Geothermal Megaprojects

Four different technologies are routinely filed under “geothermal megaprojects,” and treating them as one technology is the characteristic error in this subject. They have different physics, different track records and different failure modes, and a claim that is true of one is usually false of the others. This brief keeps them apart throughout.

Conventional hydrothermal taps rock that is already hot and already permeable, and circulates fluid that is already there. It is a mature industry with a sixty-five-year commercial record, roughly 15 to 17 GW installed worldwide, and it is not the subject of this brief except as the baseline against which everything else is measured. Enhanced or engineered geothermal systems — EGS — drill into hot rock that lacks permeability and manufacture the permeability by hydraulic stimulation, then circulate injected water through the engineered fracture network. Closed-loop or advanced geothermal systems — AGS — circulate working fluid inside a sealed pipe and never contact the formation at all, trading reservoir risk for conduction through the borehole wall. Superhot rock aims deeper still, at temperatures above roughly 375 °C where water becomes supercritical and carries several times the enthalpy per kilogram, and is a drilling problem before it is anything else.

The subject of this brief is the attempt to remove the permeability constraint and thereby make geothermal available where the geology is not already favourable. That attempt is fifty-two years old, dating to the Fenton Hill hot-dry-rock experiment begun in New Mexico in 1973–74, and it has been about to scale for most of that period. What changed between 2021 and 2026 is real and is narrower than the coverage suggests: the completion toolkit developed for shale gas — cemented casing, plug-and-perf, limited-entry perforating, tight cluster spacing, large proppant volumes — was imported into hot crystalline rock, and it worked. It produced a measured thirty-seven-fold improvement in flow at a government field laboratory, and flow rates above 100 kg/s from commercial well pairs in Utah. That is the strongest result the field has ever had.

Three promotional claims define this slot, and the brief takes each separately because the evidence treats them very differently: that you can drill anywhere; that the result is baseload everywhere; and that the oil-and-gas drilling revolution transfers directly. One of the three is substantially true in a restricted form, and it is not the one that gets the headlines. The other two are contradicted by the operating record of the sector's own flagship plant and by the arithmetic of world capacity additions.

The scope boundary with sibling briefs is worth stating. Everything about making heat come out of rock — drilling, stimulation, reservoir behaviour, well flow, induced seismicity, capital cost per kilowatt — belongs here. Planetary-scale energy systems owns the question of what a firm low-carbon generation fleet does to a grid, and treats firm power as an input rather than a subject. Small modular reactors is the natural comparison rather than the natural neighbour: both sectors rest their economic case on a learning rate that has been asserted more often than it has been measured, and reading the two briefs against each other is more informative than reading either alone.

Established Start with the base rate, because it is the number the announcements omit. Of the four canonical EGS projects — Fenton Hill, Soultz-sous-Forêts, Cooper Basin/Habanero and Rittershoffen — zero reached sustained commercial electricity generation, and one of the four reached sustained commercial output of any kind. Fenton Hill remained experimental throughout, against an original hot-dry-rock concept that posited 50 MWe at 20% thermal efficiency; its most durable documented lesson is that “natural fractures and engineered fractures are almost unrelated.” Soultz, begun in 1986, produces 1.7 MWe from three wells of roughly 5 km, at 165 °C produced and 80 °C injected with a circulation of about 100 m³/h, and is classified in the peer-reviewed literature as a demonstration and research site rather than a commercial one; power production began on 24 June 2016, three decades after the project started. Habanero drilled past 4,139 m into granite at 230–264 °C, peaked at 15.7 kg/s in its 2008–09 tests and 19 kg/s in 2013, produced at 212–215 °C, ran a 1 MWe binary pilot for 160 days in 2013 powering site operations, consumed six wells and A$144.22 million, closed on 10 December 2015 with its wells plugged, and drew from its own operators the verdict that “the project failed to satisfy key economic measures.” Rittershoffen is the single success — two wells of about 2.5 km, 170 °C produced, roughly 110 m³/h, in industrial operation since 19 May 2016 supplying about a quarter of a bio-refinery's heat demand, with low induced seismicity — and it delivers 24 MWth of heat and zero megawatts electrical.

Established Widening the sample does not improve the base rate. A peer-reviewed systematic review adds Rosemanowes in the United Kingdom (research only, no power), Hijiori and Ogachi in Japan (neither commercial), Basel (never generated, permanently abandoned) and Landau in Germany (grid-connected combined heat and power, operating). Habanero's modelled heat recovery factor was “lower than 2%.” The review's own verdicts have aged well and are quoted rather than paraphrased: “EGS is still on a learning curve. Success is not guaranteed, and this implies significant financial risks,” and “the ‘typical’ EGS system does not exist.”

Established This is not only a sceptic's reading, and that is what makes it load-bearing. A 2025 peer-reviewed review in Nature Reviews Clean Technology, carrying three industry affiliations including the head of technology at the sector's leading developer, states that while several EGS projects operate commercially in Europe, most “operate for research purposes or produce modest amounts of energy (a few megawatts electrical or a few tens of megawatts thermal).” The commercial EGS fleet is single-digit megawatts electrical, and the people building it say so in a refereed journal. A concession against interest is worth more than an outside criticism.

Established The genuine breakthrough is real, recent, and narrower than advertised. At Utah FORGE, a United States Department of Energy field laboratory in granite at a bottom-hole temperature of about 429 °F (roughly 221 °C), unpropped stimulations in 2023 produced 0.7 kg/s and propped plug-and-perf stimulations produced 26 kg/s — a thirty-seven-fold difference, with 28 kg/s during circulation testing and 3.0 million lb of proppant placed across wells 16A and 16B. Sustained production of 26 kg/s was held across a thirty-day test in August and September 2024, from roughly 1,000 ft of stimulated lateral. That is the cleanest available evidence that importing shale completion design is what made EGS flow. Fervo's Project Cape wells in Utah subsequently reported initial production above 100 kg/s — 54,000 barrels per day — across the first well pair, with sustained rates of 95 to 100 kg/s from three laterals of about 4,700 ft.

Established And the flagship plant runs well below its own design. Project Red in Nevada, the 3.5 MW demonstration on which the “EGS is proven” narrative rests, has over 614 days averaged 2.1 MW gross and 1.4 MW net against a 3.5 MW design capacity, producing at 175 °C, injecting at about 75 °C, and circulating roughly 36 kg/s against a design specification of 63 L/s, with thermal decline of about 1.4 °C over the period and uptime of 98.4% excluding surface and grid outages. Both halves of that are true and the second does not rescue the first: the plant is highly reliable at a level roughly 40% under what was engineered. The operator published these figures itself, framed as a success story, which is why they can be relied on.

Established One caution on the flow shortfall, because the units do not match and this brief will not pretend they do. The design figure is volumetric (63 L/s) and the outturn is a mass rate (36 kg/s). At 175 °C the density of liquid water is roughly 890 kg/m³, so 63 L/s at production conditions is about 56 kg/s, and the shortfall is between 36% and 43% depending on whether the design rate was specified at surface or at reservoir conditions — which the source does not say. The arithmetic is this brief's; the inputs are sourced. The conclusion is unchanged in either reading: the flagship EGS demonstration circulates roughly two-fifths less fluid than it was built to circulate, and generates roughly 40% less power.

Established The gap between contracted and generating is the sector in one line. As of its first-quarter 2026 investor release, Fervo held 658 MW of executed binding power-purchase agreements, a framework agreement with Google for up to 3 GW through 2033, 2.6 GW described as advanced development and more than 38 GW early stage — and booked $61 thousand of revenue against a $31.8 million net loss and $172.8 million of quarterly capital expenditure, with roughly $1.2 billion of capital expenditure projected for the following four quarters. It raised $1.89 billion in a May 2026 flotation at $27 a share, opening at about $7.7 billion and rising a third the same day. Contracted: 658 MW. Generating: 1.4 MW net, from one plant. Cape Phase I, roughly 100 MW in three 33-MW GeoBlocks, reached mechanical completion in the first quarter of 2026 with first power guided for the fourth quarter.

Frontier The reservoir mechanism the field was founded on does not appear to operate. The FORGE analysis found “little to no evidence of shear stimulation” even when injection was held below the minimum principal stress — replicating null results at EGS Collab and Desert Peak. Shear stimulation, in which pressurised fluid causes pre-existing fractures to slip and self-prop, is the mechanism classical EGS was named for and premised on. The analyst's reading is that crystalline rock lacks the large-scale fault systems shear-stimulation EGS assumed. If that generalises, EGS is not the concept it was named for: it is hydraulic fracturing with a heat exchanger, and its economics are the economics of proppant, well count and lateral length. The field observation is established; its status as a general theory of crystalline reservoirs is not.

Handwave The forward per-well numbers have no operating well behind them. The developer's Generation 3.0 design is marketed at 16 MW per well. Its own modelled figure for Project Cape, in a conference paper co-authored with its modelling contractor, is a maximum fifteen-year average of about 7.2 MW per producer — less than half the marketing number, and itself a model output rather than a measurement. The long-run capital-cost aspiration of $3,000/kW “within five to ten years,” against a Phase II target of $5,500/kW, has no demonstrated pathway in any source consulted for this brief.

The genuine frontier in this subject is a reservoir-physics frontier and a materials-at-temperature frontier. It is not a resource frontier: the heat is unambiguously there, everywhere, and nobody disputes it. Separating what is open from what merely sounds open is unusually easy here, because the field has run the experiments and published the negative results.

Frontier The central open question is whether the reservoir behaves over years, not whether the well can be drilled. Every thirty-year financial model in the sector rests on an assumption about how a propped fracture network in granite holds conductivity as the rock around it cools and stresses redistribute. The longest continuous operating record on a modern EGS well is 614 days, with about 1.4 °C of thermal decline — genuinely encouraging, and roughly one-eighteenth of the period being extrapolated. Cape's modelled fifteen-year average of about 7.2 MW per producer comes from a model whose authors list its own exclusions: formation heterogeneity is assumed away across intermixed granite and gneiss, and thermoelastic effects — precisely the mechanism by which a cooling reservoir changes its own fracture apertures — are out of scope. The modelled net fluid loss of 0.36 to 0.49% over fifteen years is contingent on placing producers on the outside of the well pattern.

Frontier Unpropped fractures close, and there is nothing underneath them to fall back on. In shale, natural fractures retain modest permeability after injection stops, so an imperfect completion still produces. In the FORGE granite, unpropped fractures closed quickly with limited residual conductivity. That asymmetry is why the 0.7 kg/s figure is so low and why proppant is not an optimisation in this application but a precondition. It also means proppant survival at depth and temperature becomes a first-order reservoir variable rather than a consumable specification.

Frontier There is a scale gap nobody has crossed, and an unexplained flow gradient in the middle of the evidence. FORGE stimulated about 1,000 ft of lateral with 330 ft of vertical well spacing; commercial EGS plans call for 3,000 to 6,000 ft; shale routinely runs 10,000 to 20,000 ft. And FORGE produced 26 kg/s where Cape reports 93 to 121 kg/s — a fourfold difference reported by the same analyst in the same year, so it is a real difference between the government test bed and the commercial wells rather than a units artefact, and it is unexplained in what was published. Whether the gap is lateral length, stress state, completion intensity or reservoir quality matters enormously, because it determines whether FORGE is a floor that commercial practice has already cleared or an anomaly at Cape that will not repeat.

Frontier Flow localisation is a live worry the analyst raises against his own case. At FORGE, a single perforation cluster — Stage 10 — took 25.8% of total flow, which he describes as concerning. A reservoir in which a quarter of the circulation runs through one cluster is a reservoir with a short-circuit risk: that path cools fastest, and once it does, the thermal drawdown curve of the whole system is set by the worst-behaved fracture rather than the average one. He also notes that a maximum wellhead pressure of 8,000 psi prevented the designed pumping schedule in two stages, and that granite–gneiss intermixing makes it hard to separate design effects from geology.

Frontier A reportable disagreement in the literature rather than a settled point. The same analysis refutes a 2024 attempt by another group to infer formation permeability from microseismic propagation rates. This brief does not adjudicate it; it records that two groups working on the same facility disagree about what microseismicity can be used to measure, which is directly relevant to whether seismic monitoring can serve as a reservoir diagnostic as well as a hazard control.

Frontier Closed-loop is the frontier where two independent analyses reach opposite conclusions, and the brief reports the disagreement rather than resolving it. A 2022 University of Edinburgh modelling study, independent of the developer, found Eavor-Loop power-output claims “plausible, although the upper range of their predictions would likely require production temperatures in excess of 150 °C”; it also found that adding laterals reduces efficiency through thermal interference, that groundwater assists output only at hydraulic conductivities the authors doubt exist at depth, and that load-following by varying flow rate “is not established.” A 2025 study from a Spanish research institute, also independent, found that precisely the condition Edinburgh made the claim conditional on is what fails: only 3 of 14 tested configurations held production temperature above 150 °C after one year, with a 50 °C drop inside the first twelve days in some configurations. The two agree on the mechanism — conduction-limited heat exchange area, and efficiency loss from thermal interference as laterals are added — and disagree on whether the resulting numbers clear a commercial bar. The 2025 paper is later, more adverse and in a higher-profile venue; that is a reason to weight it, not a reason to call the question closed.

Frontier Superhot rock is a genuine frontier and is at an earlier stage than any coverage suggests. The physics is attractive and uncontested: supercritical water above roughly 375 °C carries several times the enthalpy of liquid brine, so a single well could in principle replace five to ten conventional ones. Iceland has drilled two wells into it. Neither produced electricity. Pilots above 330 °C are now reported in the year-in-review literature. The open questions are not about the resource but about casing metallurgy at 400 °C-plus, silica and corrosion chemistry in supercritical discharge, and whether any drilling method reaches the required depth at the required temperature.

Frontier What is not open, stated plainly because the field's reputation implies otherwise. The existence of the resource is not open. Whether horizontal drilling works in hot granite is not open — it does, at 21 days spud-to-total-depth for a 19,448 ft well. Whether proppant-based stimulation produces commercially interesting flow rates in crystalline rock is not open — it does, at above 100 kg/s. What is open is durability, cost at scale, seismic siting, and whether any of the closed-loop or superhot routes clear a commercial bar at all.

Established The first bottleneck is temperature, and it binds on the electronics rather than on the rock. Commercial measurement- and logging-while-drilling tools are rated 150 to 175 °C, with the industry constrained near a 200 °C threshold, and every 10 °C rise roughly halves electronics lifespan. Crystalline formations add shock and vibration up to 30 G RMS. The collision is direct and unresolved in the sources consulted: the newest generation of production wells sits at 460 °F, about 238 °C, above the stated commercial ceiling. Either specialised tooling is in use or tools are being run in cooled conditions; which of the two was not established, and the brief does not guess. This is the constraint that separates EGS at 200 °C, which is being done, from superhot rock at 400 °C-plus, which is not.

Established The second is what hard rock does to conventional hardware, which is easy to quantify and rarely quantified. In central Texas pink granite at 45,000 to 50,000 psi unconfined compressive strength, a hybrid particle-drilling bit ran 990 ft at 45 ft/hr where a single conventional polycrystalline-diamond-compact bit was worn out after 122 ft — an eightfold difference in bit life. Purpose-built tooling does better: at Utah FORGE a specialist bit reached a maximum rate of penetration of 43 m/hr and averaged 23 m/hr with minimal wear observed on the carbide outer compacts. Elsewhere in 200 MPa hard rock, 20 m/hr; in New Zealand, 16 m/hr. Drilling and completion is 40 to 60% of total project cost for a superhot project, so bit life is not a detail; it is most of the cost structure.

Frontier The third is that the learning curve is real, unpeer-reviewed at its primary source, and much noisier than it appears. Spud-to-total-depth fell from 70 days to 21 days across three well generations while measured depth rose from 11,220 ft to 19,448 ft; per-well cost fell from $9.4 million to $4.8 million across the first four horizontals; and a national laboratory independently corroborates a 35% interproject rate. But the primary paper — the developer's own September 2024 preprint — carries a repository statement that it has not been peer reviewed, and no subsequent journal publication was found. Meanwhile the independent cost data show a 2.9-fold spread across eight wells at the same project. A learning curve with that much residual variance is a real trend and a poor forecasting instrument, and a sector financing itself on the forecast should say so.

Frontier The fourth is well count, which is where flow rate turns into capital. Take the sector's own numbers at face value: a modelled fifteen-year average of about 7.2 MW per producer at Project Cape. A 400 MW Phase II therefore implies of order fifty-five producers plus injectors, each a 19,448 ft well at whatever the eighth-well cost turns out to be. The arithmetic is this brief's; the inputs are sourced. That is why the drilling learning curve is not a nice-to-have in this sector but the entire economic argument, and why a 2.9-fold cost spread across eight wells is a first-order financial risk rather than an engineering curiosity.

Established The fifth is capital intensity against a revenue line that does not yet exist. The leading developer booked $61 thousand of revenue against $172.8 million of capital expenditure in a single quarter, with roughly $1.2 billion projected over the following year. Its Phase II capital-cost target is $5,500/kW; the peer-reviewed review literature projects $4,500/kW and $80/MWh for 2027. Those are projections, not outturns, and no independent levelised-cost analysis was available to this brief — a gap stated rather than filled.

Established The sixth is social licence, and it is the bottleneck with a body count. Pohang and Basel both ended their projects permanently, and the peer-reviewed recommendation following Pohang is that current regulatory frameworks may underestimate risk in critically stressed fault zones. A German project was suspended after a 2009 event and required to carry €50 million of annual insurance. Insurance cost, siting restrictions and permitting delay are the mechanism by which the seismicity record binds the economics, and they bind before a single well is drilled.

Established What is not a bottleneck, stated because the sector's rhetoric implies otherwise. The resource is not a bottleneck; hot rock is under everywhere. Drilling speed in granite is no longer a bottleneck at EGS depths and temperatures. Flow rate per well, the constraint that killed Cooper Basin at 19 kg/s, is no longer a bottleneck at Cape's 95 to 100 kg/s. The bottlenecks are durability over decades, cost variance across wells, tool survival above 200 °C, and the seismic siting question — and none of them is solved by drilling faster.

Established Nothing here waits on a physics result. The heat exists, the drilling works, the stimulation works. What EGS waits on is high-temperature downhole electronics above the 175 °C commercial ceiling, bit and cutter metallurgy for crystalline rock, proppant that holds conductivity at depth and temperature, and enough repeat drilling to move down a cost curve with less variance than 2.9-fold. All four are industrial engineering problems with active work behind them and none is a discovery.

Frontier The one genuinely open scientific question is this brief's own and belongs to nobody else. Whether a propped fracture network in granite holds conductivity over years rather than months, and how fast the rock cools around it, is the variable every economic model in the sector integrates over thirty years and the variable for which the longest observation is 614 days. No other brief on this map will answer it; only time and Cape Phase I will.

Established It depends on public money at the demonstration stage, and the record is explicit. Utah FORGE is a United States Department of Energy field laboratory and every FORGE number in this brief comes from analysis of that public facility. Cooper Basin received A$32.75 million of Australian government funding against a total project cost of A$144.22 million. The breakthrough result of the last decade came out of a government test bed, not out of private capital, and private capital then applied it.

Established What depends on it is firm low-carbon power near load. That is a generic input to several topics in this corpus rather than a specific dependency on this route: planetary-scale energy systems needs firm generation and does not care which technology supplies it, and the same is true of the data-centre siting arguments in energy corridors. The honest statement is that nothing on this map depends on geothermal specifically, and several things depend on something in its category succeeding.

Established The decisive experiment already ran and its result is the thirty-seven-fold number. Utah FORGE compared unpropped against propped stimulation in the same formation and got 0.7 kg/s against 26 kg/s. It is a clean, government-funded, published comparison, and it settled which half of the shale toolkit transfers to hot crystalline rock. It is worth noting what makes it decisive: the rock, the depth, the temperature and the operator were held constant and only the completion changed. Very little else in this field has that structure.

Established The negative result in the same experiment deserves equal billing and rarely gets it. The absence of shear stimulation at FORGE, even with injection below minimum principal stress, replicates null results at EGS Collab and Desert Peak. Three independent facilities failing to observe the founding mechanism of a technology is a strong result, and it is a result about the concept rather than about any project. Recording it is more useful than recording another flow-rate record.

Frontier The experiment still running is Project Red's decline curve. Six hundred and fourteen days of operation with about 1.4 °C of thermal decline is encouraging and far too short to answer the thirty-year question the financial models assume. The next real datum is Cape Phase I, roughly 100 MW across three GeoBlocks, with first power guided for the fourth quarter of 2026 and the remaining two blocks in the first quarter of 2027 — the first time the technology will be asked to produce at a scale where its cost claims can be tested against an audited revenue line.

Established A closed-loop experiment has also returned, and it returned negatively. A peer-reviewed thermodynamic analysis with no industry affiliation found that only 3 of 14 tested configurations held production temperature above 150 °C after one year, with a 50 °C drop inside the first twelve days in some cases. It is a modelling result rather than a field result, which is the correct caveat and also the point: no operator-side field data from any closed-loop project was available to this brief. No Eavor-Lite performance figures, no Geretsried outturn, no measured production temperatures over time. The critique is independent and the rebuttal, if there is one, has not been published in a form this brief could verify.

Frontier The natural experiment on scale is already thirty years long and it is free to read. World installed geothermal capacity went from 8,272.7 MW in 2000 to 15,411.6 MW in 2024, and the annual series is published. If EGS is about to change the sector's trajectory, that series will bend within five years, and nobody has to build anything for the test to run. The prior it must beat is an average annual addition of 417 MW across 2014 to 2024.

Handwave The superhot experiment worth watching is 100 metres long. A millimetre-wave drilling field test reached 100 m in granite in July 2025, up from “just a few centimetres deep” in the originating laboratory system, against a stated ambition of a pilot power plant “as early as 2028” targeting rock at about 400 °C. The reporting itself states that 100 m “is only a fraction of the commercial depth needed.” That is a three-orders-of-magnitude gap to the target depth with no demonstrated operation at temperature, and no 2026 update was fetched for this brief.

Frontier Two experiments this brief wanted and could not obtain, named so that a reader can go and look. No induced-seismicity record for Project Red, Cape Station or Utah FORGE was retrievable — no largest-event magnitude, no traffic-light protocol description — which means this brief cannot say whether modern EGS has avoided Basel- and Pohang-scale events or has merely not yet been tested at comparable scale in comparable crust. And Mazama Energy's Newberry work, referenced in the peer-reviewed year-in-review literature at one facility reaching 331 °C, was not independently researched; it is potentially the most significant near-term superhot datum and it is not covered here.

Established Modern EGS is a shale completion executed in crystalline rock, and the specification list is the shale list. A well is drilled vertically to the target formation and turned horizontal for several thousand feet; the lateral is cased and cemented rather than left open; plug-and-perf isolation and limited-entry perforating distribute injection across tightly spaced clusters; and large volumes of proppant are placed to hold the fractures open after pressure is released. At Project Cape the first well pair used three laterals of roughly 4,700 ft with a median propped fracture half-length of about 670 ft and a height of about 550 ft. At FORGE, 3.0 million lb of proppant was placed across two wells to convert 0.7 kg/s into 26 kg/s. None of this is the classical EGS concept, in which pressurised water was expected to shear pre-existing fractures into permanent self-propped conductivity. It is the oil-and-gas completion toolkit, and its arrival is the whole of the recent progress.

Established The canonical project record, in full, because it is short enough to enumerate and is almost never enumerated.

Established The drilling learning curve, as published by the developer, is the strongest engineering claim in the sector and it is worth seeing as a table rather than as a chart. Note that depth, lateral length and temperature all rose while time fell, which is what makes the claim interesting; note also that the source is the developer's own wire release.

Established Cost followed time down, and the independent check both confirms the trend and shows what the trend line hides. The first four Cape horizontal wells fell from $9.4 million to $4.8 million per well, a realised learning rate the developer puts at 35% against a planned 18%, with a fastest well of 21 days and a rate of penetration of 70 ft/hr on the fourth horizontal. A United States national laboratory independently documents an interproject learning rate of 35% between Project Red and Cape Station, which corroborates the developer's headline figure — and simultaneously reports Cape Station horizontal wells at $346 to $994 per foot across eight wells, a 2.9-fold spread that no learning-curve line conveys. The same laboratory revised its baseline drilling cost curves down 18 to 26% for horizontal wells against the 2017 GeoVision baseline, so the improvement is real enough to have moved the reference case.

Established Closed-loop geothermal is the clearest case on this map of a marketing claim meeting a conductivity limit. A sealed loop never contacts the formation, so it has no permeability risk, no fluid loss and, in principle, no induced seismicity. What it also has is a heat-exchange surface limited to the borehole wall and a heat-supply mechanism limited to conduction through rock of roughly 3 W/m/K. Conduction cannot replenish extracted heat at useful flow rates, so production temperature falls steeply and then keeps falling: modelled drops of 50 °C in the first twelve days in some configurations, steep decline over twelve to fifteen days and slower deterioration thereafter. Megawatt-scale generation requires at least 15 laterals and 75 kg/s; the best case examined — 75 kg/s across 30 multilaterals — yields about 1.5 MWe and requires 158 km of total drilling. Drilling alone exceeds $65 million at $400 per metre of lateral, against a total project cost of roughly $72.6 million; at a 180 °C reservoir, revenue fails to recover thirty-year lifetime costs at a competitive 6.4 ¢/kWh and needs 12.6 ¢/kWh or more even on an aggressive $100-per-metre lateral cost. Reaching above 250 °C at an average Earth gradient implies drilling to 9.4 km. The paper's stated conclusion is that closed-loop systems “are not scalable for solely electricity generation” outside high-gradient fields, “directly contradicting industry claims of scalability.”

Established Superhot rock has produced two deep wells and no electricity. Iceland's IDDP-1 at Krafla, drilled in 2008–09, reached 450 °C at about 2.1 km, struck 900 °C magma which halted drilling, stuck twice and sidetracked twice, flow-tested from 2010 discharging fluid at 440 °C and 140 bar, and suffered corrosion and silica dust erosion and deposition before closing in 2012 after several surface valves failed. Its widely quoted “up to 36 MWe, five to ten times a typical commercial geothermal well” is a potential, not an outturn, and is circulated by an advocacy organisation. IDDP-2 at Reykjanes deepened well RN-15 from 2,507 m to 4,659 m slant depth in 2016–17, the deepest high-temperature geothermal well in the world, with bottom-hole temperature approaching 600 °C and complete circulation loss — and production was never initiated because the casing failed. Peer-reviewed follow-up finds casing damage below 2,300 m that prevents deeper logging, and that 90 to 95% of discharge now comes from the damaged shallow zone at a measured reservoir temperature of 294 °C in 2022 against 290 °C in 2016. The fluid actually produced from the world's deepest high-temperature geothermal well is a conventional high-temperature resource, not a supercritical one.

The most useful neighbour is small modular reactors, and the relationship is comparison rather than dependency. Both sectors propose firm low-carbon power near load; both rest their economic case on a learning rate demonstrated over a handful of units; both have order books far in excess of their operating fleets; and both are being financed substantially by data-centre demand. Reading the two briefs against each other is the single most informative thing a reader can do with either, because the failure modes are structurally identical and the physics is entirely different.

Planetary-scale energy systems is adjacent as consumer: it treats firm generation as an enabling condition and does not care which technology supplies it. The relationship runs one way. Nothing in that brief depends on geothermal specifically, and the honest reading of the capacity arithmetic here is that geothermal is currently supplying about three to four hundredths of one per cent of annual renewable additions, so it is not yet a material input to any planetary-scale account.

Energy corridors is adjacent by substitution. Siting firm generation near demand is the alternative to moving power a long way, and the data-centre procurement pattern documented in the institutional section is precisely a bet that the first is cheaper than the second. Whether that bet is correct depends on numbers in both briefs.

Underground cities shares the hard-rock engineering — the bit wear, the vibration environment, the cost per metre in crystalline formations — though at radically different depths and temperatures. Infrastructure resilience is adjacent through induced seismicity: Pohang is a case in which an energy project became a civil-protection event, and the governance question of who assesses that risk and to what standard belongs to both briefs.

Conventional hydrothermal generation is adjacent in a way this brief has to keep insisting on, because the two are constantly merged in coverage. It is the sixty-five-year-old industry that supplies essentially all of the 15 to 17 GW installed worldwide, it is what is actually delivering the only data-centre megawatts flowing today, and it is not the subject here. Every claim in this brief about base rates, durability and cost concerns the attempt to escape the geology that conventional geothermal requires.

Established The institutional record of this technology includes shuttering a project by ministerial decision after it damaged a city, and that is the correct starting point. Following Pohang, a government-commissioned panel concluded the event was triggered, the responsible ministry announced permanent closure, and residents sued the state. The plant had been in test operation for barely a year. Whatever else EGS is, it is a technology whose downside case has already been realised in a dense urban setting, and whose regulatory treatment in most jurisdictions has not been rewritten in response.

Frontier The institution that does not exist is a probabilistic seismic-hazard regulator for subsurface stimulation. The peer-reviewed recommendation coming out of Pohang is specific: deterministic assessment is inadequate and formal uncertainty analysis is required, because current regulatory frameworks may underestimate risk in critically stressed fault zones. That is a demand for a capability — an authority that can require an operator to establish, probabilistically and before stimulating, that a given site is not on a critically stressed fault. It exists nowhere this brief could verify. In its absence, siting judgement sits with the operator, and the operator's incentive is to drill where the rock is hot.

Established Who funds the science is not who sells the product. The thirty-seven-fold result came from a Department of Energy field laboratory; the Cooper Basin post-mortem was funded in part by an Australian government agency to the tune of A$32.75 million of an A$144.22 million project. Public agencies funded the demonstrations and the negative results; private developers commercialised the positive one. That is a defensible division of labour and it has a consequence worth naming: the negative findings in this brief are disproportionately public-sector products, and the positive findings are disproportionately private-sector ones.

Established Who buys is the most consequential recent change, and the buyer is not a utility. The 2025–26 demand signal is data-centre procurement, and the ledger is worth reading with its statuses intact. Signed: 31 MW from Shell Energy at Cape Station, bringing contracted Cape volume to 500 MW as of April 2025; 115 MW from Google in Nevada via the incumbent utility, delivery stated as “by 2026”; 10 MW from Google with a Taiwanese developer alongside an equity investment; up to 150 MW from Meta with Sage Geosystems, delivery by 2027; 50 GWh a year from Panasonic in Japan; and about 13 MW from Ormat to Switch on a twenty-year term in January 2026. Announced but not contracted: up to 150 MW from Google via Ormat under a Nevada clean-transition tariff, subject to state regulatory approval expected in the second half of 2026, with projects online 2028 to 2030 and the technology not specified as conventional or advanced; and a framework with Google for up to 3 GW through 2033, which is a framework rather than contracted capacity.

Established The pattern in that ledger is the finding. The largest headline numbers are frameworks, regulatory-contingent, or “up to.” The only megawatts flowing to a data centre from a plant that already exists are the roughly 13 MW from Ormat's Salt Wells facility in western Nevada — a conventional hydrothermal plant, not an EGS one. The most advanced institutional relationship between geothermal and the data-centre industry is currently being served by sixty-year-old technology. Whether the 115 MW Nevada agreement is actually delivering was not verifiable for this brief, and given the flagship plant's 1.4 MW net output it is a question a reader should ask rather than assume.

Frontier Concentration is a fact on both sides of the market. Ten countries hold more than 93% of world geothermal capacity: the United States at 3,953 MW, Indonesia 2,742, the Philippines 2,034, Türkiye 1,797, New Zealand 1,259, Kenya 980, Mexico 976, Italy 916, Iceland 808 and Japan 607. On the buy side, one hyperscaler appears as counterparty in four of the deals above. And the leading developer's own flotation filing discloses that its anchor customer's priority provisions “may limit our flexibility to pursue alternative commercial, strategic, or financing arrangements” — a public admission that the sector's largest developer has one customer whose contractual position constrains its strategy.

Established What would change the institutional picture, stated as a testable condition. A regulated utility procuring EGS capacity through a rate-base process, rather than a hyperscaler procuring it through a bilateral corporate agreement, would put the technology in front of a body obliged to examine its cost against alternatives on the public record. The Nevada tariff proceeding expected in the second half of 2026 is the closest thing to that test currently scheduled, and its outcome will say more about EGS's commercial standing than any flow-rate announcement.

Established The ethical question here is unusually concrete, because the harm has already occurred, is documented, and is now attributed in peer-reviewed literature. On 15 November 2017 an earthquake of Mw 5.4–5.5 struck Pohang, South Korea, near an EGS project that had been in test operation since 2016. Dozens were injured, more than 1,500 residents were left homeless, the national college entrance examination was postponed, and residents sued the government. A government-commissioned panel concluded it was a triggered rather than a natural earthquake; the responsible ministry announced the plant — built for about 80 billion won, roughly US$71 million — would be permanently shuttered. A 2026 peer-reviewed analysis with no industry affiliation concludes it was induced by high-pressure hydraulic stimulation in well PX-2, which activated a pre-existing low-permeability fault.

Frontier The two numbers from that analysis matter more than the attribution, because they convert an accident into a governance failure. The stochastic poromechanical model yields a 7 to 15% exceedance probability for the mainshock — 7% under a reverse-slip-dominated stress regime, 15% under a strike-slip-dominated one — consistent with independent statistical estimates of about 6 to 16% from recorded magnitude–frequency relations. A one-in-seven to one-in-fourteen chance of a damaging earthquake is a foreseeable tail, not a freak event. And the stress threshold for fault reactivation at Pohang was about an order of magnitude larger than that proposed for natural earthquakes — roughly −0.2 to −0.1 MPa against about −0.01 MPa — meaning the intuition trained on natural seismicity understates how much perturbation a critically stressed fault will absorb before it slips, and therefore how confidently a deterministic assessment can clear a site. The authors' conclusion is explicitly a governance one: formal uncertainty analysis must replace deterministic assessment, and current regulatory frameworks may underestimate risk in critically stressed fault zones.

Established Basel is the other case, and this brief must state that its own evidence for it is weak. Main stimulation began on 2 December 2006 and was suspended six days later on 8 December after a trigger event of ML 2.9; the largest recorded event was ML 3.4 a year later; over 13,500 events were recorded with nine at ML 2.5 or above; an insurer processed roughly 2,700 claims estimated at CHF 7–9 million; and the project was cancelled in December 2009 after a three-year study predicted the town would continue to feel small earthquakes several times a year across a thirty-year project life. The only source available to this brief for that chronology is an encyclopaedia entry, which is below the standing bar the corpus sets. The SERIANEX risk assessment, the peer-reviewed seismological work on the Basel sequence, and the legal and insurance disposition for the operator were all unverified. The numbers are stated because they are load-bearing for the governance argument and suppressing them would mislead more than including them; they are flagged because a reader is entitled to know what they rest on. Independently and from a peer-reviewed source: a German project at Landau was temporarily suspended after an M2.7 event in 2009 and required to carry €50 million of annual insurance.

Frontier And here is the gap this brief most wants closed and cannot close. No induced-seismicity record for any modern EGS project was obtainable. The operator's and project's seismicity pages returned empty or were disallowed to automated retrieval; there is no figure here for the largest induced event at Project Red, Cape Station or Utah FORGE, and no description of any operator's traffic-light protocol. This brief therefore cannot say whether the modern generation of EGS has avoided Basel- and Pohang-scale events through better siting and monitoring, or has merely not yet stimulated at comparable scale in comparably stressed crust. That is a materially different question from the one the sector's safety messaging answers, and anyone drafting from this brief should refuse to fill it in from memory.

Established Evidence quality, stated as a disclosure obligation. A substantial share of what is publicly known about EGS performance comes from the developer, from its modelling contractor, or from trade press relaying both. The two most-cited technical analyses of FORGE and Cape share an author who is an interested party analysing a public facility. The peer-reviewed 2025 review carries three industry affiliations; the 2026 year-in-review carries declared industry conflicts and rests substantially on 2025 conference proceedings rather than refereed papers. This brief marks interested parties throughout, relies on them for figures they would be embarrassed to get wrong, declines to rely on their forward projections, and prefers the independent and government sources for anything comparative. It also records that a content-farm cluster generating plausible-looking geothermal deal pages with specific dollar and megawatt figures was encountered during research and was not used; readers assembling their own numbers should be aware that this subject now has a synthetic-source problem.

Frontier Public money and opportunity cost. The decisive experiment of the decade ran at a Department of Energy field laboratory; the Australian taxpayer contributed A$32.75 million to a project that closed with its wells plugged. That is public money working as intended — funding the demonstrations private capital will not — and it is also money that could have gone elsewhere in the firm-power portfolio. The open question this brief cannot answer, and states rather than dodges, is whether the sector's projections of $4,500/kW and $80/MWh for 2027 or $5,500/kW at Phase II are competitive with the alternatives, because no independent levelised-cost analysis and no comparison against firm alternatives was available. Every cost number in this brief is a developer target or a review-paper projection. None is an audited outturn, and the first one arrives with Cape Phase I.

Frontier The civilisational prize is genuine and specific. Firm, dispatchable, low-carbon power with a small surface footprint, available where the load is rather than where the wind is, with no fuel cycle, no proliferation question and no seasonal storage requirement, would be among the most useful technologies on this map. Power density at Project Cape is reported at around 30 MW per square kilometre, which is one to two orders of magnitude above wind or solar. If EGS worked at the cost its proponents project, the argument would be over.

Established Against that, the deployment arithmetic is the fifty-year problem stated as a number. World installed geothermal capacity went from 8,272.7 MW in 2000 to 15,411.6 MW in 2024 — a compound rate of 2.63% a year, a doubling time of about 27 years, and an average annual addition of 417 MW across the last decade of the series. The rate is decelerating: 2.2% across 2019 to 2024 against 3.2% across the preceding decade. In 2025 global renewables added 692 GW, reaching 5,149 GW total; geothermal added 0.22 to 0.3 GW. Ten countries hold more than 93% of the world's geothermal capacity, and the largest, the United States at 3,953 MW, still gets under 1% of its electricity from it.

Established The general principle this case illustrates is about the difference between a demonstrated capability and a deployed one. Geothermal has had a commercially proven technology since 1960, an unlimited resource, and no physics problem for the whole of that period, and it has grown at under 3% a year for a quarter of a century. What binds is not knowledge but the conjunction of subsurface uncertainty, capital intensity, well-to-well cost variance and a permitting environment shaped by two well-remembered failures. Any argument that a technology will scale because the physics works is answered, in this sector, by fifty-two years of the physics working.

Speculative The case for reversal, stated fairly. The completion-toolkit import is genuinely new; the well-cost trajectory is genuinely steep; and for the first time the sector has a demand signal from buyers who value firmness and proximity enough to pay a premium for both. If Cape Phase I holds its output and Phase II hits $5,500/kW, the base rate this brief leans on will have been broken by the first project that was ever properly capitalised. That is a real possibility and it is not the way to bet on the evidence available in September 2026.

Established What already happened, because the timeline usually starts in 2021. Hot dry rock began at Fenton Hill in 1973–74. Soultz began in 1986 and reached power production on 24 June 2016. Basel stimulated for six days in December 2006 and was cancelled in December 2009. Cooper Basin ran a 1 MWe pilot for 160 days in 2013 and closed on 10 December 2015. Rittershoffen entered industrial operation on 19 May 2016. Pohang stimulated from 2016, triggered a damaging earthquake on 15 November 2017, and was permanently shuttered. Fifty-two years of effort stand behind the current moment, and four of those six projects ended in closure.

Established 2021 to 2026: the shale-completion import. Project Red drilled in 2022 and came online in 2023. Utah FORGE ran propped stimulations and a thirty-day circulation test in August and September 2024. Cape Phase I wells were drilled through 2024 and 2025 and reached above 100 kg/s. The developer floated in May 2026 raising $1.89 billion. This is a five-year interval in which the engineering changed genuinely and the generating fleet grew by 3.5 MW of nameplate.

Frontier Late 2026 to 2027: Cape Phase I, and the first honest cost outturn. First power from GeoBlock 1 is guided for the fourth quarter of 2026, with GeoBlocks 2 and 3 in the first quarter of 2027, for roughly 100 MW total. Whether a 100 MW plant performs like the 3.5 MW one is the whole question, and the answer arrives as an audited number rather than as a press release.

Handwave 2028: Cape Phase II at roughly 400 MW, and the data-centre deliveries. Eight 50-MW GeoBlocks are guided for 2028, at a target capital cost of $5,500/kW. The Google–Ormat arrangement, if approved by the Nevada regulator in the second half of 2026, brings projects online in 2028 to 2030. Every one of these dates is a developer or counterparty statement about a plant that does not exist, and geothermal schedules in this corpus have a poor record: the sector's own flagship arrived 40% under design capacity.

Frontier Five years: a five-year decline curve on a propped granite reservoir either exists or does not. Every thirty-year financial model in the sector is currently extrapolating from 614 days. This is the milestone that would actually change what can be claimed, and it is the one nobody publishes a date for, because it consists of waiting.

Frontier Ten years: the sector either bends the capacity curve or does not. The test is arithmetic. Either EGS is adding gigawatts a year rather than the 417 MW a year the whole sector averaged across 2014 to 2024, or geothermal has settled into the niche conventional hydrothermal has occupied since 1960 — excellent where the geology is already favourable, marginal elsewhere. A United States projection of 8.5% of national power supply by 2050, against under 1% today, is the optimistic pole and is a projection rather than a plan.

Speculative Twenty-five years and beyond: superhot rock at commercial scale. This requires a drilling technology that has demonstrated 100 m against a requirement of several kilometres, at temperatures above every commercial downhole tool rating, with casing that has already failed once at 4,659 m. The independent engineering review's verdict is that direct-energy methods are “highly novel… and may not be commercializable this decade.” No date is offered here, because the sources do not support one.

• Depends on Nothing on this map. The one genuinely open scientific question — whether a propped fracture network in hot crystalline rock holds conductivity over years rather than months — is this brief's own to answer, not another brief's result to wait for. Everything else the field waits on is industrial, financial or institutional, and is recorded below.
• Requires (not on this map) Downhole measurement and logging electronics that survive above the 175 °C commercial ceiling, since every 10 °C roughly halves tool life and the newest wells sit near 238 °C. Bit and cutter metallurgy for crystalline rock, where a conventional bit lasted 122 ft against 990 ft for a specialist one. Proppant that holds fracture conductivity at depth and temperature, because unpropped fractures in granite close and there is no natural permeability underneath to fall back on. Enough repeat drilling to move down a cost curve with less than 2.9-fold well-to-well variance. A regulatory framework that requires probabilistic rather than deterministic seismic-hazard assessment before stimulation in critically stressed crust — the explicit recommendation of the peer-reviewed Pohang analysis, and an institution that does not currently exist in most jurisdictions. And a price signal for firm low-carbon power that is set by something other than a handful of hyperscaler procurement teams.
• Enables Firm low-carbon power near load, which several topics in Categories VI and VII assume generically. No typed enabling edge is claimed: those briefs depend on firm power existing, not on this particular route to it, and on current evidence the route is not the leading candidate.
• Adjacent Small modular reactors, the competing firm-power claim resting on the same unmeasured learning-rate argument; underground cities, which shares hard-rock excavation and the same tool-at-temperature constraints; infrastructure resilience, for induced-seismicity governance; planetary-scale energy systems; and the unbuilt Energy Systems slot 9, Energy Storage Revolutions, which is the natural pair for any firm-power claim.

“EGS is proven.” Handwave Zero of the four canonical EGS projects reached sustained commercial electricity generation, and the peer-reviewed 2025 review co-authored by the leading developer's own technology head describes the operating commercial fleet as producing “a few megawatts electrical or a few tens of megawatts thermal.” What is proven is that propped plug-and-perf stimulation moves thirty-seven times more fluid than unpropped stimulation in hot granite, and that a modern commercial well pair can produce above 100 kg/s. Those are real and important results and they are not the same claim as “proven.” The distinction that matters: flow is demonstrated; durability, cost at scale and commercial electricity are not.

“The shale revolution transfers directly.” Frontier Split this into two claims, because one is supported and one is not, and testing the weak version is a strawman. The strong version of the industry's claim — that shale completion design transfers — is supported by the FORGE result of 0.7 to 26 kg/s. The reservoir-mechanism claim fails: field analysis found little to no evidence of shear stimulation even below minimum principal stress, and unpropped fractures closed rapidly with no natural-fracture permeability to fall back on. The hardware-environment claim also fails: commercial downhole tools are rated 150 to 175 °C where the newest wells run near 238 °C, and a conventional bit that lasted 122 ft in granite would have lasted a great deal longer in shale. Completions transfer. Reservoir physics and tool ratings do not.

“Pohang was a magnitude 5.5 earthquake caused by geothermal drilling.” Established Three separate corrections are needed and each matters. First, the magnitude is not settled: the two 2018 papers in Science disagree in their own titles, Grigoli and colleagues using Mw 5.5 and Ellsworth and colleagues using Mw 5.4; the Korean government panel and contemporaneous reporting used 5.4; the 2026 peer-reviewed analysis uses 5.5. This brief writes it as Mw 5.4–5.5, sources differ. Second, the 2018 attribution was hedged and the confident attribution is recent: those papers are titled “a possible case of induced seismicity” and “assessing whether… was an induced event.” The unhedged statement that the earthquake was “induced by high-pressure hydraulic stimulation in the well PX-2, which activated a pre-existing low-permeability fault” is from February 2026, eight years later. Do not retroject 2026 confidence onto 2018. Third, it was not caused by drilling but by stimulation — the injection of high-pressure fluid to create permeability — which is the step closed-loop designs omit entirely and conventional hydrothermal never performs.

“Fervo has delivered a lot of capacity.” Established As of the first quarter of 2026 the delivered fleet is one 3.5 MW-nameplate plant averaging 1.4 MW net, on quarterly revenue of $61 thousand. Cape Station has delivered zero megawatts; first power is guided for the fourth quarter of 2026. The correct framing is not “contracted versus delivered” but contracted versus nothing yet delivered, and the comparison is 658 MW to 1.4 MW. This is not an accusation of bad faith — the company discloses all of it — but the framing is routinely lost in coverage.

“They published a learning curve, so it is established.” Frontier They published one, and the primary paper is an EarthArXiv preprint from September 2024 whose own repository page states that it has not been peer reviewed, with no evidence of subsequent journal publication. The independent corroboration a reader wants does exist, and it comes from a national laboratory rather than from peer review — which is a reasonable evidentiary position and is not the same thing as a refereed result. The same laboratory's data show the $346-to-$994-per-foot spread that the smooth curve conceals.

“Cape Station is a 400 MW project.” Established A naming problem with real consequences. The technical literature calls the Utah project “Project Cape”; corporate and financial material calls it “Cape Station,” split into Phase I at roughly 100 MW and Phase II at roughly 400 MW. Anyone citing “400 MW Cape Station” is citing Phase II alone; the total is about 500 MW; and the two phases have different schedules, different well designs and different cost targets. Citations that do not name the phase are not interpretable.

“Closed-loop avoids the reservoir problem.” Handwave It replaces a permeability problem with a conductivity problem, and rock loses that argument. The peer-reviewed independent case gets about 1.5 MWe out of 158 km of drilling and needs roughly double the competitive power price to pay for it. But report the disagreement rather than only the adverse verdict: an earlier independent modelling study from a British university called the developer's claims “plausible” conditional on production temperatures above 150 °C, and the 2025 study finds that exactly that condition fails in 11 of 14 modelled cases within a year. Two independent groups, opposite conclusions, agreeing on the mechanism. This brief does not resolve it, and notes that no field data from any closed-loop operator was available to weigh against either.

“Iceland has demonstrated superhot geothermal.” Established Iceland has demonstrated superhot drilling, which is a different and still-remarkable achievement. Neither IDDP well ever produced electricity. IDDP-1 closed on surface valve failures in 2012 after corrosion and silica erosion; IDDP-2 never initiated production at all because its casing failed, and what it now discharges is 90 to 95% shallow-zone fluid at 294 °C — a conventional high-temperature resource. The frequently cited “IDDP-1 could have generated 36 MWe” is a potential, never an outturn, and it is circulated by an advocacy organisation.

“Geothermal is scaling fast now.” Established World installed capacity grew at a compound 2.63% a year from 2000 to 2024, a doubling time of roughly 27 years, and the growth rate has fallen: 2.2% across 2019 to 2024 against 3.2% across the preceding decade. In 2025 global renewables added 692 GW; geothermal added 0.22 to 0.3 GW, about three to four hundredths of one per cent of the year's renewable build. Ten countries hold more than 93% of world capacity. Nothing in the last five years has bent that curve, and the curve is where a scaling claim would show up first.

“Global geothermal capacity is X GW.” Established There is no single number and anyone quoting one without a source is quoting an artefact. For the same date — end of 2025 — an intergovernmental statistical authority gives 16 GW and a trade tracker gives 17,173 MW, a gap of roughly 1.2 GW that neither source explains and that probably reflects differing treatment of off-grid, derated and non-operating plant. The brief reports both rather than choosing, because neither has better standing on the point and picking one would manufacture a precision the evidence does not support.

“Contracted gigawatts mean the technology works.” Frontier One developer holds 658 MW of binding agreements against one operating plant producing 1.4 MW net. Beyond that, the largest headline numbers in the data-centre wave are structurally softer than they read: a 3 GW figure is a framework, not contracted capacity; a 150 MW figure is subject to a state regulator's approval expected in the second half of 2026 for projects online in 2028 to 2030, with the technology not even specified as EGS or conventional; another 150 MW is an “up to.” Contracts are evidence about capital markets and corporate procurement policy. They are not evidence about reservoirs.