $0.08–0.12/W that set the benchmark every other cell has to beat. The weaknesses all come from the p-type wafer. Boron-oxygen light-induced degradation cost early PERC roughly 2–3% of output in its first weeks until gallium doping largely fixed it around 2019–2021, and LeTID, a slower degradation that appears after months at elevated cell temperature, is still the thing to look for in hot climates because it can take several percent off a module that passed factory flash test. The architecture also caps itself: metal touches silicon over a few percent of the rear, and recombination under that metal is what stops PERC going much past 23.5%.

Specify PERC only when someone is clearing remaining inventory below TOPCon on a $/W basis and the racking and land are already sunk, because at equal price per watt TOPCon gives you higher efficiency, a better temperature coefficient, and less degradation. Where PERC still matters is analysis rather than procurement: if you are diligencing an operating portfolio, most of the modules on the ground are PERC, and their degradation curve and LeTID exposure are what decide whether year-15 output matches the model. If you are repowering a PERC site, check that the racking and inverters can take a physically larger, higher-current n-type module before assuming a straight swap. For anything new, default to TOPCon.

Cell efficiency around 23%, module 21–22% · 1–2 efficiency points gained over the aluminum back-surface-field cell it replaced · roughly 1–2% of 2025 module shipments, down from the large majority in 2020 · over 1 TW installed worldwide · modules $0.08–0.12/W against a $1,150–1,600/kW utility project · degradation typically 0.5%/yr against a 25–30 year warranty.

A silicon PV plant's grid behavior comes from the tracker and the inverter, not from the cell, so nothing in this entry changes it. Output follows the sun, which means capacity credit starts near nameplate in a market with little solar and falls toward 10–20% once solar covers the afternoon and the net peak moves into the evening; CAISO and PJM have both cut solar accreditation for that reason. The inverters supply no inertia or fault current on their own, so interconnection agreements increasingly specify reactive power and grid-forming behavior as conditions of connection. The one PERC-specific point is scale: with more than a terawatt installed, the fleet's 0.5%/yr decay is quietly shrinking the accredited capacity behind contracts signed a decade ago.

Martin Green's group at UNSW demonstrated the cell in 1983 and won the 2023 Queen Elizabeth Prize for Engineering for it. Suntech and SolarWorld commercialized it around 2012–2014, and LONGi, JinkoSolar, Trina Solar, JA Solar, and Canadian Solar then built the gigawatt-scale mono-PERC lines that made solar the cheapest new generation on earth. Hanwha Qcells sold it as Q.ANTUM. Most utility plants commissioned anywhere between about 2017 and 2022 are PERC, which is why the operating fleet and the technology roadmap now look nothing alike.

PERC's cost structure is the standard silicon chain: polysilicon, ingot pulling, wafering, cell processing, module assembly, each step in a different factory and mostly in China. That chain is what got modules to $0.08–0.12/W, and it is also why nobody outside it competes on price alone. The lines are largely written down, so a PERC module can be sold at cash cost, which is what happened during the 2023–25 overcapacity glut and is why leftover PERC inventory sometimes undercuts TOPCon. The broader lesson for anyone building a business on PV is that the cell is not where the margin sits: modules are roughly a tenth of a $1,150–1,600/kW utility project, and the rest is balance of system, land, interconnection, and financing.

20 minPERC Solar Cell: Course Photovoltaics #16Photovoltaics Explained · 5k+ views

9 minWhat Is Solar PERC Technology? // Mono PERC // Poly PERC. // Passivated Emitter and Rear Contact.Solar Facts · 10k+ views

4 minThe Advantages of Mono PERC Solar Panel Technology: Passivated Emitter Rear ContactSolar Solution · 10k+ views

Crystalline Silicon Photovoltaics Research (US Department of Energy) · Photovoltaics Report, July 2026 edition (Fraunhofer ISE)

TOPCon puts a passivating contact across the whole rear of an n-type wafer instead of PERC's local openings. The stack is a tunnel oxide about 1–2 nm thick topped with a doped polysilicon layer: majority carriers tunnel through the oxide into the poly and out to the metal, while the oxide keeps them away from the silicon surface where they would recombine. Because the contact covers the full area rather than a few percent of it, resistance and recombination stop trading against each other, and production cells reach roughly 25–26% with modules at 22–24%. The wafer is phosphorus-doped n-type, which removes the boron-oxygen degradation p-type PERC had to engineer around. None of that is why TOPCon won. It won because the extra steps drop into an existing PERC line, so the industry converted its installed capacity in about three years instead of building new factories, and by 2025 TOPCon was close to 95% of shipments among the top module makers.

TOPCon buys 1–2 more points of module efficiency than PERC, a temperature coefficient near -0.29%/°C against PERC's -0.34, bifaciality of roughly 70–85%, and lower annual degradation, all at close to PERC's price per watt. It costs more silver per watt, and the polysilicon deposition wraps around the wafer edges and has to be etched off, which is where yield went during the first two years of the ramp. The degradation mode the industry is now watching is UV-induced degradation of the front surface, reported on fielded n-type modules and not well captured by the standard IEC UV preconditioning, so ask a supplier for extended UV test data rather than a certificate. The other exposure is legal: Trina, JinkoSolar, LONGi, and First Solar are all litigating TOPCon patents, and a US import ruling can move a module's landed cost more than any process improvement will.

Default to TOPCon for anything utility-scale. It is the cheapest way to buy a modern module, supply is enormous, and at equal $/W it beats PERC on every spec that matters. Move off it in three cases. If the site is hot and bright and you can model the extra energy, HJT's temperature coefficient and bifaciality may justify a premium. If area is the binding constraint, as on a roof, back-contact modules give you roughly 15% more watts in the same footprint. And if US import exposure is unacceptable, CdTe is the only non-silicon option at scale. Wherever you buy, check where the cells were made rather than just the modules, and check the specific supplier's patent position.

Tunnel oxide roughly 1–2 nm thick under a doped polysilicon layer · cell efficiency about 25–26%, module 22–24% · temperature coefficient near -0.29%/°C against PERC's -0.34 · bifaciality roughly 70–85% · close to 95% of 2025 shipments among the top module makers · converted from PERC in about three years because the added steps retrofit an existing line.

The dispatch profile is the same as any silicon PV plant, but two TOPCon properties move numbers an interconnection study cares about. Higher module efficiency and real bifacial gain put more DC capacity behind the same AC connection, and utility projects now run DC:AC ratios of 1.3–1.5; that clips a little midday energy and raises the morning and evening shoulders, which is what capacity accreditation actually rewards. Lower degradation matters over a 20–30 year PPA, since an n-type module with no boron-oxygen loss holds several more points of nameplate at year 25, so the accredited capacity a system operator counts on late in the contract is higher. Neither changes the fact that the plant is inverter-coupled and supplies no inertia unless the inverters are specified to.

JinkoSolar's Tiger Neo, Trina Solar's Vertex N, JA Solar's DeepBlue 4.0, Canadian Solar's TOPHiKu, and Astronergy's ASTRO N are the volume TOPCon lines. Jinko reached volume first in 2022–23 and the rest of the top ten followed within about two years. On the legal side, Trina has asserted TOPCon patents against Runergy and Adani in the US, and First Solar holds and licenses a TOPCon patent portfolio it acquired with TetraSun in 2013, which is an unusual position for a company that makes no silicon cells.

TOPCon's economics are the reason it exists. Upgrading a PERC line costs a fraction of a greenfield cell fab, so the industry captured a 1–2 point efficiency gain without writing off its tool base, and module prices barely moved. The consequence is chronic overcapacity: Chinese nameplate cell and module capacity has run at roughly twice global demand since 2023, prices fell below cash cost through much of 2024–25, and most listed module makers lost money over that stretch. If you are buying modules, this is the best procurement environment the industry has ever had. If you are trying to build a module business, the only defensible positions are a policy-protected market (US 45X credits and tariffs), a patent position, or a genuinely different cost structure such as CdTe.

7 minTOPCON Tunnel Oxide Passivated Contact New solar cell Technology BreakthroughSmart Solar · 10k+ views

3 minWhat is TOPCon solar module technology? Is it worth investing in?Ornate Solar · 10k+ views

7 minJinko Solar Virtual Tour-Cell FactoryJinkoSolar · 10k+ views

TOPCon – Overcoming Fundamental Bottlenecks to a New World-Record Silicon Solar Cell (Fraunhofer ISE) · International Technology Roadmap for Photovoltaic, 17th Edition 2026 (VDMA)

Heterojunction cells sandwich an n-type crystalline wafer between thin films of amorphous silicon. A few nanometers of intrinsic hydrogenated amorphous silicon on each face passivate the crystalline surface chemically, doped amorphous layers on top of those decide which carrier leaves each side, and a transparent conductive oxide carries current sideways to low-temperature silver paste fingers. Amorphous silicon passivates better than any diffused junction, so open-circuit voltage reaches 740–750 mV against roughly 690 mV for PERC, and LONGi's record HJT cell hit 26.81% in 2022. The structure is symmetric, which is why HJT tolerates wafers as thin as 100–130 µm without bowing and why its bifaciality reaches about 90%. Every process step runs below roughly 200 °C, so an HJT line shares almost nothing with a PERC line and has to be built from scratch.

HJT has the best temperature coefficient in mass production, around -0.25%/°C against PERC's -0.34, the highest bifaciality of any commercial cell, and very low first-year and annual degradation, so it delivers more kWh per installed watt than TOPCon at hot, bright, high-albedo sites. The gain is usually a few percent and it is worth modeling site by site rather than assuming. Against that, HJT consumes the most silver per watt of any mainstream cell, because low-temperature paste is both less conductive and more expensive, and its transparent conductive oxide normally contains indium, a constrained byproduct metal. The commercial failure mode has already happened more than once: with no retrofit path, an HJT maker has to fund a whole new line and then sell at a premium, and Meyer Burger ran out of money trying, shutting its US module plant and putting its main German subsidiaries into insolvency in 2025. Technically, watch the cell interconnection, because low-temperature soldering and busbar-less designs mean the joint often ages faster than the cell.

Pick HJT when the site is hot, the ground is bright, and the offtake pays for energy rather than capacity. Desert utility projects, high-altitude sites, and snow-covered ground are where the temperature coefficient and the 90% bifaciality turn into revenue. The test is arithmetic: take the module price premium over TOPCon in $/W, divide it by the modeled extra kWh per watt over 25 years, and compare the result to your PPA price. In most temperate markets that premium does not clear. Don't pick HJT on datasheet efficiency, since TOPCon has closed most of that gap; pick it on modeled energy yield, and if the yield model doesn't clear the premium, buy TOPCon.

Intrinsic amorphous silicon passivation layers a few nanometers thick on both faces · open-circuit voltage 740–750 mV against roughly 690 mV for PERC · record cell 26.81% (LONGi, 2022), production modules around 22.5–24% · temperature coefficient near -0.25%/°C against PERC's -0.34 · bifaciality about 90% · wafers as thin as 100–130 µm · every process step below roughly 200 °C.

Temperature coefficient decides how much of a plant's nameplate is actually available at the system peak, because peaks land on the hottest afternoons. At a 65 °C cell temperature, which is ordinary in summer, an HJT module at -0.25%/°C is down about 10% from its rating while a PERC module at -0.34%/°C is down about 14%, and that gap falls in exactly the hours that set capacity accreditation and the highest prices. Bifacial gain near 90% adds to the morning and evening shoulders over bright ground, shifting a little output away from the midday hours most likely to be curtailed. Everything else is the same as any silicon plant: inverter-coupled, no native inertia, and a capacity credit that falls as solar penetration rises.

Huasun is the largest HJT producer and Risen Energy's Hyper-ion is the best-known module line. Panasonic invented the structure as HIT in the 1990s and its core patents expired around 2010–2011, which is what let everyone else build it. Meyer Burger was the Western attempt to industrialize it and failed in 2025. HJT also underpins several research records, including LONGi's 26.81% cell, and heterojunction passivation is what the highest-efficiency back-contact cells are now built on.

HJT has to be paid for twice: a greenfield low-temperature line instead of a PERC retrofit, then a bill of materials carrying more silver and some indium. Silver is the swing item, and the industry's answer is silver-coated copper paste and electroplated copper metallization, both in production at some Chinese makers and both of which change the cost picture only if they hold up for 25 years outdoors. So HJT is really a bet on two things arriving together: copper metallization removing the silver premium, and a market that pays for energy yield rather than $/W. If either fails, TOPCon absorbs the volume, which is roughly what happened between 2022 and 2025.

12 minHIT (Heterostructure with Intrinsic Thin Layer) solar cell: Part 1nanolearning · 10k+ views

19 minWhat is Hetero Junction Technology in Solar Panels?WAAREE Energies Limited · 5k+ views

10 minDifferent Types of Solar Panels: What is PERC, TOPCon, N-Type, HJT,...Cleversolarpower by Nick · 10k+ views

Silicon-based heterojunction solar cells (EPFL PV-Lab) · Degradation and Failure Modes in New Photovoltaic Cell and Module Technologies, 2025 (IEA-PVPS Task 13)

Back-contact cells move both polarities to the rear, interdigitated as alternating p and n strips, so the front face carries no metal at all. That recovers the 4–6% of incident light a conventional front grid shades, and it lets the front surface be optimized purely for optics and passivation. The price is that carriers generated near the front have to reach the rear without recombining, so the design needs high-lifetime n-type wafers, excellent front passivation, and alignment-critical patterning of two doped regions plus two isolated metal grids on the back. Modern products are hybrids rather than pure IBC: LONGi's HPBC combines back contacts with a TOPCon-style passivated contact, Aiko's ABC is its own stack, and the highest research cells combine back contacts with heterojunction passivation, which is where silicon single-junction records above 27% have come from. Commercial modules land at roughly 24–25%, the highest available, in a panel with no visible gridlines.

Back contact gives the highest module efficiency on the market, about 24–25% against 22–24% for TOPCon, in a uniform black panel, which is why it owns the premium residential segment. Vendors also claim better behavior under partial shading from the way the rear metallization is segmented; treat that as a claim to test on the actual roof rather than as a spec. The weakness is manufacturing. Back contact adds masking and alignment steps to a flow TOPCon deliberately kept short, demands tighter wafer quality, and yields lower, which shows up as cost per watt rather than as a field failure. The failure mode that has actually cost buyers money is corporate: SunPower, the company that commercialized IBC, filed for Chapter 11 in 2024, and Maxeon, its spun-out manufacturing arm, had to be recapitalized under TCL Zhonghuan's control. A 25-year warranty is only as good as the entity behind it, and this segment has failed that test repeatedly.

Choose back contact when roof area is fixed and worth more watts: a house with one small south-facing plane, a commercial roof with a hard kWh target, a carport, or any job where mounting and labour per module dominate the module price. In those cases roughly 15% more watts per square meter is worth a real premium, because you save a proportional share of racking, wiring, and install labour. Don't choose it for a utility project, where land is cheap and $/W is the only number that matters; TOPCon wins there and will keep winning. Whichever you pick, diligence the warranty counterparty as carefully as the module, because the back-contact segment's insolvency record is worse than its technology record.

Module efficiency roughly 24–25%, the highest commercially available, against 22–24% for TOPCon · about 240–250 W/m² against 210–220 for a mainstream module · recovers the 4–6% of light a front metal grid shades · silicon single-junction cell records above 27% come from back-contact designs · more masking and alignment steps and lower yield than any other silicon cell.

These modules mostly sit behind the meter, so the grid question is the service connection and the distribution feeder rather than the transmission system. Fitting 15% more kW onto an area-capped roof means 15% more midday export through the same connection, and on a circuit already carrying a lot of rooftop solar the binding limit becomes hosting capacity and voltage rise, not anything at the array. Where the utility caps export rather than installed capacity, or where the tariff pays little for exported energy, those extra watts are worth much less than the datasheet implies, so read the interconnection rule before paying the premium. As with all PV, the inverter decides what grid services the system provides, and IEEE 1547-2018 now requires volt-VAR control and ride-through that older installations do not have.

Aiko Solar's ABC modules and LONGi's Hi-MO 9, built on HPBC 2.0, are the volume products, both aimed at premium rooftop in Europe, Japan, and Australia. Maxeon 7 descends from the SunPower IBC cell that ran from the mid-2000s onward. JinkoSolar has shown a back-contact product of its own, which is the signal that the format is moving from a niche toward a second mainstream option rather than staying a specialty.

The cost problem is process steps, and it is stubborn: even after crediting the higher efficiency, back contact costs more per watt than TOPCon, and the gap only closes where area carries a price. That is why the market splits cleanly, with back contact taking premium residential and small commercial while utility projects stay on TOPCon. The industry's own answer is to fold back contacts into lines that already exist, which is what LONGi's HPBC 2.0 is; if that works at scale the format becomes a feature of mainstream cells rather than a separate product with a separate cost base. The historical warning for investors is that every pure-play back-contact company so far has struggled to fund its next line, so treat manufacturing scale and balance sheet as the diligence items, not cell efficiency.

14 minSunpower IBC cell: process flownanolearning · 10k+ views

13 minHow the Back of a New Solar Panel Just Changed EverythingUndecided with Matt Ferrell · 100k+ views

Technology development for silicon solar cells (ISFH) · Crystalline Silicon Photovoltaics Research (US Department of Energy)

CdTe is the only non-silicon photovoltaic technology at real scale, and essentially all of it comes from First Solar. A few microns of cadmium telluride, now a graded cadmium selenide telluride absorber, are deposited by vapor transport onto a sheet of coated float glass, with a thin window layer in front and a back contact behind, and laser scribes cut the coating into series-connected cell strips so the module is interconnected as it is built. The absorber has a direct bandgap near 1.45 eV, close to the single-junction optimum, so about 3 µm of it absorbs what 150 µm of silicon needs. The process matters more than the physics here: glass goes into one factory and a finished module comes out, and First Solar quotes roughly four and a half hours for the whole sequence, against a multi-day chain across polysilicon, ingot, wafer, cell, and module plants owned by different companies in different countries. Commercial modules run about 19–20%, below silicon, with a record cell around 23%.

CdTe's advantages are structural. One vertically integrated plant makes the entire module, so there is no polysilicon or wafer market to be exposed to, no silver in the bill of materials, and a supply chain that sits in the US rather than in China. Its temperature coefficient (around -0.3%/°C) and its response in humid, hazy air are better than p-type silicon's, so the energy-yield gap is narrower than the 19–20% against 22–24% nameplate gap suggests. The weaknesses are equally structural. Lower efficiency means more modules, racking, and land per MW, so CdTe loses wherever area carries a price. Tellurium is a byproduct of copper refining with world supply of only a few hundred tonnes a year, which caps how large the technology can get. Cadmium content drives permitting questions and an end-of-life obligation even though the compound is stable and sealed in glass. The honest failure mode is concentration: one company, a handful of fabs, no second source, so a process problem or a policy change hits the whole technology at once.

Buy CdTe when US content, tariff exposure, or delivery certainty matter more than watts per square meter, which in 2025–26 describes a large share of US utility projects, because domestic manufacturing credits and the tariff position on Southeast Asian silicon move the landed cost more than the efficiency difference does. It also does well in hot, humid climates, where the temperature coefficient and spectral response narrow the yield gap against silicon. Don't buy it where land or racking is expensive per square meter, on constrained rooftops, or where you need bifacial gain, since most of the installed fleet is monofacial. And don't build a procurement strategy around a second CdTe supplier appearing, because the technology and the tellurium position are effectively one company's.

Absorber about 3 µm thick with a direct bandgap near 1.45 eV, against roughly 150 µm for a silicon wafer · commercial module efficiency 19–20%, record cell around 23% · temperature coefficient near -0.3%/°C, better than PERC's -0.34 · roughly four and a half hours from glass to finished module, against a multi-day silicon chain · world tellurium supply of only a few hundred tonnes a year.

The plant behaves like any other utility PV plant: variable, inverter-coupled, no native inertia, a capacity credit that falls as solar penetration rises, and the same midday curtailment exposure in high-solar markets. Two things differ at the margin. Better hot and humid performance means the fleet holds slightly more of its nameplate through the summer afternoons that set capacity accreditation, and most of the installed base is monofacial and large-format, which changes racking height and albedo assumptions in the yield model without changing the dispatch shape. In practice a system operator studying a CdTe project uses the same interconnection assumptions it would use for silicon.

First Solar is the technology: Series 6 and Series 7 modules from plants in Ohio, Alabama, and Louisiana in the US, plus Malaysia, Vietnam, and India, on the way to roughly 25 GW a year of nameplate capacity. Topaz Solar Farm and Desert Sunlight in California, both 550 MW and both completed in 2015, are the classic large CdTe plants. The company also runs a module recycling business, which exists partly because of the cadmium content and partly because tellurium is worth recovering.

CdTe's cost structure contains no polysilicon, no wafers, no silver, and no cell-to-module supply chain, so its cost curve is driven by throughput, deposition efficiency, and glass rather than by the commodity cycle that whipsaws silicon. That is why First Solar stayed profitable through the 2023–25 glut that pushed most Chinese module makers below cash cost. The business rests on two policy legs: the US 45X manufacturing credit, which pays per watt produced domestically, and the tariff structure on imported silicon cells. Both are legislative. The right way to underwrite the company is as a manufacturing business with a durable process advantage plus a policy-dependent price premium, rather than as a technology that would win on physics alone.

11 min5.4 CdTe PV TechnologyDelftX Solar · 5k+ views

4 minCadmium Telluride (CdTe) Solar PanelsLDSreliance · 50k+ views

Cadmium Telluride (US Department of Energy) · Tellurium, Mineral Commodity Summaries 2026 (USGS)

A tandem stacks a wide-bandgap perovskite cell on top of a silicon cell so each absorbs the part of the spectrum it converts best: the perovskite, around 1.68 eV, takes the blue and green photons silicon wastes as heat, and the 1.12 eV silicon cell takes the red and near-infrared the perovskite passes through. In a two-terminal monolithic stack both subcells carry the same current, so bandgap and layer thickness have to be tuned to match, and most of the engineering sits in that current match, in the recombination junction between the cells, and in the transparent contacts. This is the only credible route past silicon's practical single-junction ceiling of about 29.4%. LONGi holds the certified cell record at 35.5% (ESTI, July 2026), and Oxford PV has a Fraunhofer-certified 26.9% 60-cell module off the first commercial line. The perovskite itself is a solution- or vapor-processed film deposited at low temperature, so in principle it adds a few steps to a finished silicon cell rather than replacing the line. Efficiency is no longer the hard part.

The upside is the only large efficiency step left in silicon PV: a 30% module against a 22–24% one is roughly a quarter to a third more energy from the same area, racking, wiring, and grid connection, and the top cell is a thin film added to a wafer process that already exists. The problem is durability. Perovskites degrade under moisture, oxygen, UV, and thermal cycling; mixed-halide wide-bandgap compositions can phase-segregate under illumination; and mobile ions drift under field and heat. None of those are failure modes silicon has. Standard qualification (IEC 61215 damp heat, 200 thermal cycles, UV preconditioning) was written around silicon's failure modes and does not predict perovskite lifetime, so a passing certificate is weaker evidence than it looks. The honest status is that nobody has a perovskite module with 25 years of outdoor data, and LONGi, which holds the efficiency record, has said it has no active mass-production plan.

Track tandems; don't specify them, unless you are running a pilot and can absorb the loss. If you are a developer, the sensible position today is an option: a small deployment, or a supply agreement with a real degradation warranty and a solvent counterparty behind it. If you are an investor, the diligence question is not efficiency, because everyone will clear 30% eventually. It is outdoor data: how many module-years, at what temperature and humidity, measured how, and what the extrapolated year-25 output is with a stated confidence interval. A company that answers with a champion-cell number and an accelerated-test certificate has not answered. Buy TOPCon for anything you need to finance this decade.

Perovskite top cell around 1.68 eV over a 1.12 eV silicon bottom cell · certified cell record 35.5% (LONGi, ESTI, July 2026) against silicon's practical single-junction ceiling near 29.4% · Fraunhofer-certified 26.9% for Oxford PV's 60-cell module · a 30% module yields roughly a quarter to a third more energy per unit area than today's 22–24% modules · no perovskite module has 25 years of outdoor data, against 25–30 year warranties on silicon.

Watts per square meter is the grid-relevant number here. A tandem raises the energy a fixed site and a fixed interconnection can deliver by roughly a quarter to a third, which matters most where the interconnection queue rather than land is the binding constraint, because a repowered site or an existing point of connection gets more energy without a new study. The dispatch shape does not change at all: still a variable, inverter-coupled plant with the same curtailment exposure and the same falling capacity credit as solar penetration rises. Degradation uncertainty is itself a grid question, since capacity accreditation and contracted PPA volumes both assume a known annual decay rate, and until there is field data an offtaker has to price that volume risk somewhere.

Oxford PV runs the first commercial perovskite-on-silicon line at Brandenburg an der Havel, shipped its first modules in 2024, and has licensed its technology to Trinasolar. LONGi holds the certified cell record. Hanwha Qcells runs a tandem pilot line in Jincheon, Korea, and Trinasolar and JinkoSolar have both published module-scale tandem results. Swift Solar and Caelux are the notable US startups, though Caelux ships a perovskite-on-glass product rather than a silicon tandem.

The cost case is that the perovskite layer adds a modest amount to a silicon cell that already exists, so incremental cost per watt is small if yields are good and the added area cost is zero. None of that is proven at volume. Deposition uniformity over a full 210 mm wafer, encapsulation that keeps moisture out for decades, and yield through the extra steps are all open questions, and the equipment industry for large-area perovskite deposition barely exists. Funding has come mostly from strategics and national programs rather than project finance, which is the correct read of the risk. Anyone underwriting a tandem business should model the first factory as first-of-a-kind capex and size a warranty reserve big enough to survive being wrong about degradation, because that is the loss that would end the company.

3 minHow Do Tandem Solar Cells Work?Swift Solar · 10k+ views

2 minAn Introduction to Perovskite Tandem Solar CellsHanwha Group · 100k+ views

11 minAre perovskite cells a game-changer for solar energy?DW Planet A · 1m+ views

Perovskite Solar Cells (US Department of Energy) · Perovskite Tandem Solar Cells (Helmholtz-Zentrum Berlin)

An onshore wind turbine is a three-bladed, upwind, pitch-regulated rotor on a tubular steel tower, driving either a three-stage gearbox and a doubly-fed or full-converter generator, or a direct-drive permanent-magnet generator with no gearbox at all. Power available in the wind rises with the cube of wind speed and the square of rotor diameter, and wind speed rises with height, so every generation of machine has grown: onshore units are now typically 4–7 MW with 150–180 m rotors on 100–160 m towers, extracting 45–50% of the wind's kinetic energy against a theoretical Betz maximum of 59.3%. The design trend that matters is falling specific power, meaning more swept area per rated kilowatt. A bigger rotor on the same generator reaches rated output at a lower wind speed, so the machine sits at full load more of the time, and capacity factor is what the market pays for. Good new US projects run 40–50%, against a 34.2% fleet average in 2025 that still carries a lot of older, smaller machines.

Onshore wind is the second-cheapest source of bulk energy, $37–86/MWh at $1,900–2,300/kW, and its output shape complements solar, because it blows harder at night and in winter. It also uses very little of the land it occupies, typically a couple of percent of the project footprint, so farming continues around the turbines. Cost stopped being the constraint around 2020. What stops projects now is siting: local opposition and county ordinances, setbacks, radar and aviation objections, the transmission needed to move power out of windy low-population areas, and a US interconnection queue where the median project reaching operation in 2025 had waited over five years. The failure modes that show up in operating portfolios are blades and gearboxes, including leading-edge erosion, structural blade failures on the newest and largest models, and mid-life gearbox replacement, so a 20–30 year model without a serious mid-life capital line is wrong.

Build onshore wind where measured wind supports a capacity factor above about 35%, land is available at scale, and there is a transmission path. Make the site decision from a met mast or lidar campaign rather than a wind atlas, because a 1% error in estimated mean wind speed is roughly a 2% error in energy and the error compounds straight into the debt sizing. Pair it with solar wherever both resources exist, since the combined portfolio has a higher capacity credit and a flatter output than either alone. Choose utility solar instead when land is constrained, the wind resource is marginal, or you need a faster and more predictable build, since solar takes about 15 months of construction against 18 for wind with far less permitting risk. Choose offshore only when land is genuinely unavailable or the load is coastal, because capex per kW roughly doubles.

Machines typically 4–7 MW with 150–180 m rotors on 100–160 m towers · rotor extracts 45–50% of the wind's kinetic energy against a Betz maximum of 59.3% · capacity factor 30–55% by site, US fleet average 34.2% in 2025 · $1,900–2,300/kW and $37–86/MWh unsubsidized · roughly 18 months of construction · 167 GW added worldwide in 2025 · a 1% error in estimated wind speed is roughly a 2% error in energy.

Wind's capacity credit runs well below its capacity factor, typically 10–25% of nameplate in US markets and falling as more wind is added, because the hours it can be relied on shrink faster than its energy does. Modern turbines connect through power electronics and contribute no natural inertia, so grid codes have had to specify fault ride-through, reactive support, and now grid-forming behavior explicitly. The complementarity with solar is real and worth modeling, since a wind-plus-solar portfolio carries a higher combined capacity credit than the sum of its parts, which is why hybrid interconnection requests have become common. The binding constraint is transmission: good wind sits far from load, and curtailment on windy nights when the export path is full is routine in ERCOT, SPP, and MISO.

Vestas, Siemens Gamesa, GE Vernova, and Nordex supply the Western market and Goldwind, Envision, and Mingyang ship most of the world's turbines by capacity. Typical current machines include the Vestas V163-4.5 MW and GE Vernova's 3.x and 6.x onshore platforms in the US, with 6–10 MW class machines standard in China. Texas runs over 40 GW of installed wind, and Iowa, Kansas, Oklahoma, and South Dakota each take more than a third of their electricity from it.

Roughly two-thirds of project capex is the turbine and tower; the rest is foundations, roads, the collection system, the substation, and interconnection. Western OEMs spent 2022–2024 losing money on onshore turbines after signing fixed-price contracts into an inflationary supply chain, and Siemens Gamesa's 4.X and 5.X quality problems cost it billions in warranty provisions, which is a useful reminder that a turbine warranty is a credit exposure to the manufacturer. Chinese turbine prices have run at a fraction of Western ones and are reshaping every market that admits them. For a developer the value has moved off the machine and onto things that are scarce: a measured wind resource, land control, a queue position, and a PPA with a creditworthy buyer. Repowering an existing site with a larger rotor on the same interconnection is currently one of the better returns in the sector, because the expensive permission is already in hand.

12 minInside a Wind Turbine: How It Really WorksFirst Gear Animations · 100k+ views

5 minHow do Wind Turbines work?Sabin Civil Engineering · 5m+ views

4 minWind turbine generators, HOW DO THEY WORK?DOB-Academy · 100k+ views

Land-Based Wind Market Report (US Department of Energy) · Renewable Power Generation Costs in 2025, executive summary (IRENA)

Fixed-bottom offshore wind puts 14–15 MW turbines with 220–240 m rotors on steel foundations driven into the seabed in water up to about 60 m deep. The standard foundation is a monopile, a single tube up to roughly 11 m in diameter weighing 1,500–2,500 tonnes, driven or drilled in by a jack-up or heavy-lift vessel; jackets, four-legged lattice structures, take over where soils or depths defeat a monopile. Wind offshore is stronger and steadier than onshore, so capacity factors run 45–55%, and the plants sit near coastal load centers that would otherwise need long inland transmission. Everything else about the sector is harder than onshore: an installation vessel fleet numbering a few dozen ships worldwide, offshore substations, subsea export cables, and maintenance access limited by significant wave height.

The resource is the strength. A 45–55% capacity factor, output that holds up on winter evenings when solar is gone, and generation delivered close to demand are worth a lot to a coastal system. The cost side repriced badly. At $3,450–6,550/kW and $70–157/MWh, offshore wind runs two to three times onshore capex per kW for about 1.5 times the capacity factor, and from 2023 to 2026 the sector worked through contract renegotiations, cancellations, and multi-billion-dollar write-downs as interest rates, vessel day rates, and steel prices moved against contracts signed years earlier. The structural problem is that a fixed-price, non-indexed offtake contract signed three to five years before financial close puts all the inflation and rate risk on a developer whose asset is almost entirely front-loaded capital, so a 200 basis point move in rates changes the price the project needs by tens of dollars per MWh. The operational failure mode is cables: subsea export and array cable faults are the largest single source of offshore wind insurance claims by value, and a single export cable is a single point of failure for a whole wind farm.

Build fixed-bottom offshore where land-based siting is genuinely unavailable, the shelf is shallow and close to a coastal load center, and a policy mechanism will pay $70–157/MWh, which in practice means the North Sea, the Baltic, the US Northeast, Taiwan, Japan, and coastal China. Do not build it as a cheaper alternative to onshore wind; at these capex ratios it never will be. If you are structuring the offtake, insist on inflation indexation and set the strike price close to financial close rather than years ahead, because that mismatch, not the technology, caused most of the sector's losses. If the water is deeper than about 60 m, the comparison is floating rather than fixed.

Turbines 14–15 MW with 220–240 m rotors · monopiles up to roughly 11 m in diameter and 1,500–2,500 tonnes, in water to about 60 m · capacity factor 45–55% · $3,450–6,550/kW and $70–157/MWh unsubsidized · 24-month construction assumption, 3–6 years in practice · roughly 2–3× onshore capex per kW for about 1.5× the capacity factor.

A 45–55% capacity factor that peaks in winter gets a much better capacity credit than onshore wind, often 30–45% of nameplate depending on the market, and offshore output correlates less with solar than any other renewable in a Northern European or US Northeast system. The grid connection is a project in its own right: an offshore substation, an export cable (HVAC out to roughly 80 km, HVDC beyond), and onshore reinforcement at the landing point, together commonly around 15% of project cost and often the item that sets the schedule. Because the turbines are inverter-coupled, systems with a lot of offshore wind have had to buy inertia separately, and the UK and Germany now procure synchronous condensers and grid-forming capability explicitly. Landing several gigawatts at a single coastal point also concentrates risk, which is why operators are moving toward meshed offshore networks instead of radial links.

Hornsea 2 (1.32 GW, UK) and Dogger Bank (3.6 GW when complete, using GE Vernova's Haliade-X) are the largest projects; Vineyard Wind 1 and Revolution Wind are the US flagships. The repricing shows up in specific deals: Ørsted canceled Ocean Wind 1 and 2 in 2023 and took a write-down of several billion dollars, Avangrid paid to terminate the Commonwealth Wind PPAs, and the UK's AR5 auction in September 2023 drew no offshore wind bids at all before AR6 cleared at a higher strike price. Federal stop-work orders on Empire Wind and Revolution Wind in 2025, both later lifted, added a political risk that US project models had not previously carried.

Roughly a third of the cost is turbines and the rest is foundations, installation vessels, cables, substations, and construction finance, so offshore wind's cost curve depends on heavy industry and shipping rather than on a factory learning curve. That is why its costs rose with steel, vessel day rates, and interest rates instead of falling the way solar's did. Revenue is almost entirely policy: CfDs in the UK and Europe, OREC contracts in US states, feed-in tariffs in Asia. Two lessons come out of 2023–26 and both are contractual. Index the strike price, and set it as close to financial close as you can, because the gap between bid and FID is where the money was lost. Anyone selling into the sector should note that the scarce inputs are installation vessels, HV cable, and large-forging capacity, not turbines.

4 minHow it all comes together at sea: installing an offshore wind farmSiemens Gamesa · 100k+ views

2 minMonopile installation - Sandbank Offshore Wind FarmTWD · 50k+ views

2 minHow Offshore Wind Farms WorkØrsted · 50k+ views

Offshore Wind Market Report: 2024 Edition (National Laboratory of the Rockies) · Making Offshore Wind Work (World Bank and ESMAP)

Floating offshore wind puts the same turbine on a moored platform instead of a fixed foundation, which opens water deeper than about 60 m, where most of the world's offshore wind resource actually sits: off Japan, Korea, California, the Mediterranean, and the Atlantic edge of Europe. Three platform types compete. Semi-submersibles float on buoyancy in widely spaced columns and can be assembled and towed from an ordinary quay, which is why most projects use them. Spars are ballast-stabilized cylinders, very stable but needing deep sheltered water for upending and turbine assembly. Tension-leg platforms are held down by vertical tendons kept in tension, which makes them light and stiff but hard to tow and install. All three need catenary or taut mooring lines with drag-embedded or suction anchors, plus a dynamic export cable that hangs in the water column and flexes for the life of the project.

Deep-water sites have the best wind resource available anywhere, and Hywind Scotland's 30 MW has reported annual capacity factors above 50%. Floating also opens coastlines that have no shallow shelf and therefore no other local option. The costs are all in the parts that are not the turbine. Platform steel or concrete runs into the thousands of tonnes per unit, mooring and anchors scale with depth, dynamic cables are a specialized product with limited fatigue history, and the assembly port needs deep draft and very high quayside bearing capacity, which most ports do not have. Only a few hundred megawatts are installed worldwide, nearly all at demonstration scale, so every cost estimate is an extrapolation rather than a measurement. The failure mode the industry is still learning is fatigue: platform, moorings, and dynamic cable all see load cycles a fixed foundation never sees, and no floating system yet has a 25-year service record.

Floating is the answer only where the water is too deep to fix a foundation and the coastline has load worth serving, which means Japan, Korea, California, Norway, Portugal, and parts of the Mediterranean. Everywhere else, fixed-bottom is cheaper and much better understood. If you are evaluating a floating project, the diligence questions are industrial rather than aerodynamic: which fabrication yard, at what serial rate, out of which port, with which mooring and anchor design for that specific seabed. The cost curve comes from repeating one hull design dozens of times in a yard, so a project without a committed serial fabrication route is a demonstration with a large price tag attached.

Water deeper than about 60 m, where most of the world's offshore wind resource sits · a few hundred MW installed worldwide, almost all demonstration scale · capacity factors above 50% at deep-water sites, with Hywind Scotland reporting above 50% in service · platform steel or concrete in the thousands of tonnes per unit · three competing hull types (semi-submersible, spar, tension-leg) · capex above $7,000/kW on current projects.

Output is variable but flatter and stronger than any other wind, since deep-water sites hold above 50% capacity factor with modest diurnal variation, so the capacity credit sits at the top of the wind range. The connection is the hard part. A dynamic export cable has to survive 25 years of motion in the water column, and for California and Japan the onshore side is a bigger problem than the offshore one, because the landing points are on coastlines with weak transmission and no spare capacity. Port capability is a grid prerequisite here in a way it is not for other technologies: Humboldt Bay has to be rebuilt before California's leases can be assembled at all, and that work sits on the same critical path as the transmission upgrade.

Hywind Scotland (30 MW of spars, 2017), Hywind Tampen (88 MW, 2022–23, the largest so far, powering Norwegian offshore oil platforms), WindFloat Atlantic (25 MW semi-submersible off Portugal), and Kincardine (50 MW, Scotland) are the operating projects. Equinor's Hywind and Principle Power's WindFloat are the two best-known hull designs. BOEM's December 2022 California lease sale sold five leases for about $757 million covering roughly 4.6 GW off Humboldt and Morro Bay, and Japan's Goto and NEDO demonstrations are the Asian equivalent.

Floating wind's cost problem is a fabrication problem. The turbine is the same machine as fixed-bottom, so the cost curve depends on building hulls in a serial yard, standardizing moorings, and getting dynamic cable into volume production, none of which respond to turbine R&D. That makes the economics look like shipbuilding: the fifth hull out of a yard costs far less than the first, and a yard without an order book never gets there. Public money is doing the early work through UK CfD allocations, ScotWind, Norway's Utsira Nord, and US state procurement, and the honest expectation is that costs stay well above fixed-bottom into the 2030s. For a supplier, the durable positions are mooring and anchor systems, dynamic cables, and quayside capacity, all of which are scarce and none of which is a turbine.

9 minFloating wind turbines: Offshore energy's secret weaponDW Planet A · 100k+ views

3 minInstalling Floating Wind Turbines | Foundation Types ExplainedFidar - Offshore Animation · 5k+ views

4 minA Guide to a Floating Offshore Wind FarmOffshore Renewable Energy Catapult · 10k+ views

Floating Offshore Wind Shot (US Department of Energy) · Floating Wind JIP Stage 3 Phase I Summary report, July 2025 (Carbon Trust)

A reservoir hydropower plant stores water behind a dam and releases it through turbines when the operator wants power, which makes it the original dispatchable renewable and still the largest low-carbon source of electricity in the world: roughly 1,253 GW installed producing about 4,600 TWh a year, near 14% of global generation, at a global average capacity factor around 42%. The machine follows the head. Pelton wheels suit high-head sites above roughly 300 m, Francis turbines cover the broad middle from about 30 to 300 m, and Kaplan units with adjustable blades handle heads under 30 m at high flow. A unit goes from standstill to full load in a minute or two, and the energy store is the reservoir itself, so flexibility costs nothing extra once the dam exists. Construction takes 6–12 years and the capital cost is entirely site-specific, but the civil works last 50–100 years, and that long tail is most of the reason hydro's lifetime cost looks so low.

Hydro gives a grid nearly everything it wants from a single asset: cheap energy, fast dispatch, real inertia, black-start capability, and an 80-year life. It is also the only technology on this sheet whose flexibility arrives as a by-product rather than as a purchase. The weaknesses are geography and water. Good sites are largely taken in North America and Europe, and what remains is in Africa, South Asia, and the Andes, where projects are large, politically difficult, and slow. Drought is the operating failure mode, and it correlates across a whole basin at once, so energy and firm capacity fall together exactly when a system can least afford it: Brazil in 2021, southwest China in 2022, and Zambia and Zimbabwe on the Kariba reservoir through 2024–25 each lost gigawatts of supposedly firm output for months. Add sedimentation slowly eating storage volume, displacement of people, methane emissions from tropical reservoirs, and relicensing obligations for fish passage and minimum flows that permanently cost output.

If you have a site with head, flow, and a route through permitting, build it, because nothing else gives you dispatchable clean energy with a 50–100 year life at $50–100/MWh. That is a narrow if. For most systems the practical hydro decisions are not new dams but three cheaper moves: uprating existing units at refurbishment, adding generation to non-powered dams, and negotiating relicensing terms that preserve operating flexibility. When you compare hydro against a gas peaker for flexibility, remember that hydro supplies it as a by-product of energy it was going to sell anyway, so the marginal cost of a start is close to zero. If what you need is to shift energy across hours rather than to generate it, that is pumped storage, which is covered on the energy-storage sheet. And in a basin with a drying hydrological trend, model the dry-year output rather than the average.

About 1,253 GW installed worldwide producing roughly 4,600 TWh a year, near 14% of global electricity · global average capacity factor around 42% · Pelton above roughly 300 m of head, Francis 30–300 m, Kaplan below 30 m · standstill to full load in a minute or two · construction 6–12 years, civil works lasting 50–100 years · $50–100/MWh at $1,500–3,000/kW, both highly site-specific.

This is the entry where the plant supplies far more than energy. A synchronized hydro unit provides real rotating inertia and fault current, frequency regulation, spinning reserve, and voltage support, and it can black-start a grid; many units can also be dewatered and spun on air as synchronous condensers, supplying inertia and reactive power with no water at all. Capacity credit is close to nameplate as long as the reservoir has water, which is the whole caveat, because in a drought year the credit and the energy fall together across an entire basin. Grids built on hydro (Norway, Quebec, British Columbia, Brazil) have never had to buy flexibility as a separate product, and interties such as NordLink and the North Sea Link exist largely to sell that flexibility to neighbors who do.

Three Gorges (22.5 GW) and Baihetan (16 GW) in China, Itaipu (14 GW, Brazil and Paraguay), Guri (10 GW, Venezuela), and Grand Coulee (6.8 GW, US) are the largest plants. Norway takes nearly all its electricity from hydro and uses the reservoirs to balance Northern Europe. On the failure side, the Kariba reservoir between Zambia and Zimbabwe fell so low in 2024 that both countries ran extended load shedding, and Brazil's 2021 drought pushed it onto expensive thermal generation for most of a year.

Hydro is almost pure capital: a very large civil works bill, then decades of near-zero fuel cost and O&M of roughly 1–2% of capex a year. That structure makes cost of capital and construction schedule the dominant variables, since a 6–12 year build accrues a lot of interest and cost overruns on large dams are the norm rather than the exception. Once the asset is paid off it is close to unbeatable, which is why century-old plants are still profitable and why relicensing fights are worth what they cost to fight. The live business in mature markets is refurbishment: replacing 60-year-old runners and generators typically buys several percent more output from the same water and the same permit, at a small fraction of new-build cost. New capacity is going in mostly in China, India, Africa, and Southeast Asia, where the binding constraints are financing and resettlement rather than engineering.

4 minHow a hydro generator worksTranspowerNZ · 500k+ views

6 minHow Francis Turbines Work (Hydropower)saVRee · 100k+ views

10 minHow to start and sync 432,000,000 Watt turbine manually in hydroelectric power plantDams explorer · 100k+ views

Types of Hydropower Plants (US Department of Energy) · 2026 World Hydropower Outlook (International Hydropower Association)

A run-of-river plant takes water from a river through a diversion weir and intake, carries it down a canal or penstock to a powerhouse, and returns it downstream, with little or no storage in between. Output therefore follows the river: high in the snowmelt or monsoon season, low in the dry months, and controllable only within whatever daily pondage the scheme has. That is the single distinction from a reservoir plant, and it changes the product from dispatchable power into variable but forecastable power. The machines are low-head types, Kaplan and bulb turbines in the large river schemes and cross-flow, Francis, or Pelton units in small mountain ones, and the civil works rather than the turbine account for most of the cost. Annual capacity factors typically run 40–60%, with a far lower figure in the dry season, so a plant that averages 50% over the year may deliver 20% for months at a time.

Large run-of-river plants on big rivers are genuinely cheap and much easier to permit than a storage dam, because there is no large reservoir, far less land inundated, and usually no resettlement. They are also synchronous machines, so they supply inertia and fault current that a PV or wind plant does not. The weakness is that the technology does not scale down gracefully. Below about 10 MW the cost per kW roughly doubles, because the intake, penstock, access road, powerhouse, and grid connection are largely fixed costs that a smaller machine has to carry. Seasonality is the other problem, since a plant whose low season coincides with the system peak is worth much less than its capacity factor suggests. The environmental failure mode is the dewatered reach between intake and tailrace, which is what drives minimum-flow requirements and fish-passage retrofits, and both of those permanently reduce output.

Run-of-river is worth building when a large river has consistent flow, the drop can be captured with modest civil works, and the seasonal shape matches the load, which is the case in hydrologically stable systems with summer peaks and snowmelt-fed rivers. Get a long flow record before anything else, ideally 20–30 years, because inter-annual variability in a river is much wider than in a wind or solar resource and a short record will flatter the project. Be skeptical of the small-hydro pitch. Below 10 MW the $/kW is bad enough that rooftop solar or solar plus storage usually beats it, and "distributed and easy" stops being true once you have priced the access road and the interconnection. Choose a reservoir plant instead when you need dispatchability and can carry the permitting; choose run-of-river when you want a cheaper, less contested project and can sell variable energy.

Annual capacity factor typically 40–60%, and far lower in the dry season · cost per kW roughly doubles below about 10 MW · Kaplan and bulb turbines for low head, cross-flow and Pelton for small high-head sites · $50–100/MWh at $1,500–3,000/kW for large schemes, $3,000–6,000/kW for small ones · 3–6 years to build · civil works rather than the turbine are most of the capital cost.

Output is variable but the generator is synchronous, so unlike PV and wind the plant supplies real inertia, fault current, and voltage support whenever it is running, which is why operators value a small hydro fleet more than its energy alone would suggest. Forecasting is good over days, because runoff responds to measured rainfall and snowpack, and poor over seasons, which is where the planning risk sits. Capacity credit depends entirely on whether the river's high season matches the system peak: a snowmelt plant in a summer-peaking system is credited well, and the same plant in a winter-peaking system is not. Schemes below about 10 MW connect at distribution voltage, which avoids the transmission interconnection queue but caps the project at whatever the local feeder can host.

Chief Joseph (2.6 GW on the Columbia) and Beauharnois in Quebec are large classic run-of-river plants. Belo Monte in Brazil is the cautionary version: 11.2 GW of installed capacity on a river with a large seasonal swing and almost no storage, so its average output is a fraction of nameplate. China operates on the order of 47,000 small hydro stations and Europe more than 20,000, most of them under 10 MW and many now facing fish-passage and minimum-flow retrofits under the Water Framework Directive.

The economics are dominated by civil works and by scale. A large run-of-river plant on a big river lands in the same $1,500–3,000/kW band as a reservoir plant with a shorter build and less permitting risk, which makes it one of the better renewable investments wherever the hydrology supports it. Small hydro is a different business: at $3,000–6,000/kW for a few megawatts, it only works with a feed-in tariff, a renewable obligation certificate, or an off-grid customer paying diesel prices. Operating costs are low and the assets last, so most of the activity in Europe and North America is refurbishment, re-equipping non-powered dams, and buying existing plants for their cash flow. The main risk to an existing asset is regulatory, because a relicensing round that imposes higher minimum flows or a fish-passage retrofit permanently reduces annual output, sometimes by more than 10%.

11 minHow Run of the River Hydroelectric Power Station Works3D-knowledge · 50k+ views

2 minVoith: Functioning of bulb turbines (EN)Voith Group · 10k+ views

15 minThe Genius of Small Hydro TurbinesUndecided with Matt Ferrell · 1m+ views

Types of Hydropower Turbines (US Department of Energy) · U.S. Hydropower Market Report, 2025 update (Oak Ridge National Laboratory)

"Tidal energy" covers two businesses that share only a driver. Tidal stream turbines sit in a fast current and work like submerged wind turbines, and after twenty years of exotic prototypes the industry has converged on three-bladed horizontal-axis machines of 1–2 MW: MeyGen in the Pentland Firth runs four 1.5 MW turbines for 6 MW total, and Orbital Marine's O2 is a 2 MW floating platform with two retractable rotor arms. Tidal range plants are barrages across an estuary that fill on the flood tide and discharge through low-head bulb turbines on the ebb, which is a civil-works problem rather than a turbine problem: La Rance in Brittany has run 240 MW since 1966 and Sihwa Lake in South Korea 254 MW since 2011. Water is about 800 times denser than air, so a tidal rotor takes the same power from a much smaller swept area at much lower speed, and it carries much higher structural loads for the same reason. The output is completely predictable years ahead from the tide tables and completely uncontrollable, and it falls to zero four times a day at slack water.

Predictability is the real product. You can publish a tidal plant's 2045 output today, which no other renewable can do, and the resource is uncorrelated with sun and wind, so it fills gaps a solar-and-wind portfolio leaves. The failure mode is economic rather than technical. Everything sits in seawater moving at several meters per second, so installation needs a dynamic-positioning vessel working a slack-water window, maintenance means recovering a machine of a few hundred tonnes to the surface, and biofouling, cavitation, and cable abrasion never stop. MeyGen has produced over 84 GWh since 2017, and a good year has been around 10 GWh, which on 6 MW is a realized capacity factor near 20% rather than the 30–40% the site is designed for — availability, not resource, is the binding constraint. Barrages avoid all of that and fail differently: they change the tidal prism of an estuary and the habitat that depends on it, which is why the repeatedly proposed multi-gigawatt Severn barrage keeps being rejected.

Pick tidal stream when you have a genuinely rare site (a constricted channel with peak spring currents above roughly 2.5 m/s, water 25–50 m deep, and a grid connection near the shore) and a buyer who will pay a policy price for predictable, non-correlated output. Cost is not the argument: MeyGen Phase 2 won a UK contract for difference at £178.54/MWh in 2012 prices, roughly three times what offshore wind has cleared in recent rounds. If you want cheap marine generation and the site allows it, build offshore wind, which uses the same vessels, cables, and ports at several times lower cost per MWh. Build a barrage only where the estuary is already heavily engineered and the ecological argument has already been settled; on a natural estuary, assume permitting kills it.

Viable stream sites need peak spring currents above roughly 2.5 m/s · MeyGen 6 MW from four 1.5 MW turbines, over 84 GWh generated since 2017 · realized capacity factor near 20% against a 30–40% design target · Orbital O2 2 MW on a floating platform · La Rance 240 MW since 1966 and Sihwa Lake 254 MW since 2011 · MeyGen Phase 2 CfD strike price £178.54/MWh in 2012 prices · four slack-water nulls a day.

Tidal is variable but perfectly forecastable, so an operator can schedule around it years in advance; what it cannot do is respond to demand, and its four daily nulls drift about 50 minutes later each day, so they walk straight through the evening peak twice a month. Capacity credit is therefore low, but forecast error is near zero, which removes the reserve cost wind and solar impose on a system. Arrays connect through inverters and supply no inertia, and barrages, which use synchronous bulb units, do. The bigger integration cost is location: the best channels are in remote places (the Pentland Firth, the Bay of Fundy, the Channel Islands), so a subsea export cable and a rural network reinforcement are usually part of the project.

MeyGen in the Pentland Firth (6 MW from four SIMEC Atlantis/Andritz turbines, now operated by Proteus Marine Renewables, with 59 MW of further CfD awards across the UK's AR4, AR5, and AR6 rounds); Orbital Marine Power's O2 at EMEC in Orkney; Nova Innovation's Shetland Tidal Array; Minesto's tidal kites in the Faroe Islands; Sabella in the Fromveur passage, France; La Rance (EDF) and Sihwa Lake (K-water) for tidal range; and the Bay of Fundy, home of the FORCE test site, which everyone cites as the world's best resource and where no array has yet survived commercially.

Almost all of the cost is capital spent in a hostile place: foundations or floats, subsea cable, and vessel time, with the turbine itself a minority of the bill. That means the cost curve depends on serial installation and on maintenance intervals stretching out, not on rotor aerodynamics, and it is why Orbital's floating platform is an explicit bet that a machine you can service alongside a quay beats one you have to lift off the seabed. The sector is small enough that a single supplier failing resets it, which has happened repeatedly (Marine Current Turbines, OpenHydro, Cape Sharp Tidal). Public money is effectively the whole market: the UK's ringfenced tidal-stream CfD budget is what keeps a pipeline alive. If you are underwriting a tidal business, the question is whether a few hundred MW of orders arrive before that ringfence is reviewed, because nothing about the technology gets cheap at 10 MW a year.

12 minTidal energy could be huge – why isn't it?DW Planet A · 1m+ views

5 minCheap, renewable, clean energy. There's just one problem.Tom Scott · 1m+ views

2 minTidal Power 101Student Energy · 1m+ views

Tidal Energy (Pacific Northwest National Laboratory) · Annual Report: An Overview of Ocean Energy Activities in 2025 (IEA Ocean Energy Systems)

Wave energy converters turn the orbital motion of surface water into electricity, and after fifty years of prototypes there is still no sustained multi-megawatt commercial plant anywhere in the world. The resource is real and large: good west-facing coasts in the North Atlantic and the Southern Ocean carry roughly 20–70 kW per meter of wave crest, and it peaks in winter, when demand does. The machines fall into families that have never merged — point absorbers (a buoy heaving against a reaction mass or the seabed), attenuators (hinged rafts lying along the wave), oscillating water columns (an air chamber driving a Wells turbine), overtopping devices (a ramp filling a small reservoir), and oscillating surge flaps. The hard engineering is the power take-off: waves deliver slow, reversing, irregular motion at periods of 5–15 s, so the machine has to rectify very high force at very low speed into something a generator can use, usually through hydraulics, a ballscrew, or a pneumatic stage. Every conversion step costs efficiency and adds a wearing part in salt water.

Wave power is spatially dense, seasonally well matched to winter demand in northern Europe, and available to countries with deep water close inshore and no shelf for fixed offshore wind. The failure mode is survival economics. A hundred-year wave carries orders of magnitude more energy than the machine's rated condition, so the structure and moorings are sized by a load the plant never produces any energy from, and that steel sits in the capital cost for the asset's whole life. Devices sturdy enough to survive are too expensive to pay back; devices cheap enough to pay back get destroyed, and the field's history is mostly the second case (Pelamis at Aguçadoura, Oceanlinx, Aquamarine Power's Oyster). The absence of a converged archetype after five decades is itself the diagnostic. Wind settled on the three-bladed upwind turbine within about fifteen years because the economics selected clearly; wave has never produced that signal.

Treat wave as a research position and a niche-power product rather than a generation choice. It makes sense today where the alternative is shipped-in diesel: navigation and monitoring buoys, offshore instrumentation, aquaculture, and small island microgrids. It also makes sense as a breakwater retrofit, where the civil structure is already paid for by coastal protection — Mutriku in the Basque Country has run an oscillating-water-column plant inside a harbour wall since 2011. If you need utility-scale marine generation, build offshore wind, which is on a real cost curve, or tidal stream if the site has current; both are years ahead. If you are an investor, the milestone worth waiting for is one device family running a multi-unit array through several winters at a declared availability figure. CorPower's DNV prototype certificate in July 2026 is progress toward that, not evidence of it.

Resource roughly 20–70 kW per meter of wave crest on good west-facing coasts · wave periods 5–15 s, so the power take-off works at around 0.1 Hz · Mutriku 296 kW from 16 Wells turbines, in service since 2011 · CorPower's Aguçadoura demonstration builds toward about 1.2 MW across up to four C4 units · VianaWave, the first proposed 10 MW commercial farm, targeted for roughly 2029–30 · realized capacity factors under 25%.

Wave is not currently a grid problem, because nothing is connected at scale. When it is, the shape helps: waves lag the wind that generated them by hours to a day and are far smoother than wind minute to minute, so a mixed wind-and-wave portfolio ramps less and has fewer zero hours than wind alone. Capacity credit is still low, because a settled high-pressure week flattens waves and wind together. What has to be built alongside is the same as for tidal — a subsea export cable, a shore substation, and reinforcement into a remote coastal node — and on a project of a few megawatts those costs dominate.

Mutriku (Spain — 296 kW of Wells turbines built into a harbour breakwater, the longest continuously operating wave plant); CorPower Ocean's C4 at Aguçadoura, Portugal, which took the world's first DNV prototype certificate for a wave energy converter in July 2026, with the 5 MW Valiant project at EMEC's Billia Croo in Orkney and the 10 MW VianaWave farm behind it; Eco Wave Power's EWP-EDF One at Jaffa Port in Israel and its US installation at the Port of Los Angeles; Wave Swell Energy's UniWave200 at King Island, Tasmania; Ocean Power Technologies' PowerBuoy, now sold as an autonomous offshore power and sensing platform rather than as a generator; and the cautionary set — Pelamis Wave Power, Oceanlinx, and Aquamarine Power, all insolvent.

There is no commercial cost curve to point at, only device-level estimates that no operating array has validated, so treat any wave LCOE as a projection rather than a measurement. The cost structure is the problem. A wave farm carries offshore wind's marine cost base — vessels, moorings, subsea cable, port and installation — spread across machines of 200 kW to 1 MW instead of 15 MW, so fixed marine cost per MW is roughly an order of magnitude worse. That is why the credible commercial paths are the ones that dodge that cost base: breakwater-integrated columns where a coastal-protection budget pays for the structure, and small autonomous units sold against diesel logistics. Funding is almost entirely public (Horizon Europe, the US DOE Water Power program, Wave Energy Scotland), and it has run for decades without producing a commercial plant. If you are diligencing a wave company, ask what it costs to recover and reinstall one unit, because that number, not the LCOE model, decides whether an array is operable.

11 minWhy wave power isn't everywhere (yet)DW Planet A · 100k+ views

3 minWhy Can't We Get Power From Waves?MinuteEarth · 1m+ views

2 minHow it works- CorPower Ocean Wave Energy ConvertersCorPower Ocean · 100k+ views

Wave devices (European Marine Energy Centre) · Wave (Tethys, Pacific Northwest National Laboratory)

A flash plant produces hot brine from a hydrothermal reservoir above roughly 180 °C, drops its pressure in a separator so part of the flow flashes to steam, runs the steam through a turbine, and reinjects everything else. This is the conventional geothermal plant and the most deployed type, a bit under half of world geothermal capacity, with a rarer dry-steam variant at the few fields that produce steam with no liquid phase. Production wells run 1–3 km deep and a good one supports 5–10 MW, so a 50 MW plant is a small field of wells feeding a central steam gathering system. Conversion efficiency is only about 15–20%, because the resource is lukewarm by power-plant standards, but the heat is free and always there, which is why capacity factors of 80–90% are normal — the highest of any renewable. The difficult part is not the power block, which is an ordinary steam turbine bought from Toshiba, Mitsubishi, Fuji, or Ormat; it is finding the reservoir, drilling it, and keeping it productive.

Geothermal delivers firm, synchronous, round-the-clock power at 80–90% capacity factor on a small footprint, with CO2 emissions well below any combustion plant and no fuel price to hedge. It also has the worst risk profile in generation at the front end. Exploration wells cost several million dollars each and are drilled before anyone knows whether the field produces, which is equity risk no lender will take, and a field can still disappoint after the plant is committed. Once running, the problems are chemical rather than mechanical: silica and calcite scale up pipes and reinjection wells, hydrogen sulfide and CO2 come out of solution as non-condensable gases that must be extracted (costing parasitic power) and abated, and reservoir pressure declines if reinjection does not keep pace with production. The Geysers in California is the standard illustration — the world's largest geothermal complex at roughly 725 MW net, derated over decades by pressure decline, and now recharged with treated municipal wastewater piped in from nearby towns.

Build a flash plant when you have a proven hydrothermal resource above about 180 °C, which in practice means Indonesia, the Philippines, Kenya, Turkey, Iceland, New Zealand, Mexico, Italy, Japan, or the US West. Then the economics are good: $66–109/MWh at 80–90% availability beats anything else that is firm and clean. Two things decide whether the project happens. First, who funds the exploration wells — usually a state utility, a development bank, or a drilling-risk insurance scheme, because commercial lenders will not. Second, whether the measured resource temperature holds above the flash threshold; if it comes in below roughly 180 °C, switch to a binary plant, which takes the same brine cooler and reinjects all of it. If your grid needs firm clean power and you have no hydrothermal resource at all, your comparison set is nuclear, enhanced geothermal, and gas with capture, not this.

Resource temperature above roughly 180 °C · production wells 1–3 km deep supporting 5–10 MW each · conversion efficiency roughly 15–20% · capacity factor 80–90% · unsubsidized LCOE $66–109/MWh at $5,000–6,460/kW · roughly 36 months of construction once the field is proven · world geothermal capacity about 17.2 GW at the end of 2025, with the top ten countries holding over 90% of it.

A flash plant is a synchronous steam plant, so it supplies inertia, fault current, and voltage support the way a coal or nuclear unit does, and system operators give geothermal a capacity credit close to its availability. It is economically must-run rather than technically inflexible: the wells want steady flow, cycling promotes scaling, and the fuel costs nothing, so operators run it flat and let something else follow load. Fields sit in volcanic terrain that is usually far from load, so a long transmission line is part of the project — Kenya's Olkaria and Indonesia's Sumatran fields both needed one. Curtailment exposure is low today but rising in high-solar markets, where a must-run plant absorbs midday negative prices it cannot avoid.

The Geysers (California, roughly 725 MW net, operated by Calpine, dry steam); Larderello (Italy, in production since 1913); Cerro Prieto (Mexico); Olkaria (Kenya, roughly 800 MW and the backbone of a grid that runs largely on geothermal and hydro); Sarulla and Gunung Salak (Indonesia); Wairakei and Ngatamariki (New Zealand); and Hellisheiði (Iceland), which sells power, district heat, and hosts the Carbfix project mineralizing CO2 into basalt.

Cost is front-loaded and mostly underground: drilling and the steam gathering system typically run a third to a half of project capital, and the power block is a commodity purchase. That structure creates the financing problem, because the riskiest dollars are spent first, before any revenue can be contracted, and it is why every successful geothermal country built a public risk-sharing mechanism (Kenya's GDC drills wells and sells steam; Indonesia and Turkey used state tariffs and drilling funds). Growth has been slow for the same reason: world capacity reached about 17.2 GW at the end of 2025 after adding only 223 MW in the year, against 647 GW of solar. Conventional geothermal is not going to scale much further, because the good hydrothermal fields are largely taken, and the growth story has moved to enhanced and closed-loop systems that try to remove the resource constraint rather than the cost.

4 minEnergy 101: Geothermal EnergyU.S. Department of Energy · 1m+ views

6 minHow Double Flash Geothermal Power Station Works3D-knowledge · 10k+ views

3 minHow Dry Steam Geothermal Power Station Works (EiM series)3D-knowledge · 10k+ views

Geothermal Electricity Generation (US Department of Energy) · The Future of Geothermal Energy, 2024 (IEA)

A binary plant never lets the geothermal brine flash. The brine runs through a heat exchanger that boils a secondary organic fluid — isobutane, isopentane, or a refrigerant — and that fluid drives the turbine and condenses in a sealed loop, while all of the brine goes back down the reinjection well. That buys three things at once. It opens up resources of 100–180 °C that a flash plant cannot use, it keeps non-condensable gases and dissolved solids underground so the plant has essentially no emissions and no scaling in the turbine, and it makes air cooling practical, so the plant can sit in a desert with no water. The price is thermodynamic: conversion efficiency is 10–13% against 15–20% for flash, so a binary plant needs more brine, more wells, and much larger heat transfer area per MW, and its own pumps and condenser fans eat 15–30% of gross output. Nearly every next-generation geothermal project uses an ORC surface plant, so this is the surface-side reference for enhanced and closed-loop systems.

Binary units are modular, standardized, and quick to install — Ormat ships repeatable skids rather than engineering each plant from scratch — and they make power from resources that would otherwise be non-commercial, including sub-boiling brine and oilfield produced water. Full reinjection also makes them the easiest geothermal plants to permit: no steam plume, no hydrogen sulfide, no water consumption when air-cooled. The weaknesses follow from the low efficiency. Capital per MW is high because the heat exchangers and air-cooled condensers are large, and output swings with ambient temperature. An air-cooled binary plant can lose a fifth or more of its output on a hot afternoon, which is exactly when the grid needs it, and that seasonal derate is the number most often missing from a pro forma. The working fluids are flammable hydrocarbons, so the plant carries fire protection and leak detection that a steam plant does not.

Use binary whenever the resource is below about 180 °C, whenever full reinjection is required by permit or by reservoir management, and whenever there is no cooling water. Above that temperature, flash is cheaper per MW because you are not paying for a second working fluid and its heat exchangers, and a flash plant with a binary bottoming unit captures both. If your resource is 100–150 °C, run the ambient-temperature sensitivity before committing, because at a hot, dry site the summer derate can move annual energy enough to change the deal. If you are evaluating a next-generation geothermal developer, remember that the surface plant is a solved, purchasable ORC and all of the real risk is subsurface — do not let a claimed cycle efficiency distract from the flow rate and temperature the wells actually deliver.

Resource temperature 100–180 °C · conversion efficiency 10–13%, against 15–20% for flash · parasitic load 15–30% of gross output · unit sizes from about 1 MW to 20 MW, ganged into larger plants · capacity factor 80–90% · LCOE around $100/MWh and up, the expensive end of geothermal's $66–109/MWh band · roughly 38% of world geothermal capacity · Ormat operates about 1,140 MW and some 500 wells, the largest owner-operator in the industry.

Binary behaves like flash on the grid — firm, synchronous, high capacity credit, run flat because the heat is free — with one difference that belongs in a planning study. An air-cooled unit's output falls with ambient temperature, so its contribution at a summer peak is below nameplate and a resource adequacy model should derate it rather than credit it at 90%. Binary units are small and modular enough to be added at an existing steam field or on an oilfield water-handling site, which sometimes lets them reuse an interconnection instead of joining the queue. That matters more than it sounds when the median US project waits over five years for a connection.

Ormat's fleet is most of the category — Puna (Hawaii), Steamboat and McGinness Hills (Nevada), Olkaria III (Kenya), and Ijen (Indonesia, 35 MW, commercial in 2025); Turkey's large binary fleet, built almost entirely on 130–180 °C resources; Unterhaching and Landau in Germany; Chena Hot Springs in Alaska, which makes power from 74 °C water and marks the low-temperature extreme of the technology; and Exergy, Turboden, and Atlas Copco as the other ORC suppliers.

Binary plants are the closest geothermal gets to a manufactured product: a handful of suppliers build repeatable ORC modules, so surface costs are far more predictable than the drilling that feeds them. Ormat's business model shows where the margin sits — it designs and manufactures the equipment, then owns and operates the plants, capturing both the product margin and the generation revenue, and it holds roughly 1,140 MW of its own capacity. Capital still lands inside Lazard's $5,000–6,460/kW geothermal band, whose LCOE range is $66–109/MWh, and binary sits at the expensive end of it because of the extra heat-transfer area and the parasitic load — call it $100/MWh and up, against flash steam (014) nearer the bottom of the same band. The growth market is not new hydrothermal fields, which are scarce; it is ORC surface plants sold into enhanced and closed-loop geothermal projects and into oilfield and industrial waste heat, where the heat source is already paid for by something else.

10 minHow Binary Geothermal Power Station Works3D-knowledge · 10k+ views

3 minThe ORC Process - Atlas Copco Solutions for Geothermal and Waste Heat Recovery PlantsAtlas Copco North America · 10k+ views

Geothermal power plants (US Energy Information Administration) · Simulation of Air-Cooled Organic Rankine Cycle Geothermal Power Plant Performance (Idaho National Laboratory)

An enhanced geothermal system makes a reservoir instead of finding one. You drill into hot, impermeable rock — typically granite at 3–5 km and 180–230 °C — kick a horizontal lateral, and hydraulically stimulate it in stages to open a fracture network between an injector and a producer, then circulate water through those fractures and run the hot water through an ORC surface plant. Every technique in that sentence came out of shale: multi-stage plug-and-perf fracturing, horizontal drilling, polycrystalline diamond bits, and fiber-optic distributed acoustic and temperature sensing to see where the flow actually goes. The consequence is that geothermal stops being a resource business restricted to volcanic provinces and becomes a drilling-cost business that works wherever hot rock is reachable, which is most places. Fervo Energy is the company that made this credible, going from a 3.5 MW pilot in Nevada in 2023 to a 500 MW field under construction in Utah, with a 2 GW ambition behind it.

EGS keeps everything good about geothermal — 80–90% capacity factor, firm synchronous output, small footprint, no fuel price — and removes the exploration lottery, because you drill to a temperature you can predict from a thermal gradient rather than hunting for natural permeability. The cost trend is real: Fervo's drilling times fell roughly 70–75% between 2022 and 2025, and its cost per kW fell from about $7,000 in Cape Station Phase 1 to about $5,500 in Phase 2, against a $3,000/kW target. Two risks are unresolved and both are underground. Thermal drawdown is the first: circulating water cools the rock nearest the dominant fractures, and nobody has run a stimulated reservoir for the 20–30 years a project is financed over, so the decline curve is a model rather than a measurement. Induced seismicity is the second: stimulation makes earthquakes, usually too small to feel, occasionally not. Basel was canceled after a magnitude 3.4 event in 2006, and Pohang was shut after a magnitude 5.5 in 2017 that injured people and did hundreds of millions of dollars of damage. One bad event near a city sets the whole field back.

Consider EGS when you need firm clean power on a date, you are somewhere with a decent thermal gradient and no hydrothermal resource, and you can accept a technology whose long-run decline rate has not been measured. That describes the buyers actually signing contracts: hyperscalers and utilities comparing it against new nuclear at $141–220/MWh and gas turbines they cannot get delivered before 2030. If the comparison is purely price per MWh against a combined cycle at $48–109/MWh, EGS loses today. Site it away from population and instrument it for seismicity from the first stimulation, with a published traffic-light protocol, because the binding risk is social rather than technical. And if you need firm capacity within three years and you can get gas, buy the gas plant — EGS is a five-year proposition even when the drilling goes well.

Target rock typically 3–5 km deep at 180–230 °C · Project Red 3.5 MW in 2023; Cape Station's first 100 MW due around October 2026 with 400 MW more slated for 2028 · Phase 1 about $7,000/kW and Phase 2 about $5,500/kW, against a $3,000/kW target · drilling times down roughly 70–75% since 2022 · capacity factor 80–90% · Fervo's May 2026 IPO priced at $27 a share and opened near a $10B valuation · Pohang's 2017 induced earthquake reached magnitude 5.5.

EGS is a synchronous ORC plant, so it supplies inertia and gets a capacity credit close to its availability, and a planner can treat it as a small firm unit. Its more interesting property is that a stimulated reservoir doubles as a short-duration store: throttle the production well and pressure builds in the fracture network, then release it for a few hours of above-rated output, which Fervo has demonstrated and which turns flat 100 MW into something that can lean into the evening peak. That flexibility is worth more than the energy in a solar-heavy market, and it is the main reason a utility would take EGS over an equally firm alternative. What has to be built alongside is conventional — a gathering system, the ORC plant, and a transmission connection — but in the Great Basin the transmission is usually new line rather than a spare bay.

Fervo Energy's Project Red in Nevada (3.5 MW, supplying Google) and Cape Station in Beaver County, Utah (100 MW in 2026, 500 MW contracted, with utility and hyperscaler offtake including Southern California Edison and Shell Energy); Utah FORGE, the DOE field laboratory at Milford that de-risked much of the technique; Soultz-sous-Forêts in Alsace, the long-running European EGS demonstration; Basel and Pohang, the two projects that induced seismicity ended; and Sage Geosystems and XGS Energy as adjacent subsurface approaches that have also signed data-center offtake.

EGS economics are drilling economics. The wells are most of the capital, so the cost curve is set by rate of penetration, bit life, and how much productive lateral you can stimulate per dollar, and all three improve with repetition in the same rock. That is why Fervo's cost fell between phases of one field rather than between fields, and why the $3,000/kW target depends on continuing to drill the same formation with a crew that has already learned it. The demand side is unusually favourable: data-center buyers want firm clean power on a fixed date and will pay well above a gas benchmark for it, so the sector had contracted revenue before it had a cost curve. The risk worth watching is the gap between a $10B valuation and a decline curve nobody has measured. If drawdown at Cape Station turns out steeper than modeled, the whole category reprices at once, because every EGS pro forma leans on the same assumption.

10 minEnhanced Geothermal: A Groundbreaking Energy SourceTDC · 100k+ views

2 minHow Fervo Energy’s Next-Generation, Enhanced Geothermal Systems WorkFervo Energy · 5k+ views

24 minThe Big Business of Drilling Deep Into the Earth | Bloomberg PrimerBloomberg Originals · 500k+ views

Enhanced Geothermal Systems (US Department of Energy) · FORGE, the Frontier Observatory for Research in Geothermal Energy (US Department of Energy)

These are the two ways to get geothermal heat without stimulating a reservoir. A closed loop drills a sealed circuit — usually two vertical wells joined by many horizontal multilaterals — and circulates working fluid inside the pipe, so nothing ever contacts the rock. That removes induced seismicity, water loss, and downhole scaling in one move, and it works in rock with no permeability at all. It also removes convection, so heat now reaches the pipe by conduction alone through rock with a thermal conductivity of roughly 2–3 W/m·K, which means power per meter drilled is low and you buy output by drilling more wellbore. Superhot and supercritical concepts go the other way and chase temperature instead of surface area, targeting 400 °C and above, where water goes supercritical and a single well could in principle carry five to ten times the power of a conventional one. Both reduce to the same question: what does a meter of hole cost, and how much power does that meter return?

Closed loops solve the two things regulators and neighbours object to in enhanced geothermal, and they work anywhere, which is a very large addressable market if the arithmetic closes. It has not closed yet. Eavor's Geretsried plant in Bavaria delivered the first closed-loop electricity to a commercial grid in December 2025, but at roughly 0.5–2 MW gross against a design near 8 MW electric, with only six of twelve planned lateral pairs drilled — which is precisely the conduction-limited outcome the physics predicts if you stop drilling early. Superhot rock has a bigger prize and a harder problem: no drill bit, casing, cement, or downhole electronics survives long above about 400 °C, which is why Quaise is trying to vaporize rock with millimeter waves instead of grinding through it. Quaise has raised $230M in total, is running a 100 kW gyrotron toward a 1 MW unit, and is about a kilometer deep at its central Texas field site, with first grid power from Project Obsidian in Oregon targeted for 2030. Neither approach has a plant delivering rated output, and both are betting on a drilling cost curve that has not yet bent.

Do not pick either of these for a project you need this decade. Track them, and treat them as options on drilling cost rather than as generation you can procure. If you are a developer choosing between enhanced geothermal and a closed loop at a specific site, the deciding question is permeability and seismic tolerance: where the rock will fracture and the community will accept stimulation, EGS returns far more power per meter drilled; where it will not, a closed loop is the only option, and you should size it on conduction rather than on a vendor's peak-output curve. If you are an investor, the honest test is cost per meter of hole and power per meter of hole, both measured on a completed well, because everything else in the pitch sits downstream of those two numbers. For superhot, the milestone to wait for is a bit and casing that survive a full run at temperature.

Rock thermal conductivity roughly 2–3 W/m·K, which is what caps a closed loop's output per meter · Eavor's Geretsried loop delivered first grid power in December 2025 at roughly 0.5–2 MW gross against an 8 MW design, with 6 of 12 lateral pairs drilled · water goes supercritical above about 374 °C and 220 bar, the target for five to ten times the power per well · Quaise has raised $230M and is about 1 km deep with a 100 kW gyrotron, moving to 1 MW · Project Obsidian targets first grid power in 2030 · capacity factor 80–90% if either works, since the heat source never varies.

On paper both deliver what conventional geothermal delivers: firm, synchronous output with a capacity credit a planner can count at peak. The closed loop adds one useful trick, because the thermosiphon can be throttled and the loop's thermal mass used to hold back a few hours of output, so it load-follows better than a hydrothermal plant does. Since neither depends on a hydrothermal resource, both can in principle be sited next to load or next to an existing interconnection, and avoiding a five-year queue is most of their strategic appeal. Nothing here is proven at grid scale, so any system plan that counts on it before 2030 is counting on a demonstration rather than a resource.

Eavor's Geretsried project in Bavaria (first closed-loop grid power, December 2025, funded largely by a €92M EU Innovation Fund grant) and the earlier Eavor-Lite demonstration in Alberta; Quaise Energy's millimeter-wave field site in central Texas and Project Obsidian in central Oregon; XGS Energy's conductive closed-loop wells in Nevada and California; Iceland's IDDP-1 at Krafla, which drilled into magma at around 900 °C in 2009, and IDDP-2 at Reykjanes; and Sage Geosystems, which applies the same "engineer the wellbore, not the reservoir" idea to pressure storage as well as heat.

The whole case is drilling cost per meter, and today it is roughly an order of magnitude too high for a conduction-limited loop to compete. A closed loop needs tens of kilometers of hole per plant, so a 20% cut in drilling cost moves the answer more than any surface-plant improvement could. Both approaches therefore depend on the same oilfield learning curve that enhanced geothermal is already riding, with the difference that EGS gets convective heat transfer from a fractured reservoir for free and these do not. Funding is grant-heavy and strategic — the EU Innovation Fund at Geretsried, venture rounds at Quaise and XGS, oil-major balance sheets in the background — which is the right shape for technologies whose value is optionality. If you are building a business here, the defensible asset is a drilling capability, not a plant design, because the surface plant is a purchased ORC either way.

4 minThe Thermodynamics of Eavor-Loop™Eavor · 10k+ views

18 minThe Most Advanced Drill in Human HistoryDr Ben Miles · 500k+ views

14 minHow to Build a Superhot Geothermal Power PlantQuaise Energy · 10k+ views

Next-Generation Geothermal (National Laboratory of the Rockies) · Bridging the Gaps for Superhot Rock Geothermal Energy (Clean Air Task Force)

A trough plant is a field of parabolic mirrors, each tracking the sun on one axis and focusing sunlight onto a vacuum-jacketed receiver tube running along its focal line. Synthetic oil — a biphenyl and diphenyl oxide eutectic — is pumped through the tubes at up to about 393 °C, above which it decomposes, and hands its heat to an ordinary water-steam Rankine cycle. That temperature ceiling is the whole story of the format. It holds the steam cycle to roughly 37–38%, and it makes molten-salt storage awkward, because you need an oil-to-salt heat exchanger and the salt tanks then work over a temperature difference of about 95 °C instead of a tower's 275 °C, so you buy far more salt per stored kWh. Troughs are still the dominant CSP format, roughly two-thirds of installed capacity worldwide, and the SEGS plants in the Mojave have run since the 1980s, which gives the format an operating record no other solar-thermal design has. Annual solar-to-electric efficiency lands around 14–16%.

Troughs are the most bankable CSP design, with the longest track record, the simplest optics (one-axis tracking on identical modules), and the shortest list of surprises. Adding 6–7.5 hours of storage is routine, and a trough plant with storage is genuinely dispatchable solar driving a synchronous turbine into the evening. The weaknesses all trace back to 393 °C: lower efficiency than a tower, storage that costs more per kWh because of the small temperature difference and the extra heat exchanger, and a heat transfer fluid that is flammable, degrades over time, and leaks, with hot-oil fires having taken plants offline. The deeper problem is industrial rather than thermal. A trough plant is site-built steel, glass, and concrete, so it never got a factory learning curve; PV module prices fell about 90% in a decade while trough capital cost did not move much at all.

Build a trough plant when you want dispatchable solar with a synchronous turbine in a high direct-normal-irradiance desert and a policy buyer is paying for it. That describes essentially every recent trough project: Morocco, the UAE, China, and Spain under its old feed-in tariff. If you are choosing purely on the cost of evening energy, build PV with a 4–8 hour battery instead, which delivers the same product for less almost everywhere and is why Western trough development stopped. If you do want CSP and need more than about 8 hours of storage or process heat above 400 °C, use a power tower, where 565 °C salt makes both cheap. And if the site's direct normal irradiance is below roughly 2,000 kWh/m² a year, do not build CSP at all — troughs concentrate beam radiation and get nothing from diffuse light, so a hazy or cloudy climate hurts them far more than it hurts PV.

Heat transfer fluid capped near 393 °C, giving a steam cycle around 37–38% · annual solar-to-electric efficiency roughly 14–16% · storage typically 6–7.5 hours, across a salt temperature difference of about 95 °C against a tower's 275 °C · roughly two-thirds of world CSP capacity, which is itself still under 10 GW · unsubsidized LCOE in the $100–200/MWh band at $3,000–7,000/kW · sites need direct normal irradiance above roughly 2,000 kWh/m² a year · SEGS has operated in the Mojave since the 1980s.

A trough plant with storage is a synchronous steam plant, so it supplies inertia, fault current, and voltage support, and an operator can dispatch it — that is the product, and it is why grids with high solar penetration still run CSP tenders. Capacity credit is high with storage and poor without it, since a storage-less trough plant fades with the sun exactly like PV. It ramps in minutes rather than seconds and has a minimum stable load below which the turbine has to come off, so it is a load-following resource and not a frequency-response one. Curtailment exposure is low, because the operator can hold heat in the tanks through the midday price collapse and generate in the evening, and that ability to move energy across the day without a battery is most of what a hybrid solar tender is buying.

SEGS in the Mojave (354 MW across nine plants built 1984–1990, the founding fleet); Andasol 1–3 in Spain (50 MW each, the first troughs with molten-salt storage, 7.5 hours); Solana in Arizona (280 MW with 6 hours of storage) and Mojave Solar (280 MW); Spain's roughly 2.3 GW trough fleet built under the 2008–2013 feed-in tariff; Noor I and Noor II at Ouarzazate, Morocco (160 MW and 200 MW); Shams 1 in Abu Dhabi (100 MW); and the 600 MW of troughs inside Dubai's Noor Energy 1, the largest single CSP project ever built.

Trough cost is civil and mechanical work at scale — mirrors, receiver tubes, drives, pylons, hot-oil piping, and a steam plant — and none of it rides a manufacturing curve the way a PV module does, so the format's cost has been roughly flat in real terms while its competitor's fell by an order of magnitude. The supply chain is correspondingly thin: Rioglass for mirrors, a short list of receiver-tube makers, and Abengoa (since restructured), ACS Cobra, and Chinese EPCs for construction. Where troughs still get built, the buyer is a state tender that explicitly values dispatchability, local content, or industrial policy, not a merchant developer choosing on price. If you are building a business around CSP components, the durable demand is more likely to be industrial process heat and hybrid tenders in China and the Gulf than merchant power anywhere.

2 minEnergy 101: Concentrating Solar PowerU.S. Department of Energy · 100k+ views

18 minVIRTUAL VISIT OF A PARABOLIC TROUGH SOLAR THERMAL POWER PLANTRENOVETEC INSTITUTE · 5k+ views

Linear Concentrator System Concentrating Solar-Thermal Power Basics (US Department of Energy) · How CSP Works: Tower, Trough, Fresnel or Dish (SolarPACES)

A power tower surrounds a central receiver with a field of two-axis heliostats that hold the sun's image on it all day. Molten nitrate salt is pumped up the tower, heated from about 290 °C to about 565 °C, and stored in a hot tank, and steam is raised from that tank whenever the operator wants power. The design decision that matters is that the salt is both the heat transfer fluid and the storage medium, so there is no oil-to-salt heat exchanger and the tanks work over a 275 °C temperature difference instead of a trough's 95 °C. Storage therefore costs little more than a bigger tank and more salt, which is why 10–15 hours is normal and capacity factors of 45–70% are achievable, making this the only solar technology that is dispatchable without batteries. The turbine is an ordinary synchronous steam machine, so the plant also supplies inertia and fault current, which is a real product on a grid full of inverters.

Everything the tower does better than a trough comes from temperature: a better steam cycle, cheap storage, and heat grade high enough for industrial use. In operation the salt-storage towers do what they promise, and Cerro Dominador in Chile and Noor III in Morocco both run into the night. The weaknesses are execution and market. Every tower is a first-of-a-kind construction project, with a receiver that sees enormous flux and daily thermal cycling, thousands of heliostats to calibrate and wash, and salt that freezes at about 220 °C, so trace heating has to run for the plant's life or the loop solidifies. Crescent Dunes is the cautionary case: 110 MW with 10 hours of storage, roughly a billion dollars of capital, a hot-salt tank leak, Chapter 11 in January 2026, and a sale for $7 million in cash that March. Ivanpah shows the other failure, which is being on the wrong side of the PV cost curve — 392 MW with no storage at all, and PG&E moved in January 2025 to terminate its contracts and shut two units, though the California PUC has so far rejected those termination agreements.

Pick a tower over a trough whenever you want more than about 8 hours of storage or heat above 400 °C, because 565 °C salt makes both cheap and 393 °C oil makes both expensive. Pick a tower over PV plus batteries only in a narrow set of cases: very long daily duration (12+ hours, where salt's cost per kWh-thermal beats cells), a desert grid that specifically needs synchronous inertia and fault current, or a tender that bundles CSP with PV and pays explicitly for the firming. Outside those, PV plus a 4–8 hour battery wins on both cost and schedule, which is why Western tower development effectively stopped after Ivanpah and Crescent Dunes. If you are bidding into China's hybrid CSP-plus-PV tenders the calculation is different, because the CSP block is being bought as the firming asset for a multi-gigawatt renewable base rather than as a generator standing on its own.

Salt heated from about 290 °C to about 565 °C, a 275 °C storage temperature difference · 10–15 hours of storage typical, giving capacity factors of 45–70% · nitrate salt freezes near 220 °C, so trace heating runs permanently · capital cost above $7,000/kW with LCOE in the $100–200/MWh band · China connected 9 CSP plants in 2025 for 27 total and about 1.74 GW · Crescent Dunes, 110 MW with 10 hours of storage, sold out of bankruptcy for $7M in March 2026.

This is the most grid-friendly solar plant that exists: a synchronous steam turbine supplying inertia, fault current, reactive power, and in some configurations black start, with 10–15 hours of stored heat behind it, so an operator can schedule it like a gas plant and accredit it near its availability. It ramps in minutes rather than seconds, so it is a load-following resource that still needs something faster alongside it for frequency response. Curtailment exposure is close to zero, because the plant banks heat instead of exporting at midday, and that property is exactly what China's hybrid tenders are buying when they pair a 100 MW CSP block with several hundred MW of PV and wind. What has to be built alongside is transmission, since the direct-normal-irradiance sites are deserts a long way from load.

Cerro Dominador in Chile (110 MW with 17.5 hours of storage); Noor III at Ouarzazate, Morocco (150 MW, 7 hours); Redstone in South Africa (100 MW, 12 hours, online 2024–25); Ashalim/Megalim in Israel (121 MW); Crescent Dunes in Nevada (110 MW, the bankruptcy case) and Ivanpah in California (392 MW, direct steam, no storage, being closed unit by unit); and Delingha, Dunhuang, and the 2023–25 wave of Chinese hybrid CSP-plus-PV bases in Qinghai, Gansu, and Xinjiang, which is where nearly all current construction sits.

Nearly all of a tower's cost is the heliostat field and the receiver-and-tower assembly; the storage is the cheap part, and the energy-storage sheet covers two-tank molten salt on its own terms. That split explains the market. Storage hours are almost free to add and the collector is not, so CSP loses to PV whenever a buyer only wants energy, and wins occasionally when a buyer wants energy and firmness in one contract. Western supply has largely exited — BrightSource, SolarReserve, and Abengoa are gone or restructured — and China now holds the industrial base, with 27 plants connected, about 1.74 GW installed, and a 15 GW target in its next Five-Year Plan. If tower costs come down, that is where it happens. For anyone building a business on CSP components, the addressable market is Chinese and Gulf tenders plus high-temperature industrial heat, not the merchant power market Ivanpah and Crescent Dunes were built for.

4 minConcentrating Solar Power-Power TowersKeepItCleanCreative · 100k+ views

4 minSolar Reserve Solar Thermal Solar Power tower with molten salt storage based on Solar TwoMatthew Wright · 10k+ views

13 minThis Billion Dollar Solar Plant was an EPIC FailureTwo Bit da Vinci · 1m+ views

Power Tower System Concentrating Solar-Thermal Power Basics (US Department of Energy) · How solar thermal energy storage works with concentrated solar (SolarPACES)

A combined-cycle plant is a gas turbine whose exhaust, still around 600 °C, raises steam in a heat-recovery steam generator to drive a second, steam turbine. That bottoming cycle is the whole invention: it lifts fuel-to-electricity conversion from the 35–44% a gas turbine manages on its own to 60–64% on a lower-heating-value basis with a modern H-class machine, the highest of any thermal power plant ever built. A 1x1 block (one gas turbine, one steam turbine) runs 400–850 MW depending on frame size, and a 2x1 runs two gas turbines into one steam turbine for 800–1,700 MW. This is the reference dispatchable plant, the thing every clean-firm proposal gets benchmarked against, because it is cheap per MW, cheap per MWh at reasonable gas prices, follows load over minutes to hours, and can be built in about two years once the turbine arrives. Everything hard about a CCGT in 2026 is in that last clause.

The efficiency is the strength and it compounds. At the 6,475–6,550 Btu/kWh Lazard models, a combined cycle burns roughly 40% less gas per MWh than a simple-cycle peaker, which is why it can run 60% of the year and still make money, and its CO2 per MWh is less than half a coal plant's. It is also flexible enough for a renewables-heavy grid: a warm start to full load in 30–60 minutes, ramp rates of tens of MW a minute, and a minimum stable load around 30–40%. The weaknesses are exposures rather than engineering flaws. Fuel is a pass-through, so every $1/MMBtu of gas price moves the cost of power by about $6.50/MWh and a gas LCOE is really a gas-price forecast. Frequent cycling is what actually breaks these plants: the heat-recovery steam generator's thick-walled headers and drums crack from thermal fatigue when the plant starts daily instead of seasonally, and that maintenance bill is usually missing from the original model. And a new plant is a 30-year carbon commitment in jurisdictions that keep changing their minds about carbon.

Pick a CCGT when you need bulk dispatchable energy at a capacity factor above roughly 40%, you have firm gas supply, and you can get a delivery slot. It is the cheapest firm power available at $48–109/MWh and nothing clean is close. Two things should stop you. First, delivery: roughly 80 GW of turbines were ordered in 2024 against about 30 GW a year of OEM capacity, GE Vernova's backlog passed 116 GW by mid-2026, Siemens quotes lead times over 40 months, and slots are being negotiated for 2029–30 with reservation deposits around 20% of the equipment price. If you need power before 2030, you may simply not be able to buy one. Second, the benchmark you are really competing against: an existing CCGT's marginal cost is $24–39/MWh, so a new plant has to beat the incumbent fleet's fuel bill, not just its own LCOE. If your capacity factor will be under about 15%, buy a frame peaker or a battery instead, because the steam bottoming cycle only pays for itself on running hours.

60–64% LHV efficiency on H-class machines, against 35–44% in simple cycle · heat rate 6,475–6,550 Btu/kWh in Lazard's 2025 model · unsubsidized LCOE $48–109/MWh · capex $1,200–1,600/kW nominal, $2,400–2,600/kW for post-2028 operation, with installed cost having moved from about $1,000/kW to $2,000–2,800 · US fleet capacity factor 58.4% in 2025 · about $6.50/MWh of cost per $1/MMBtu of gas price · marginal cost of an existing plant $24–39/MWh.

A CCGT is a large synchronous machine, so it supplies inertia, fault current, reactive power, and governor response as a by-product of running, and operators accredit it near 100% of nameplate for resource adequacy — which is why a system can replace one CCGT with several times its nameplate in solar and still be short of firm capacity. It follows load well but not instantly: 30–60 minutes from a warm start, tens of MW a minute once running, and a floor near 30–40% load, so it needs batteries or engines alongside it for second-to-minute regulation. Its main integration cost is upstream rather than on the wires, in firm gas supply and pipeline capacity for a winter peak, which is what failed in Texas in 2021 and across the US Southeast in 2022. On the wires it is an advantage: one large interconnection close to load, against the new transmission an equivalent amount of remote renewable capacity would need.

GE Vernova's 9HA and 7HA, Siemens Energy's SGT-9000HL, and Mitsubishi Power's M501/701 JAC are the three H-class lines that take nearly all new orders; EDF's Bouchain and Chubu Electric's Nishi-Nagoya were the first plants certified above 62% and 63% net efficiency; the roughly 280 GW US combined-cycle fleet is the largest dispatchable resource in any grid on earth; and the 2024–26 US buildout is dominated by data-center-adjacent projects such as the Chevron and Engine No. 1 development in West Texas and the Homer City redevelopment in Pennsylvania.

The cost structure inverted between 2021 and 2026. A CCGT used to be a cheap, quickly delivered plant whose economics were dominated by fuel; now the equipment is scarce, the installed capital number has roughly doubled from about $1,000/kW to $2,000–2,800, and gas is still the swing cost on top. Three OEMs — GE Vernova, Siemens Energy, and Mitsubishi Power — hold effectively the entire H-class market, and having been burned by the 2015–2020 downturn they have chosen to convert scarcity into price and 20% reservation deposits rather than into new factories. For a developer, that means the scarce assets are the turbine slot and the interconnection, not the engineering. For anyone selling clean-firm power, the consequence is favourable: the benchmark competitor cannot be delivered until 2030, which is why geothermal, nuclear, and fuel cells are signing contracts at prices a combined cycle would have beaten easily in 2021.

5 minCombined Cycle Power Plants Theory Overview (complete guide for power engineering)Technical Engineering School · 100k+ views

3 minJourney to the heart of Energy - How a combined cycle gas turbine power plant worksEDF · 50k+ views

5 minHeat Recovery Steam Generator (HRSG) ExplainedsaVRee · 50k+ views

Natural gas combined-cycle power plants increased utilization with improved technology (US Energy Information Administration) · Cost and Performance Baseline for Fossil Energy Plants, Volume 1: Bituminous Coal and Natural Gas to Electricity (NETL, 2022)

A frame peaker is a heavy-duty industrial gas turbine exhausting straight to the stack, with no steam bottoming cycle behind it. It is the same class of machine that sits at the front of a combined-cycle plant — a large single-shaft turbine on a rigid frame, typically 50–300 MW — minus the heat-recovery steam generator and steam turbine that would have recovered the exhaust heat. Dropping that equipment roughly halves the capital cost per kW and cuts efficiency from over 60% to 35–40%, a heat rate of 10,275–11,175 Btu/kWh, about 65% higher than a combined cycle's. It reaches full load in roughly 10–30 minutes, and fast-start packages do better than that. You do not buy this plant to make energy. You buy it to have capacity available, and most of its revenue comes from being ready rather than from running.

A frame peaker is the cheapest way to own a dispatchable megawatt: $1,150–1,450/kW, no water-steam cycle to maintain, dual-fuel capability so it can burn stored distillate when the gas system is short, and black-start capability in some configurations. Its weaknesses are the reason it exists in the first place. Fuel cost per MWh is about 65% above a combined cycle's, so LCOE at a realistic duty cycle is $149–251/MWh and the marginal cost of an existing peaker is $47–170/MWh. NOx and noise make urban siting hard, and peakers sit disproportionately in load pockets, which is where the air-quality and environmental-justice objections are loudest. The failure mode that matters most is start reliability: a plant that runs 200 hours a year has to work on the coldest morning of the decade after months of sitting idle, and start failures during cold snaps have driven the largest reliability events in US grids, including Uri in 2021 and Elliott in 2022.

Pick a frame peaker when you need firm capacity for the tail of the load-duration curve at the lowest cost per kW, you expect to run it under about 15% of the year, and you can get the emissions permit. Judging it on LCOE is a category error: the product is availability, so the number to model is $/kW-year against the capacity market or scarcity revenue, not $/MWh. Its competitor is now a four-hour battery, which beats it on ramp (seconds against tens of minutes), on emissions, and on siting, and which has already taken most of the daily evening-peak duty in California and Texas. What the battery cannot do is run for three days, which is exactly the stretch that pays for a peaker's whole year. If you need starts in under ten minutes, or many starts a day, buy an aeroderivative and accept the higher cost per kW; if you need capacity above roughly 40% utilization, buy a combined cycle; and below about 20 MW, buy reciprocating engines.

Simple-cycle efficiency 35–40% at 10,275–11,175 Btu/kWh, about 65% above a combined cycle's heat rate · capex $1,150–1,450/kW · LCOE $149–251/MWh at a 10–15% capacity factor, and often far lower utilization than that · marginal cost of an existing peaker $47–170/MWh · full load in roughly 10–30 minutes · unit sizes typically 50–300 MW · PJM's 2026/27 capacity auction cleared at the $329.17/MW-day cap in every zone.

A peaker's entire purpose is grid integration. It is a synchronous machine accredited near 100% of nameplate, it supplies inertia and fault current whenever it runs, and it is often the black-start resource for a region. In a solar-heavy market its job has moved from covering the summer afternoon, which batteries now do, to covering the multi-day event — a still, cold week, or a winter morning after storage has been drained — so its value is concentrated into a few dozen hours a year and depends entirely on starting when called. That makes fuel security part of the interconnection question: firm pipeline capacity or on-site distillate storage is what turns a nameplate megawatt into an accredited one, and several ISOs have tightened accreditation rules to say so. Because peakers are sited in load pockets to relieve transmission constraints, they generally need less new network than an equivalent renewable project would.

GE Vernova's 7E/7F and 9E/9F frames and Siemens Energy's SGT6-5000F in simple-cycle packages are the workhorses of the fleet; the PJM, ERCOT, and CAISO peaker fleets built during the 1999–2003 merchant boom are most of the US installed base; New York City's in-city peakers (Ravenswood, Astoria, Gowanus) are the documented case of urban peakers under environmental-justice pressure, with the state's peaker rule forcing repowering or retirement; and the Texas Energy Fund's loans for new dispatchable capacity are the current example of a state paying directly for peaking plant.

Peaker economics are capacity-market economics, and the capacity market has repriced hard. PJM's 2026/27 auction cleared at the $329.17/MW-day cap in every zone after two auctions of steep increases, which turns a plant with almost no energy revenue into a financeable asset. ERCOT, which has no capacity market, pays for the same thing through scarcity pricing, a much lumpier and less bankable revenue stream. Capital cost per kW is among the lowest of any thermal plant, though a bank of reciprocating engines undercuts it at the bottom of its range, but a peaker order sits in the same OEM queue as a combined cycle, so the delivery constraint applies here too and has pushed buyers toward aeroderivatives and reciprocating engines that can actually be delivered. The strategic question for anyone owning one is how many of its remaining hours storage takes next: four-hour batteries have already absorbed most evening-peak duty in California, and each additional hour of storage duration eats the peaker's niche from the short end.

2 minHow a gas turbine works | GE VernovaGE Vernova · 100k+ views

10 minWhat is a Gas Turbine? (For beginners)RealPars · 500k+ views

6 minWhat is a Peaker Power Plant?Clean Energy Group / Clean Energy States Alliance · 1k+ views

Use of natural gas-fired generation differs in the United States by technology and region (US Energy Information Administration) · 2026/2027 Base Residual Auction Report (PJM)

An aeroderivative is a flight engine adapted to drive a generator: the gas generator core comes from an aircraft turbofan or turboshaft, and the fan is replaced by a free power turbine on its own shaft. Units run 20–120 MW, with GE's LM2500 near 30 MW, the LM6000 near 50 MW, and the intercooled LMS100 near 115 MW. Because the core is light and compact, the machine reaches full load in five to ten minutes, holds decent efficiency well down the load curve, and gets 38–44% in simple cycle, several points better than a heavy frame. It is also small enough to travel: a TM2500 is a road-transportable 34 MW power plant that can be generating within weeks of arriving on site. The trade is cost and maintenance, because you pay more per kW and more per operating hour than for a frame machine, and the engine is maintained the way an aircraft engine is — removed and exchanged at a depot rather than worked on in place.

Speed is the product: five-to-ten-minute starts, high ramp rates, many starts a day without the maintenance penalty a frame machine takes, and part-load efficiency good enough that the unit is still useful at half load. The small package means it can be trucked, barged, or airlifted, sited on a compact pad, and installed in months rather than years, which is why aeroderivatives dominate temporary and emergency power. The weaknesses follow from it being an aircraft engine. Hot-section intervals are consumed by starts as well as by hours, and refurbishment means exchanging the whole gas generator, so maintenance is a scheduled capital event rather than a running expense and a spare-core pool is a real working-capital item. Efficiency is far below a combined cycle, so running one for baseload burns money. And single-unit output tops out near 120 MW, so anything larger becomes a multi-unit plant with the complexity that implies.

Pick an aeroderivative when the duty cycle is defined by starts and speed rather than by running hours: fast-start peaking, non-spinning reserve, grid support in a weak network, or firming a large renewable build. If you need the same energy for many hours a day, a combined cycle is about 20 efficiency points better and cheaper per kW. If you need capacity and can live with a 10–30 minute start, a frame peaker is cheaper per kW and cheaper to maintain, and above roughly 150 MW it is the obvious choice. Below about 20 MW, reciprocating engines beat it on efficiency, part-load behavior, and cost. The one place it currently has no competition is bridge power: a data center with a signed lease and a 2030 interconnection date can put mobile aeroderivatives behind the meter in months, which is a schedule no other thermal plant can meet.

Unit output 20–120 MW, with the LM2500 near 30 MW, the LM6000 near 50 MW, and the LMS100 near 115 MW · simple-cycle efficiency 38–44%, against 35–40% for a heavy frame · full load in 5–10 minutes · the TM2500 a road-transportable 34 MW package rated up to about 39% efficiency · installed cost in the $1,500–3,000/kW range, above a frame peaker's $1,150–1,450 · capacity factor typically under 15%, and often under 5% on backup and bridge duty.

An aeroderivative is a synchronous machine accredited near nameplate, and its five-to-ten-minute start makes it one of the few generators that can genuinely supply non-spinning reserve and replace a tripped unit inside a contingency window. On weak or islanded networks it does more: it can run as a synchronous condenser or in grid-forming mode to hold frequency and voltage for a large inverter fleet, which is why island systems and remote mining grids buy them. Its most consequential property in 2026 is that it can be sited behind the meter and never touch the queue at all — a data center running mobile turbines islanded from the grid skips a five-year interconnection wait, and that arbitrage, not the thermodynamics, is what is driving current orders. In that configuration the binding constraint is the air permit, not the wires.

GE Vernova's LM2500, LM6000, and LMS100, plus the mobile TM2500, now offered in a water-free dry-low-emissions version at 34 MW; Siemens Energy's SGT-A65, derived from the Rolls-Royce Industrial Trent; Pratt & Whitney's FT8 SwiftPac; ProEnergy's repurposed CF6-80C2 aviation cores packaged at roughly 50 MW a unit for data-center bridge power; TM2500 fleets deployed as emergency power in Ukraine, Puerto Rico, and Argentina; and the mobile-turbine plants installed at US AI data-center sites in 2025–26, which have also produced the first serious air-permit fights over behind-the-meter generation.

An aeroderivative costs more per kW than a frame machine and considerably more per operating hour, and the market accepts that because it is buying availability and schedule rather than cheap megawatt-hours. Maintenance is the distinguishing line item: the gas generator is exchanged on an hours-and-starts schedule like an aircraft engine, so owners buy long-term service agreements and lease spare cores, and the depot network is part of the product rather than an afterthought. The 2024–26 demand shock came from the same turbine shortage that hit combined cycles — with frame slots gone to 2029–30, buyers who need power now pay premiums for aeroderivatives, refurbished cores, and rental fleets, and specialists like ProEnergy have built a business converting retired aviation cores into 50 MW power packages. That premium is cyclical. If OEM capacity catches up or interconnection speeds up, the bridge-power market shrinks quickly, so a business built on it should assume the window closes rather than widens.

11 minLM6000 FAMILIARIZATION Episode 01 - IntroductionPROENERGY · 10k+ views

2 minLM6000 Aeroderivative Gas TurbineTechnical Training Professionals · 10k+ views

2 minLM2500 Combustion Turbine for Combined Cycle and Simple Cycle TrainingTechnical Training Professionals · 50k+ views

Catalog of CHP Technologies, Section 3: Technology Characterization – Combustion Turbines (US Environmental Protection Agency) · Gas Turbines in Simple Cycle & Combined Cycle Applications (NETL Gas Turbine Handbook)

A gas turbine burning hydrogen, either blended into methane or on its own. Combustion is mostly a solved problem: modern F- and H-class machines are now sold with combustors rated for 30–50% hydrogen by volume, and GE Vernova's LM6000VELOX aeroderivative package is rated for 100% using a reconfigured single-annular combustor with water injection. What makes hydrogen awkward in a turbine is the flame. It burns roughly seven times faster than methane, which pushes the flame upstream into the premixing hardware (flashback) in exactly the dry-low-NOx combustors machines use to meet their emissions limits, and it burns hotter, which makes more thermal NOx. Hydrogen also carries about a third of methane's energy per unit volume, so fuel skids, valves, and piping grow roughly threefold for the same MW. That volumetric penalty is why a 30% hydrogen blend by volume displaces only about 11% of the fuel energy, and therefore cuts CO2 by about that much.

The plant is an ordinary gas turbine, so you get a synchronous machine, fast starts, a supply chain that already builds it, and the option to keep running on methane. The failure mode is fuel, not hardware. A kilogram of hydrogen holds 33.3 kWh on a lower-heating-value basis, so a 60%-efficient combined cycle gets about 20 kWh of electricity per kilogram; at $4–6/kg delivered that is $200–300/MWh in fuel before any capital. Pure hydrogen also cannot ride the existing gas network, which typically caps blends at 5–20% by volume, so a hydrogen plant needs dedicated production, a dedicated pipe, and bulk storage. Salt caverns are the only cheap bulk storage, which turns fuel into a geography problem. The most instructive data point is the showcase that vanished: South Australia canceled the 200 MW Whyalla hydrogen plant in 2025 and disbanded its hydrogen office when the money went to rescuing the local steelworks.

Buy hydrogen capability rather than hydrogen. If you are ordering a turbine that will run for thirty years in a market with a plausible carbon price, spec a combustor rated for 30–50% hydrogen: the adder over a standard machine is small and it preserves the option. Actually running on hydrogen needs three things at once, and if any is missing the project does not work: cheap delivered hydrogen, bulk storage (a cavern, not tube trailers), and a market that pays for a few hundred hours a year of firm zero-carbon output. That describes a seasonal-firming asset at a 5–15% capacity factor, not a replacement for a combined cycle. If the job is bulk low-carbon energy from gas, gas with carbon capture is cheaper because it keeps the cheap fuel. If the job is firm capacity on a date, gas, geothermal, and nuclear all beat it on delivered cost per firm kW.

Blends of 30–50% hydrogen by volume on most modern F- and H-class combustors, 100% on GE Vernova's LM6000VELOX · about a third of methane's energy per unit volume, so a 30% blend by volume cuts CO2 by only about 11% · hydrogen's flame speed roughly 7× methane's, which drives flashback and thermal NOx · 33.3 kWh/kg LHV, so roughly 20 kWh of electricity per kg at 60% combined-cycle efficiency · $4–6/kg delivered means $200–300/MWh of fuel alone · existing gas networks cap blends at 5–20% by volume.

It is the same synchronous machine as any gas turbine, so it supplies inertia, fault current, and voltage support and gets close to full capacity credit, but only as far as the fuel behind it stretches. That makes storage rather than the turbine the thing that sets firm capacity: an accredited 200 MW hydrogen plant needs enough cavern or tank inventory to cover the longest lull the system plans for. Everything upstream has to be built alongside the plant (electrolysis, a dedicated hydrogen pipe, and storage) because pure hydrogen cannot be delivered through the methane network. Interconnection is usually the easy part, since these units are retrofits or new machines at existing gas sites with a connection already in place.

Intermountain Power Project "IPP Renewed" at Delta, Utah: two Mitsubishi M501JAC turbines, 840 MW combined cycle, in commercial operation from 2025 on a 30% hydrogen blend by volume with a stated path to 100% by 2045, fed from salt caverns at the adjacent ACES Delta project. Mitsubishi Power's 30% co-firing demonstration on a large frame machine at Takasago, and the 50% blend it and Georgia Power reached at Plant McDonough-Atkinson. GE Vernova's LM6000VELOX, ordered by ATCO for the 200 MW Whyalla plant in South Australia before the state canceled it. Germany's planned "H2-ready" gas plant auctions, which pay for turbines specified to convert later.

The turbine is barely a premium. Hydrogen-capable combustors add a modest amount to a machine you were buying anyway, which is why almost every new order in 2025–26 is specified hydrogen-ready. All the real cost is in the fuel chain, and at $4–6/kg the fuel alone is $200–300/MWh, several times a combined cycle's all-in cost, before the electrolyzers, storage, and pipe. So the technology sells as an option rather than as an operating mode: buyers pay a little now to avoid stranding a thirty-year asset and defer the decision on whether to ever burn hydrogen. Treat delivered hydrogen cost as an input here; the production side is on the hydrogen-and-efuels sheet. If that cost falls far enough, the first real duty for these plants is the last few hundred hours a year that storage and renewables cannot cover, and the revenue for that comes from a capacity market or scarcity pricing, not from selling energy.

3 minCan gas turbines run on hydrogen fuel? | GE VernovaGE Vernova · 10k+ views

4 minWorld's First 30% Hydrogen Co-firing Demonstration Using State-of-the-Art Large Frame Gas TurbineMitsubishi Power · 1k+ views

Hydrogen Gas Turbines: ETN Global Report, October 2024 (ETN Global) · A Literature Review of Hydrogen and Natural Gas Turbines: Current State of the Art with Regard to Performance and NOx Control (NETL)

Two different ways to make a gas plant's CO2 capturable. Post-combustion capture bolts an amine absorber onto a combined cycle's exhaust: the solvent picks CO2 out of a dilute flue gas (about 4% CO2 for a gas turbine, against 12–14% for coal), then steam bled from the bottoming cycle regenerates the solvent and releases the CO2 for compression. That gets 90–95% of the CO2 at an energy penalty of roughly 8–12% of plant output, most of it the regeneration steam. Oxy-fuel supercritical CO2, the Allam cycle, takes the other route: burn the gas in nearly pure oxygen so the combustion products are only CO2 and water, run the CO2 itself as the working fluid, condense out the water, and take the surplus CO2 off at pipeline pressure with no stack at all. The price there is an air separation unit, which eats a similar share of the output, plus a turbine nobody has yet built at commercial scale. Either way, the plant is only half the project: the CO2 needs a pipeline and a permitted place to go.

This is the only route that keeps a dispatchable gas plant and its cheap fuel while removing most of the CO2, and post-combustion capture is a retrofit, so it can reuse a site, a turbine, and an existing grid connection. The weaknesses compound. Capture costs 8–12% of output plus a large capital adder, so the plant makes less power and costs much more per kW. The capture island has its own minimum stable load and its own trip modes, so availability falls and the unit ramps worse than the combined cycle it started as. The usual failure is not the chemistry. It is a finished plant with nowhere to put its CO2 because the storage permit or the pipeline slipped, or a capture unit that gets bypassed whenever the power price beats the value of the credit.

Only consider gas with capture when you already have a credible CO2 disposal path. In practice that means the US Gulf Coast, Alberta, or the North Sea, where a storage site, a Class VI-equivalent permit, and a pipeline are realistic on the project's schedule; anywhere else the disposal path is the project and the power plant is the easy part. Then check the revenue, because nothing here clears without a 45Q-style credit, a carbon price, or an enhanced-oil-recovery buyer, and the answer moves more with that number than with any engineering choice. Pick post-combustion if you are retrofitting or want technology risk near zero; pick oxy-fuel only if pipeline-pressure CO2 with no stack is worth first-of-a-kind risk to you. If what you actually need is clean firm power on a data-center timeline rather than a decarbonized gas fleet, geothermal and nuclear are the alternatives large offtakers compare against, and both avoid CO2 logistics entirely.

90–95% CO2 capture on post-combustion amine · 8–12% of plant output lost to the energy penalty, mostly solvent-regeneration steam · about 4% CO2 in gas-turbine exhaust against 12–14% for coal, which is why capture per tonne costs more on gas · roughly 0.34 t CO2/MWh from a modern combined cycle, so 90% capture is worth about $26/MWh at the US 45Q rate of $85 a tonne · NET Power's Project Permian about 80 MW for $475–575M, roughly $6,000–7,000/kW · commercial operation targeted for early 2029 after a 2026 investment decision.

The generator is a synchronous gas turbine, so inertia, fault current, and capacity credit look much like a plain combined cycle's. The capture island is what changes the plant's behavior: solvent regeneration draws steam from the bottoming cycle, the absorber has a minimum stable load of its own, and the extra equipment adds forced-outage modes, so the unit ramps more slowly and is available fewer hours than the same turbine without capture. What has to be built alongside it is a CO2 pipeline and a permitted injection site, which is usually a longer and less certain path than the power plant, and one the grid operator has no visibility into.

NET Power's Project Permian in West Texas, redesigned around Siemens A35 turbines with Entropy's post-combustion capture, about 80 MW in phase one at an estimated $475–575M, with a final investment decision expected in the second half of 2026 and operation targeted for early 2029. Equinor and partners' Northern Lights in Norway as the storage half of the equation. Entropy's capture units on gas plants in Alberta. On the solid-fuel side, Boundary Dam 3 in Saskatchewan is still the only large operating coal capture unit anywhere, and it is covered under pulverized coal.

Capture roughly doubles a gas plant's capital cost per kW while cutting its net output, so the levelized cost lands well above $200/MWh at any realistic capacity factor, and the plant needs to run a lot of hours to spread that capital. NET Power's numbers are the cleanest public data point: roughly $6,000–7,000/kW for a first phase, against $2,000–2,800/kW for the combined cycles the market is currently buying. The revenue side is policy. A modern combined cycle emits about 0.34 tonnes of CO2 per MWh, so capturing 90% is worth roughly $26/MWh at the US 45Q rate of $85 a tonne, which is real money and still not enough to cover the capital adder on its own. Anyone investing here is betting that a large buyer will pay a premium for firm gas power with the CO2 accounted for, on a schedule nuclear cannot meet and geothermal can meet only in certain geologies.

2 minCan solvent-based carbon capture work? CO2 reduction explained | GE VernovaGE Vernova · 1k+ views

2 minNET Power, LLC Allam Cycle Overviewgwilliambrownjr · 10k+ views

Global Status of CCS 2025: Executive Summary (Global CCS Institute) · Class VI Wells Used for Geologic Sequestration of Carbon Dioxide (US Environmental Protection Agency)

Coal is ground to a talc-fine powder, blown into a furnace with air, and burned to raise steam for a Rankine-cycle turbine. Efficiency is set almost entirely by steam conditions: subcritical plants run 33–37%, supercritical plants (above water's critical point, around 221 bar) reach 40–43%, and ultra-supercritical plants at about 600 °C and 300 bar reach 44–47%. Everything downstream of the boiler is emissions control, which on a modern plant is a large share of the capital: selective catalytic reduction for NOx, a wet or dry scrubber for SO2, an electrostatic precipitator or baghouse for particulates, and activated carbon injection for mercury. Coal was still the largest single source of world electricity in 2025 at about 33%, though generation fell for the first time since 2020. The US fleet ran at a 48.7% capacity factor that year, which is what happens to plants built for baseload when cheaper gas and renewables dispatch ahead of them.

A coal unit is a large synchronous machine with weeks of fuel sitting in a pile on site, so it supplies inertia, fault current, and energy security without a pipeline, and most of the existing fleet is paid for. Against that, a new build costs $3,405–7,210/kW and $71–173/MWh, which loses to gas and renewables nearly everywhere, and heat rates of 8,750–12,000 Btu/kWh put emissions near 0.9 tonnes of CO2 per MWh. The plants are also bad at the job grids now need, with minimum loads around 40–50% of rating, ramp rates of a few percent of rating per minute, and cold starts measured in hours. The failure mode for an existing unit is cycling. A boiler designed to run flat and then cycled daily to make room for solar accumulates thermal fatigue in thick-wall headers and drums, and forced-outage rates and maintenance costs climb until the unit is unreliable in exactly the weeks it is needed.

For a new plant in a market with any alternative, don't. Gas beats it on capex, build time, and emissions; solar and wind beat it on energy cost; and the places where new pulverized coal still pencils out are where domestic coal is cheap, gas is imported at LNG-linked prices, and capital is state-directed, which in practice means parts of Asia. The live decision for most utilities is retire versus extend on units already built. Extend when the marginal cost of $31–114/MWh sits below the local power price often enough, the next environmental retrofit is affordable, and the grid still needs the unit for local voltage support. Retire when a large retrofit bill arrives, because spending tens of millions on a forty-five-year-old asset to keep making the most expensive energy on the system rarely returns it. If the site has a good interconnection and a rail line, converting the boiler to biomass or replacing it with gas keeps the most valuable thing on the property, which is the grid connection.

Subcritical 33–37% efficiency, supercritical 40–43%, ultra-supercritical 44–47% at about 600 °C and 300 bar · heat rate 8,750–12,000 Btu/kWh, roughly 0.9 t CO2/MWh · new build $3,405–7,210/kW at $71–173/MWh · marginal cost of an existing plant $31–114/MWh · about 33% of world electricity in 2025 at a 48.7% US capacity factor · minimum load around 40–50% of rating, with cold starts measured in hours.

A coal unit is one of the largest synchronous machines on most systems, so it carries a lot of inertia and fault current and gets close to full capacity credit. It is a poor partner for variable renewables: a 40–50% minimum load, ramp rates of a few percent of rating per minute, and multi-hour cold starts mean it cannot get out of the way at midday or come back for the evening peak. Retirements often turn out to be transmission problems rather than energy problems, because the unit was also holding up local voltage, which is why operators keep issuing reliability-must-run contracts to plants that are otherwise uneconomic. Nothing much has to be built alongside a coal plant on the grid side, but the fuel needs rail or barge access and a stockyard, and that logistics chain is most of why coal plants sit where they do.

China connected new ultra-supercritical units through 2025 even as its coal share of generation fell; NTPC Khargone in India runs 2×660 MW ultra-supercritical; Germany's Datteln 4 (1,100 MW hard coal) came online in 2020 into a market already retiring coal. The US fleet has fallen by more than half from its 2011 peak, and the Intermountain Power Project in Utah replaced its two coal units with a hydrogen-capable combined cycle on the same site and connection. Some 2025 US retirements were held open past their dates by federal emergency orders, including J. H. Campbell in Michigan.

The fleet runs on sunk capital. A paid-off unit's marginal cost is $31–114/MWh depending on coal price and heat rate, and that number rather than LCOE decides whether it runs on a given day. New capacity is a separate question with an almost uniform answer: $3,405–7,210/kW at $71–173/MWh, against $2,000–2,800/kW for a combined cycle, with a forty-year asset life staked on the politics of carbon. Where coal still gets built the driver is usually a domestic fuel resource and directed capital rather than a cost comparison. For anyone building a business around the fleet, the money is in what comes next: emissions retrofits, flexibilization, ash-pond remediation, decommissioning, and repowering sites for gas, batteries, or data centers, because the grid connection and transmission rights outlive the boiler.

7 minHow Coal Fired Thermal Power Stations WorksaVRee · 100k+ views

5 minHow a Coal Power Plant Works | Step-by-Step ExplanationWebMaxi ENG · 10k+ views

23 minThe Epochal Ultra-Supercritical Steam TurbineAsianometry · 100k+ views

U.S. coal-fired generating capacity retired in 2025 was the least in 15 years (US Energy Information Administration) · Cost and Performance Baseline for Fossil Energy Plants, Volume 1: Bituminous Coal and Natural Gas to Electricity (NETL, 2022)

A conventional steam plant burning wood pellets, agricultural residue, or bagasse instead of coal, either purpose-built or converted from an existing coal unit. Drax in North Yorkshire is the canonical conversion: four units of roughly 645 MW each, about 2.6 GW in total, switched from coal to pellets by rebuilding the mills, adding enclosed conveyors and storage domes, and changing the burners. Electrical efficiency is 30–38%, below coal's 33–47%, because fuel moisture absorbs heat, the flame is cooler, and the ash chemistry (alkali metals and chlorine) forces conservative steam conditions to avoid slagging and superheater corrosion. Co-firing is the cheap version: blend up to roughly 10–20% biomass by heat into an existing pulverized coal boiler without major changes, and go higher only by rebuilding the fuel handling. The plant is dispatchable and synchronous, which is why grids value it more than its cost per MWh suggests.

You get a weather-independent, dispatchable, synchronous plant burning a fuel that most regimes count as zero-carbon at the stack, and a coal conversion reuses the boiler, turbine, and grid connection, which is worth a great deal when a new interconnection takes over five years. The problem is the fuel bill. Delivered pellets typically run several times the price of natural gas per unit of heat, and the efficiency gap widens that per MWh, so the plant only runs when a subsidy or a mandate covers the difference. The second exposure is accounting: whether burning a tree is carbon-neutral depends on harvest-and-regrowth timing and on the supply chain's own emissions, the question is genuinely contested, and regulators rather than engineers set the answer. The failure mode follows from both. A thirty-year asset whose fuel costs more than the power it makes fails the moment its support scheme changes, and Drax's UK support has already been renegotiated down and capped on running hours.

Convert an existing coal unit to biomass when the site has a grid connection you could not get today, a rail or port link that can move a million tonnes a year of pellets, and a support scheme with enough remaining term to pay back the conversion. Purpose-built biomass makes sense mainly at modest scale next to a fuel source that would otherwise be a disposal problem: a sawmill, a sugar mill burning its own bagasse, an agricultural residue nobody else wants, ideally with a heat customer next door. If you are paying market price for pellets shipped across an ocean, run the numbers against gas with carbon capture and against simply closing the plant, because biomass usually loses without a subsidy. If the fuel has a negative price rather than a positive one, you are looking at waste-to-energy instead.

30–38% electrical efficiency, against 33–47% for coal · Drax's four converted units total about 2.6 GW · co-firing up to roughly 10–20% biomass by heat in an otherwise unmodified pulverized coal boiler · delivered pellets several times the price of natural gas per unit of heat · roughly $3,000–5,000/kW for a new purpose-built plant, and a fraction of that for a coal conversion · $100–200/MWh unsubsidized in North America and Europe.

Biomass is a synchronous steam plant, so it supplies inertia, fault current, and voltage support and gets close to full capacity credit as long as the fuel keeps arriving. A converted coal unit keeps the original grid connection and transmission rights, which in a market with a five-year interconnection queue is often the single largest reason the conversion happens at all. It ramps like the coal plant it used to be, better than nuclear and worse than gas, so it works as mid-merit dispatchable capacity rather than as fast reserve. What has to be built alongside it is fuel logistics: port or rail capacity, enclosed conveyors, and covered storage sized for weeks of supply, because pellets degrade and self-heat when they get wet.

Drax in North Yorkshire, four units converted from coal to about 2.6 GW of pellet capacity, now on a reduced UK support deal that also limits how much it runs. Ørsted's Avedøre and Studstrup conversions in Denmark, which sell heat into district networks alongside power. MGT Teesside at 299 MW, purpose-built on pellets. Brazilian and Indian sugar mills cogenerating from bagasse through the crushing season. Enviva as the pellet supply side, and its 2024 Chapter 11 filing as a reminder that the fuel chain has economics of its own.

The cost structure is fuel. Capital is roughly $3,000–5,000/kW for a purpose-built plant and much less for a conversion, but at 30–38% efficiency the delivered pellet price sets the marginal cost, and it sits well above gas nearly everywhere, which puts the unsubsidized levelized cost at $100–200/MWh in North America and Europe. That makes this a subsidy business: renewable obligations and contracts for difference in the UK, feed-in premiums in Japan and Korea, renewable heat schemes across Europe. Those terms get renegotiated, and Drax's post-2027 arrangement cut its support and capped its running hours, turning a baseload asset into a dispatchable one. If you are financing biomass, the two questions that matter are how much term is left on the support scheme and whether the fuel supply chain survives without this plant, because a pellet mill built for one offtaker has no other customer.

3 minHow Does a Biomass/Wood-fired Boiler Plant Work? | HoSt Bioenergy SystemsHoSt Group | Bioenergy Systems · 50k+ views

6 minDrax's transition from coal to biomass 🌿⚡️Drax · 5k+ views

3 minShipping the Wood Pellets to Fuel the UK's Largest Power PlantScience Channel · 50k+ views

Task 32: Biomass Combustion, publications and co-firing database (IEA Bioenergy) · The government's support for biomass (UK National Audit Office, 2024)

A mass-burn waste-to-energy plant takes unsorted municipal solid waste, drops it onto a moving grate, and burns it above 850 °C to raise steam. A typical unit makes 20–60 MW of electricity at only 20–25% efficiency, which would be indefensible if electricity were the product. It isn't. The plant is paid a gate fee for every tonne that arrives and sells the ferrous and non-ferrous metal it recovers from the ash, and power is a by-product. The low efficiency is a materials constraint rather than a design choice: chlorine in the waste corrodes superheater tubes at high metal temperatures, so steam is held near 400 °C and 40 bar instead of the 600 °C a coal plant uses. Flue-gas treatment for dioxins, acid gases, NOx, and mercury is a large share of the plant, and the bottom and fly ash are regulated waste streams in their own right.

The fuel arrives on a schedule, pays to be there, and does not care about weather, so the plant runs at an 85–90% capacity factor with no fuel-price exposure at all. Diverting waste from landfill also avoids landfill methane, which is the strongest part of the environmental case. Against that, capex per kW of electrical capacity is the highest of any thermal plant, because you are buying a waste-handling and emissions-control facility that happens to make power. Public opposition is intense and siting takes years. And the plant's economics depend on two things it does not control: the local landfill price and recycling policy. The characteristic failure is a long-term put-or-pay waste supply contract signed by a municipality that later diverts tonnage to recycling, leaving the plant short of fuel and the city paying for waste it never delivers.

Build waste-to-energy where landfill is expensive or illegal and the waste has to go somewhere anyway: dense European and East Asian cities, islands, and jurisdictions with landfill taxes or bans. Then check whether a heat customer exists, because a plant selling steam or hot water into a district network recovers roughly three times as much useful energy as one making only electricity, and in most European projects that is what decides whether it works. Do not build it as a generator. Judged on $/MWh it loses to almost everything else on this sheet, and the comparison that matters is against the landfill gate fee it displaces, not against a gas plant. In the US, where landfill is cheap and plentiful, essentially nothing new has been built since the 1990s, and for those conditions that is the correct answer.

20–60 MW per unit at 20–25% electrical efficiency · furnace above 850 °C but steam held near 400 °C and 40 bar, because chlorine corrodes superheater tubes above that · 85–90% capacity factor, since the waste arrives whatever the weather · revenue is a gate fee per tonne plus recovered metals, with power as a by-product · selling heat into a district network roughly triples useful energy recovery · capex above $7,000/kW of electrical capacity, and often well above.

The plant runs flat because waste keeps arriving and the boiler does not like cycling, so it behaves as baseload with a high capacity credit and no useful flexibility to sell. It is a synchronous steam turbine, so it does contribute inertia and fault current, and because it sits near the city that produces the waste it usually connects at distribution or sub-transmission voltage and needs no new transmission. In Europe the thing built alongside it is a district heat network, which is a larger capital commitment than the grid connection and locks the plant into heat-led dispatch, with electricity following whatever the heat load leaves over.

Amager Bakke (Copenhill) in Copenhagen, roughly 63 MW of electricity plus around 250 MW of district heat, under a public ski slope. Singapore's Tuas complex, which handles most of the island's waste because landfill there is not an option. Reworld, formerly Covanta, operating most of the US fleet, which has added essentially no greenfield capacity since the Palm Beach Renewable Energy Facility No. 2 opened in 2015. Shenzhen East in China, one of the largest plants in the world at around 5,000 tonnes a day. Hitachi Zosen Inova, Babcock & Wilcox Vølund, and Martin as the grate and boiler suppliers.

The revenue stack inverts the usual generator model. A gate fee per tonne is the largest line, recovered metals are second, heat is often third in Europe, and electricity is frequently the smallest. That makes the plant a waste-infrastructure asset priced off the local landfill alternative, so it is profitable where landfill taxes are high (much of northern Europe) and marginal where they are not (most of the US). Capital cost per kW of electrical capacity is the highest of any thermal plant, above $7,000/kW and often well above, and what makes it financeable is a twenty-year municipal waste contract rather than the power market. The strategic risk is recycling and waste-reduction policy: EU circular-economy rules and plastic reduction targets shrink both the tonnage and its calorific value, and a plant sized for 2005 waste streams runs into both at once.

2 minHow it works - Waste-to-EnergyWaste-to-Energy Facts · 100k+ views

3 minWaste-to-Energy Virtual TourReworld™ · 50k+ views

11 minVirtual plant tour of energy-from-waste plant, Meath County IrelandBabcock and Wilcox Vølund A/S · 5k+ views

Energy Recovery from the Combustion of Municipal Solid Waste (US Environmental Protection Agency) · What a Waste 3.0: Global Snapshot of Solid Waste Management toward Circularity until 2050 (World Bank)

Nuclear fission supplies about 9% of world electricity and is the only source on this sheet that pairs near-zero carbon intensity with a capacity factor above 90%. This entry is deliberately one row, and it covers only what a fission plant is on a grid and what it costs to build. Everything about how the reactors actually work (light water, small modular, high-temperature gas, molten salt, fast spectrum, and fusion) lives on the nuclear-reactors sheet, which spends twenty-five entries on the design space. The economics here are easy to state and hard to change: a US new build runs $9,020–14,820/kW and $141–220/MWh unsubsidized, almost all of it capital, while an existing plant's marginal cost is $30–38/MWh. Lazard models an 84-month construction period; recent Western projects have taken 10–15 years from decision to power. The US fleet ran at 91.0% in 2025, which nothing else on this sheet approaches.

Output is firm, carbon-free, and nearly insensitive to fuel price, the plants run 60–80 years with license renewals, and a 91% capacity factor means one kilowatt of nuclear produces about as much energy in a year as nearly four kilowatts of utility solar. The weakness is entirely in the capital: $9,020–14,820/kW spent over a decade before a single megawatt-hour is sold, with interest during construction adding roughly 30–50% on top of the overnight cost. Safety is not the failure mode; schedule is. A slip compounds through financing cost, and Vogtle 3 and 4 came in around $35B for 2.2 GW, roughly double the original estimate, and Virgil C. Summer was abandoned after about $9B with nothing built. Serial builders in Korea and China deliver the same job for a fraction of the cost, which is evidence that the problem is execution rather than physics.

Pick nuclear when a buyer needs firm carbon-free power for decades and can carry a decade of construction risk, which in practice means a state program, a regulated utility with cost recovery, or a hyperscaler willing to sign a twenty-year power purchase agreement above market. If your grid is small, if your balance sheet cannot absorb a single asset above $10B, or if you need the power before 2035, it is the wrong answer, and gas, geothermal, or solar plus storage will beat it on delivered cost and date. Restarting or uprating an existing plant is a much better deal than building a new one whenever it is available, because you buy a 91% capacity factor at a fraction of new-build cost. For which reactor design to pick, go to the nuclear-reactors sheet: the choice between light-water, small modular, and advanced designs moves capex and schedule far more than anything on this page.

About 9% of world electricity and a 91.0% US capacity factor in 2025 · US new build $9,020–14,820/kW at $141–220/MWh unsubsidized · marginal cost of an existing plant $30–38/MWh · 84-month construction in Lazard's model, 10–15 years in recent Western practice · interest during construction adds roughly 30–50% to the overnight cost on a ten-year build · Vogtle 3 and 4 about $35B for 2.2 GW · plant life 60–80 years with license renewals.

Nuclear gets the highest capacity credit on this sheet, close to its full nameplate, and a large synchronous turbine-generator supplies substantial inertia, fault current, and voltage support. The awkward part is unit size: a single 1,100 MW machine sets the largest-contingency reserve the whole system has to carry, so small grids either cannot take one or have to pay for that reserve, which is a real argument for smaller units. Nuclear can technically follow load, and French units do it routinely, but since almost all of its cost is capital and almost none is fuel, turning it down saves nothing and the plant is economically must-run. A new plant also needs firm high-voltage transmission out of a site usually chosen for cooling water rather than grid access, and that line is frequently on the critical path.

Vogtle 3 and 4 in Georgia, the only US new build this century, at roughly $35B for 2.2 GW. The Crane Clean Energy Center restart at Three Mile Island under a Microsoft power purchase agreement, and Palisades in Michigan, the first US restart of a plant that had already shut down. Barakah in the UAE, four units delivered close to schedule by a Korean consortium. Hinkley Point C in the UK, years late and far over budget. China connecting several new units a year on a serial program.

Nearly all of nuclear's cost is capital spent before any revenue arrives, so the cost of capital and the schedule matter more than anything about the plant itself. On a ten-year build, interest during construction adds 30–50% to the overnight cost, which is much of why the same design costs $2.5–4k/kW in Korea and China and three to five times that in recent Western first-of-a-kind projects. What changed between 2023 and 2026 is the buyer rather than the cost. Hyperscalers signing twenty-year agreements above market, plus restart deals for shut plants, moved the question from "can nuclear beat gas on price" to "who will pay a premium for firm clean power on a fixed date," and that is a question nuclear can sometimes answer. For an investor the leverage sits in schedule and financing structure, not in reactor design, which again is the nuclear-reactors sheet's subject.

16 minThe Economics of Nuclear EnergyReal Engineering · 1m+ views

18 minIs Nuclear Power “Too Expensive”?Kyle Hill · 500k+ views

Nuclear explained: U.S. nuclear industry (US Energy Information Administration) · Capital Cost and Performance Characteristics for Utility-Scale Electric Power Generating Technologies, AEO2025 (US Energy Information Administration)

The same cells and modules as a utility solar farm, installed on a roof or a small parcel behind a customer's meter, at two to three times the cost per watt. The hardware is not what makes it expensive. Modules sell for $0.08–0.12/W, so on a US residential system at $3–4/W they are under 5% of the price. The rest is soft cost: customer acquisition and sales commission, permitting and inspection, small crews working a different roof every day, inverters bought in ones rather than by the container, and overhead spread across a 7 kW project instead of a 200 MW one. Lazard's community and C&I band is $1,600–3,300/kW at $81–217/MWh, and US residential sits above the top of it. Capacity factors run 15–20% against utility solar's 24%, because roofs face where they face, rarely track, and get shaded.

Distributed solar competes against the retail tariff rather than the wholesale price, and US retail rates run roughly 12–35 c/kWh against a wholesale price a fifth of that, so a system with a poor LCOE by generator standards can still be a good purchase for the customer. It also needs no transmission and no bulk interconnection queue, which is why a commercial site can have it running in months while a utility project waits five years. The weaknesses follow from the same fact: the economics are set by a tariff rule, and tariff rules change. California's NEM 3.0 cut export compensation sharply in 2023 and residential sales fell hard afterward. The failure mode is a business built on one compensation regime. A residential installer carrying high customer-acquisition cost and thin margins does not survive its state moving from net metering to net billing, which is roughly the story of the 2023–25 wave of US installer bankruptcies.

Put solar on a roof when the customer pays a high retail rate, has a large daytime load, and owns the building. C&I is the strongest case in that list: a warehouse, cold store, or supermarket with a big flat roof, a load curve that matches generation, and a demand charge to shave, which works even without net metering because most of the output is consumed on site. Residential works where the retail rate is high and export compensation is still reasonable, and increasingly needs a battery to move the export into the evening, which changes the numbers. If the goal is decarbonizing a grid at least cost, buy utility-scale solar instead; the two-to-three-times gap in cost per watt is structural, not a learning-curve problem. And if the customer's real goal is riding through outages, they are buying a battery or a genset, because a plain grid-tied array shuts down when the grid does.

$1,600–3,300/kW at $81–217/MWh for Lazard's community and C&I band, with US residential above it · modules $0.08–0.12/W, under 5% of a $3–4/W residential system · 15–20% capacity factor against 24% for utility solar · competes against a US retail tariff of roughly 12–35 c/kWh rather than a wholesale price · installation measured in days for residential and weeks to months for C&I.

Distributed solar sits behind the meter, so it needs no transmission and skips the bulk interconnection queue, which is most of its speed advantage. The constraint moves to the distribution feeder: once generation approaches the feeder's minimum daytime load, reverse power flow raises voltage at the end of the line, and the utility either requires inverter volt-var and volt-watt control (IEEE 1547-2018 in the US) or caps hosting capacity outright. Its capacity credit falls as penetration rises for the same reason utility solar's does, because it pushes the system's net peak later into the evening. A plain grid-tied system also shuts down in an outage for line-worker safety, so what increasingly gets built alongside it is a battery and an islanding-capable inverter.

Sunrun and Freedom Forever in US residential; Enphase and SolarEdge microinverters and power optimizers, which exist because roofs shade unevenly; Tesla Powerwall and Enphase batteries as the standard NEM 3.0 response. On the C&I side, rooftop programs at IKEA, Walmart, Amazon, and Prologis, with developers like Standard Solar and Safari Energy financing them. Australia leads the world on penetration with more than 4 million rooftop systems, installing the same hardware for a fraction of the US cost per watt.

Almost all of the cost is labor, sales, and paperwork, so it does not fall when module prices fall. That is the structural difference from utility solar: modules dropped by an order of magnitude over fifteen years while US residential prices barely moved, because the dominant line items are local services priced in local wages. The countries that fixed the paperwork got the savings, and Australia's standardized approvals and accredited-installer scheme deliver systems at roughly a third of the US price using the same modules. The 2025–26 US policy changes matter more here than any technology development: the residential 25D credit ended on 31 December 2025, which pushes homeowners toward third-party-owned lease and power purchase structures that can still use the commercial credit, and consolidates the market toward installers with cheap capital. If you are building a distributed solar business, the durable advantages are a low customer-acquisition cost and a permitting process you have learned to run fast, not the equipment.

4 minHow Does Net Metering Work?SolarReviews · 10k+ views

11 minWhy SOLAR Panels Cost 3X MORE in USA vs Europe - The Hidden Truth About Solar Pricing 2025Solartime with Martyna · 50k+ views

Documenting 15 Years of Reductions in U.S. Solar Photovoltaic System Costs (National Renewable Energy Laboratory) · One Year In: Tracking the Impacts of NEM 3.0 on California's Residential Solar Market (Lawrence Berkeley National Laboratory)

Wind turbines from a few hundred watts to a few hundred kilowatts, rotors of roughly 1–30 m on towers of 10–40 m, usually sited at the point of use. The category stayed small for a physical reason rather than a commercial one. Wind power scales with the cube of wind speed and the square of rotor diameter, and wind speed rises with height, so a machine on a 15 m tower in turbulent, obstacle-cluttered near-ground air might see half the wind speed a 150 m utility turbine sees. Half the speed is an eighth of the power per square meter of rotor, on a rotor with roughly a thousandth of the swept area. The result is 15–25% capacity factors at $5,000–10,000/kW installed, against 30–55% at $1,900–2,300/kW for a utility machine. Turbulence shed by nearby buildings and trees makes the output worse and the fatigue loading harder at the same time, which is why building-mounted turbines almost never perform as advertised.

A small turbine generates at night and through winter, exactly when solar does not, so on off-grid sites at high latitude it can meaningfully shrink a battery bank and cut a generator's fuel burn. It is also close to the only renewable that works in polar winter. Against that, cost per kWh is several times rooftop solar at the same site, the machine has bearings, a gearbox, and a yaw system to maintain where a panel has none, and mast foundations and guy wires make siting harder than bolting modules to a roof. The failure mode is the resource assessment. Buyers site from a regional wind map instead of from measurements at hub height, and the cube law makes that expensive: a 30% overestimate of mean wind speed is a 120% overestimate of energy. Vertical-axis machines resurface every few years with the same claims and lose on the same physics, plus a peak power coefficient below the 0.45–0.50 a modern horizontal-axis rotor reaches against the Betz limit of 0.593, and blade loading that reverses twice per revolution.

Buy small wind only where the site is genuinely off-grid, the wind resource has been measured rather than assumed, and solar alone fails for part of the year. That set is real but narrow: telecom repeaters, remote monitoring and navigation aids, research and military stations at high latitude, and boats and buoys. On any site with a roof and a grid connection, buy solar; it is several times cheaper per kWh, has no moving parts, and permits more easily. If you need more than about 100 kW and own land with a real resource, put up a single utility-class machine on an 80 m or taller tower, which is a completely different economic proposition and is what "distributed wind" should mean. And if a vendor pitches a rooftop or ducted turbine in a built-up setting, ask for metered production data from an installed unit, because the near-ground resource almost never supports the brochure.

Rotors of 1–30 m on towers of 10–40 m, from a few hundred watts to a few hundred kilowatts · 15–25% capacity factor · $5,000–10,000/kW installed, against $1,900–2,300/kW for utility onshore wind · power scales with the cube of wind speed and the square of rotor diameter · a 30% error in assumed wind speed is a 120% error in energy · power coefficient near 0.45–0.50 for a modern horizontal-axis rotor against the Betz limit of 0.593, and lower for vertical-axis machines.

On a grid-connected site a small turbine is a distribution-level, inverter-coupled resource with essentially no capacity credit, because a few kilowatts of uncorrelated output is not something a planner can count on at peak. It needs the same interconnection agreement and inverter standards as rooftop solar (IEEE 1547 in the US) and no transmission at all. Off-grid, where most of these machines actually live, the thing built alongside the turbine is the rest of the system: a battery bank, a charge controller with a dump load for high winds, and usually a solar array and a small generator too, because no single off-grid source covers the whole year.

Bergey WindPower's Excel 10 and 15 for farms and off-grid homes; Primus Windpower's Air series on boats and remote telecom sites; Northern Power Systems' 100 kW machine on schools and farms; Ryse Energy and Eocycle in the off-grid and hybrid market. The Ross Island wind farm serving McMurdo and Scott Base in Antarctica is the polar niche done properly, with three 330 kW turbines cutting diesel deliveries. The US Department of Energy's Distributed Wind Market Report tracks a market of tens of megawatts a year, next to tens of gigawatts of utility wind. The urban and building-mounted end of the field has produced a long run of companies that raised money and shipped little, from Mariah Power's Windspire onward, and the recurring "wind tree" products are the same pattern.

Small wind never got the cost curve solar did. The market never grew large enough to support volume manufacturing, and a turbine is a mechanical assembly with a tower, foundation, bearings, and gearbox rather than a laminated sheet that comes off a line. Installed cost is $5,000–10,000/kW and has not moved much in fifteen years, while rooftop solar fell by a large multiple over the same period, which is the whole reason solar took the distributed market. The remaining suppliers are small and several have been through bankruptcy or acquisition. If you are building a business here, the defensible position is service and system integration for off-grid customers paying $0.30–1/kWh for diesel, not a better rotor. The value is in a hybrid system that keeps a remote site running through winter, and the turbine is one component of it.

12 minSmall Wind Turbine: Success or Failure? Do not make those MistakesWindSunWisdom · 100k+ views

11 minAre Vertical Axis Wind Turbines Better?Engineering with Rosie · 500k+ views

Small Wind Guidebook (US Department of Energy WINDExchange) · Distributed Wind Energy Technology Data Update, 2025 Edition (Pacific Northwest National Laboratory)

Diesel and gas piston engines driving generators, from about 100 kW up to 20 MW in a single block and ganged into plants above 200 MW. Their properties are better than their reputation. Large lean-burn gas engines reach 45–50% electrical efficiency, which beats a simple-cycle gas turbine's 35–44%; they synchronize in tens of seconds and reach full load in a couple of minutes; and because a plant is built from many identical units, you shed output by shutting engines off rather than by throttling one machine, so plant efficiency barely falls at part load. That last property is what makes a bank of engines the flexible-generation choice on grids with a lot of wind and solar. Capex runs $700–1,500/kW, which overlaps a frame peaker's band at the top and undercuts it at the bottom, and units ship with lead times measured in months while a gas turbine order waits five years.

Fast starts, flat part-load efficiency, low capex, modularity, and availability now are a strong combination for a grid that needs flexibility rather than energy. Engines are synchronous too, so they contribute inertia and fault current and can black-start a system. The weaknesses are emissions and fuel cost. NOx and particulates per MWh are far higher than a modern turbine's, so permits often cap run hours, and US emergency-standby engines are typically limited to about 100 hours a year of non-emergency operation. Diesel is expensive per MWh: at $3 a gallon and 42% efficiency the fuel alone is well over $150/MWh, which is why diesel units are backup and utility-scale engine plants burn gas. The characteristic failure is organizational rather than mechanical. A standby fleet that has never been load-tested properly does not start when the utility supply actually fails, and nobody finds out until that day.

Choose engines when you need capacity or flexibility below about 200 MW, when you need it soon, or when the plant will run at low and highly variable output. Below roughly 20 MW they beat aeroderivative turbines on efficiency and cost per kW. Above that, compare on expected starts and load profile rather than full-load efficiency, because part-load and start behavior is where engines win and where a turbine's heat-rate advantage disappears. For backup at a data center or hospital, diesel is still the default: the fuel stores on site indefinitely and the duty cycle is a few hours a year, so efficiency does not matter. For a utility-scale flexible plant use gas engines, and check the local NOx permit before anything else, because it sets the SCR scope, the run-hour cap, and sometimes whether the project is possible at all. If the site needs steady output rather than flexibility and cannot get an engine permit, a solid-oxide fuel cell is the usual alternative.

100 kW to 20 MW per unit, ganged into plants above 200 MW · 45–50% electrical efficiency on large gas engines, against 35–44% for a simple-cycle turbine · synchronized in tens of seconds and at full load in a couple of minutes · $700–1,500/kW installed, undercutting a frame peaker's $1,150–1,450/kW at the low end · US emergency-standby permits typically cap non-emergency running near 100 hours a year · diesel fuel alone above $150/MWh at $3 a gallon and 42% efficiency.

Engines are synchronous machines, so they supply real inertia, fault current, and voltage support, and a plant of them can black-start a system. Their value to an operator is start speed and turndown: a 200 MW plant built from twenty 10 MW engines can cover a wind forecast error in minutes and sit at a small fraction of rating without an efficiency penalty, which makes it useful as contingency reserve and renewables balancing rather than as an energy source. Standby engines behind a customer's meter are a different animal. The operator cannot see them, they carry no capacity credit, and their run-hour permits usually forbid the economic dispatch that would make them useful, unless they are enrolled in a demand-response program, which is what a growing number of data-center fleets now do. Interconnection is easy because the plants are small and sited at load, and what gets built alongside them is emissions control (SCR and an oxidation catalyst) plus fuel storage or a gas lateral.

Wärtsilä's 31 and 34 series flexible plants, including the Cooperative Energy installation at Benndale, Mississippi, and a large installed fleet across Latin America and the Caribbean where engines cover both baseload and peaking. INNIO Jenbacher and Caterpillar's MWM units in the 1–10 MW combined-heat-and-power market. Cummins, Kohler, and Caterpillar diesel sets as the standard data-center and hospital backup at 2–3 MW per unit with N+1 redundancy. Wärtsilä and Rolls-Royce mtu hydrogen-capable engine development as the low-carbon path, including Wärtsilä's 100% hydrogen engine power plant.

Engines are a commodity product from a handful of makers (Wärtsilä, INNIO Jenbacher, Caterpillar, Cummins, Rolls-Royce mtu), sold with a long-term service agreement that is a real part of the cost, because overhauls are scheduled by running hours and a plant that runs a lot pays a lot. Capex of $700–1,500/kW is low for a thermal plant, and for anything running more than a few hundred hours a year the fuel bill dominates everything else. The 2024–26 opportunity is availability: with gas turbine slots sold into 2030–31 and reservation fees in the tens of millions, an engine plant can be ordered, delivered, and commissioned inside two years, which has pulled a lot of data-center bridge power and utility peaking demand toward them. The risk to that business is permitting rather than technology, because the same NOx and particulate rules that keep engines cheap to buy make them expensive or impossible to run in non-attainment areas.

3 minHOW A DIESEL GENERATOR WORKS -ANIMATIONENGINEERING SPIRIT · 1m+ views

7 minFlexible Gas Power Plant Wärtsilä 31SG | Cooperative Energy, BenndaleWärtsilä Corporation · 10k+ views

Catalog of CHP Technologies, Section 2: Technology Characterization – Reciprocating Internal Combustion Engines (US Environmental Protection Agency) · Controlling Air Pollution from Stationary Engines (US Environmental Protection Agency)

A stationary fuel cell oxidizes natural gas or hydrogen electrochemically instead of burning it, so the products are DC electricity, water, and CO2, with essentially no NOx, SOx, or particulates. Solid oxide is effectively the whole market, and Bloom Energy is effectively the whole solid-oxide market. A Bloom stack runs a ceramic electrolyte at around 800 °C, reforms natural gas internally, and converts roughly 60% of the fuel's energy into electricity, which beats a combined cycle and, more importantly, holds at that efficiency in a 300 kW module, where no thermal cycle does. With heat recovery the total can exceed 90%. Because there is no combustion it permits where an engine cannot, and because it is modular and sits behind the meter it can be installed in months without entering the interconnection queue. It ramps slowly, since thermal cycling cracks ceramics, so it runs as baseload.

What sells it is speed and permits: near-zero criteria pollutants, a footprint that fits on a data-center pad, and delivery in months against five years for a grid connection or a new gas turbine. The efficiency at small scale is genuinely unusual and it is the reason the product exists. The honest limits: it is only as clean as the gas it burns, and at about 60% efficiency it emits roughly 0.30 tonnes of CO2 per MWh, close to a modern combined cycle rather than close to zero. It supplies no inertia and little fault current because it connects through an inverter, it will not follow load, and it needs a firm gas supply, so it trades delivery risk on the grid for delivery risk on the pipeline. The number that decides the pro forma is stack degradation: solid-oxide stacks lose output over time and get replaced on roughly a five-year cycle, and that recurring capital hides inside the service agreement rather than in the headline capex.

Choose a fuel cell when the binding constraint is a permit or a date rather than cost per MWh. That describes the data-center case exactly: gas available at the site, a five-year interconnection queue, and a NOx-constrained air district where a bank of engines cannot get permitted. It also fits hospitals, universities, and manufacturing sites that want on-site power with useful heat and cannot tolerate a diesel plant's emissions. Do not buy it for flexibility, because it will not follow load, and do not buy it as a decarbonization measure on its own, because on natural gas its CO2 per MWh is close to a combined cycle's. If you can wait for a grid connection, grid power under a PPA is usually cheaper. If you need flexibility rather than steady output, reciprocating engines are cheaper and faster wherever the permit allows them.

About 60% electrical efficiency on natural gas at any module size, and over 90% total with heat recovery · roughly 0.30 t CO2/MWh, close to a modern combined cycle · near-zero NOx, SOx, and particulates, which is what makes it permittable · stack replacement on roughly a five-year cycle, carried in the service agreement · roughly $3,000–5,000/kW installed before that agreement, at $100–200/MWh delivered on gas · installed in months against a US interconnection median above five years.

A fuel cell connects through an inverter, so it supplies no inertia, little fault current, and nothing for system strength. Its value is that it barely needs the grid: sited behind the meter next to the load it avoids transmission, distribution upgrades, and the interconnection queue, and with a microgrid controller it will island and keep the site running through an outage. It does not ramp, so it cannot provide reserve or follow load, and a system operator counts it as a permanent reduction in demand rather than as dispatchable capacity. What has to be built alongside it is a gas lateral with enough pressure and capacity, which on a large data-center installation is a pipeline project in its own right.

Bloom Energy's Energy Servers at Equinix, AT&T, and Home Depot sites, and the 2026 agreements that changed the category: up to 2.8 GW with Oracle, up to 1 GW with AEP under a $2.65B deal, a $5B partnership with Brookfield for AI data centers, and roughly $7.65B of data-center contracts inside a single quarter. Doosan Fuel Cell's phosphoric-acid fleet in South Korea, one of the largest stationary fuel cell fleets anywhere, built on a dedicated portfolio standard. FuelCell Energy's molten carbonate plants in Connecticut and South Korea. Bloom's older Delaware and California installations are the longest-running public field data on degradation.

The equipment is a manufactured product, so it rides a factory cost curve rather than a construction one, and installed cost lands around $3,000–5,000/kW before the service agreement. That agreement is the number to focus on, because it covers stack replacement on about a five-year cycle: a fuel cell's cost is part capex and part a recurring capital charge, and any model that treats it as ordinary O&M understates it. On natural gas the delivered cost lands around $100–200/MWh, above a combined cycle and below a peaker, and the buyers signing in 2025–26 are not buying on that number. They are buying delivery inside a year, a permit in an air district that will not take engines, and independence from a queue whose US median wait is over five years. If gas turbine lead times normalize or queues shorten, that premium compresses, so the durable question for the business is whether manufacturing cost falls fast enough to compete on price once speed stops being scarce.

3 minHow A Bloom Energy Server WorksBloom Energy · 100k+ views

6 minSolid oxide fuel cell - make electricity from natural gaschemicum · 50k+ views

4 minUnderstanding Solid Oxide Fuel Cell (SOFC) | Advantages, Disadvantages, ApplicationsOwl WiS · 10k+ views

Comparison of Fuel Cell Technologies (US Department of Energy) · Solid Oxide Fuel Cells (US Department of Energy)