Enhanced Geothermal Systems — Deep Hot Rock as Firm Power

EGS engineers a geothermal reservoir where nature did not provide one: drill into hot but impermeable crystalline rock, fracture it between an injection and a production well, and circulate water through the fractures to bring 175–200 °C brine to the surface, where a binary (organic Rankine cycle) plant converts roughly 10–15% of the heat into electricity and reinjects the rest. It is the opposite end of the geothermal spectrum from the shallow ground-source systems in Geothermal, which use ~20 °C earth as a sink. EGS uses ~200 °C earth as a source, and its product is firm power, not cold.

This article compiles the 2023–2026 evidence that EGS has finally worked at commercial scale, the cost curve behind that, the resource and policy frame, the fifty-year failure record that preceded it, and what all of it means for heat-driven cooling — which is worked out in Geothermal-Driven Sorption Cooling. Most operating data comes from one vendor via a non-peer-reviewed preprint and company blogs, corroborated on drilling by NREL; the failure ledger is peer-reviewed.

How a modern EGS works

Wells. DOE defines EGS as a well pair connected by hydraulically engineered fractures plus horizontal drilling, at 4,500–12,000 ft, with the reservoir in crystalline basement at ≥8,000 ft. Fervo’s design — the only EGS in commercial operation — drills a vertical section to ~2.4–2.9 km (8,000–9,500 ft) into hard rock (metasediments and granodiorite at Project Red, granitic basement at Cape Station), then turns horizontal for 1–1.4 km (3,250 ft at Red, ~4,700 ft at Cape), cased and cemented to the toe. Many wells share one pad — Cape’s first vertical plus 14 horizontals came from just two pads — which Fervo’s vendor page sells as lower footprint, lower drilling risk and access to “previously challenging” geologies. An injector and a producer are laid parallel, ~500 ft apart laterally and ~200 ft apart vertically; Cape stacks these in a multi-bench “wine-rack” of 3 injectors and 2 producers per bench, ~700 ft between benches.

Fracturing. The decisive change from 1970s–2010s EGS is the completion. Historical projects “bullheaded” unpropped water into long open-hole sections, and flow localized into one or two natural faults whose conductivity depended on holding pressure. Fervo instead uses multistage plug-and-perforate with proppant — 80 stages across the three-well Frisco triplet at Cape with zero screenouts, tight (~30 ft) cluster spacing, treating pressures mostly 7,000–9,000 psi under a 10,000 psi cap, and zipper fracs — producing many closely spaced propped fractures (modelled half-length ~670 ft, height ~550 ft) that stay conductive when pressure drops. Fervo’s own technology page never uses the words “stimulation” or “fracturing”; they appear only in its cited paper titles. Habanero’s lesson that “hot rocks are not dry” (its granite fault was overpressured by 34.5 MPa and kicked on drilling) is part of why cemented casing rather than open hole matters.

Sensing. Permanent DAS/DTS fiber behind casing in an observation well (Cape’s Delano 1-OB, DTS maximum 229 °C), plus downhole P/T gauges, map where flow actually goes and let the operator balance stages. Fervo frames fiber-optic sensing and analytics as its core differentiator alongside horizontal drilling; the same DAS data underpins its 2025 paper on seismicity during flexible operation.

Conversion. Cape Station is designed to produce at 390 °F (199 °C) specifically “to achieve optimal power conversion efficiency with modern ORC geothermal power plant technology,” which fixes the 8,000–9,000 ft target depth. Binary/ORC plants dominate all new geothermal builds (DOE Liftoff); flash plants are reserved for hotter hydrothermal fluids. Brine effectiveness at Cape is 94 kWe per kg/s at 195 °C; Project Red’s two-year average of 2.1 MW gross at ~36 kg/s and 175 °C implies ~58 kWe per kg/s. A research-agent derivation (not in Fervo’s text) puts Red’s gross ORC efficiency near 14% and Cape’s near 17–19%, consistent with the 10–15% net range in the review literature. Reinjection temperature is not published in any Fervo document; it enters only through the heat-in-place term (T_res − T_inj).

Reservoir life, water and flexibility. Well spacing trades near-term flow against thermal breakthrough: 300 ft injector–producer spacing declines rapidly, 600 × 200 ft is best over 15 years (~7.2 MW net per producer), tighter spacing wins only over 2–5 years; Fervo chose 500 × 200 ft. Project Red ran 500+ days with no thermal decline, then dropped ~2.5 °F, and shows a thermal recovery factor of ~20% after 600 days — the r = 0.2 that Fervo now uses in heat-in-place estimates. Net fluid loss at Cape is modelled at 0.36–0.49% over 15 years with producers on the outside of the pattern; Red recaptures ~70% (a near-field site beside the Blue Mountain hydrothermal plant). The fractured reservoir also works as a pressure and thermal buffer: Fervo’s in-reservoir storage papers (Ricks et al. 2021, 2022, 2024) and Cape’s demonstrated dispatch cycles let the plant follow wind and solar, with DOE citing round-trip storage efficiencies of 59–93%.

The 2023–2026 results

	Project Red (NV)	Cape Station (UT)
Rock, depth	phyllite/quartzite/diorite/granodiorite, ~2.4 km TVD, TD 3.4 km	granitic basement, 2.4–2.9 km TVD, TD 4.0–4.7 km
Lateral	~990 m	~1,430 m
Max reservoir T	191 °C	229 °C (DTS)
Production T	169 °C test → 175 °C steady	195 °C, rising through the test
Flow per producer	up to 63 L/s (test); ~36 kg/s avg in operation	120 kg/s initial → ~93 kg/s sustained, self-flowing (no pumps)
Electric output	3.5 MWe peak; 2.1 MW gross / ~1.4 MW net average	8–10 MWe sustained, >12 MWe peak per producer; ~16 MW gross per well projected
Status	on grid since 31 Oct / Nov 2023; 98.4% uptime, zero workovers over 614+ days	Phase I 100 MW (three ~33 MW GeoBlocks) COD Q4 2026; Phase II ~400 MW 2028; permitted to 2 GW

The benchmarks these beat: the best sustained historical EGS circulation was ~30 kg/s at Soultz (inflated by flow localizing in a few faults), Utah FORGE’s 2024 1,000 ft lateral pair reached 28 kg/s, and Habanero peaked at 19 kg/s. Power density is the headline: Cape’s multibench design gives 8.4 MWe/km³ of stimulated rock in the paper’s body (9.1 in its abstract) versus prior DOE estimates of 0.7–0.9 for 175–300 °C EGS — 5–10× higher.

Contracts and capital (as of Sept 2026). Fervo has ~862 MW under PPA — Google 396 MW from Cape Phase II (Sept 2026, the largest next-gen geothermal PPA, with an option on ~600 MW more and a right of first refusal to 3 GW), SCE 320 MW, NV Energy/Google 115 MW under a Clean Transition Tariff, Shell 31 MW — and a stated backlog of ~$7.2B. Financing went from venture equity ($462M Series E, Dec 2025; ~$1.5B raised since 2017) to $421M of non-recourse bank debt (Mar 2026, which the CFO noted “has historically been considered out of reach” for first-of-a-kind projects) to a $1.89B NASDAQ IPO at ~$10B valuation (May 2026) in eleven months. Fervo’s pitch to data-center buyers is speed: ~18 months from production drilling to COD against a ~5-year average US interconnection wait, Utah’s SB 132 enabling behind-the-meter delivery, and modular GeoBlocks (“if one GeoBlock underperforms, the remaining units continue to function”). Sources disagree on GeoBlock size — ~33 MW in Canary Media and Fervo’s release versus 50 MW in a Fervo blog, probably net versus gross. Meta has 300 MW with Sage Geosystems and XGS Energy. The buyers are hyperscalers, and every deal is for electricity, not heat.

Alternatives. Closed-loop / AGS (Eavor, XGS) never lets fluid touch rock; heat transfer is limited to borehole area, so DOE estimates a 5 km well needs ~20 km of laterals for 1–5 MW (TRL ~6). Eavor’s Geretsried plant delivered first grid power in Dec 2025 — four loops of ~16 km of wellbore each, 4.5–5 km deep in >200 °C granite, thermosiphon-driven — at 8 MWe plus 64 MWth of district heat on a ~€250/MWh feed-in tariff; ITIF calls AGS “best suited for district heating,” and it is the one next-gen project with a large thermal product. XGS’s 150 MW closed-loop plant for Meta (New Mexico, 2030) claims zero operational water. Superhot rock (Quaise) targets 315–415 °C at ~5 km with millimeter-wave drilling (field record 100+ m in 2025; commercial depths need 10–20 km); Project Obsidian (Oregon) aims for ≥50 MW around 2030.

The cost curve

Drilling is 30–57% of a geothermal plant’s overnight cost (DOE says ≥40%), more for EGS, where drilling plus stimulation can exceed half of project cost. The learning curve is steep and documented by NREL independently of Fervo:

Well	Days spud-to-TD	$/ft
Project Red 1 (2022)	71	995
Project Red 2	58	1,107
Cape 3 (2023)	34	633
Cape 4	26	472
Cape 6	20	346
Cape 8 (2024)	22	357

Red wells cost ~3× the Cape average — an inter-project learning rate of ~35% against a planned 18% — and Cape’s cumulative cost came in at 84% of budget. Rate of penetration rose from GeoVision’s assumed 25 ft/hr to ~70 ft/hr with PDC bits and 34-hour bit life. NREL’s 2025 model also raises casing-running speed (300 → 800 ft/hr), cuts contingency (15 → 10%) and doubles bit cost, landing horizontal-well costs 18–26% (vertical 12–24%) below the 2017 GeoVision baseline. NREL warns, though, that Utah FORGE’s 2022 speed gains did not translate to cost (FORGE drilled at 102 ft/hr but with 6.5-hour bit life and 16 bits per well): speed and cost are not the same curve.

The headline per-well halving is quoted with different bookends by different sources: Fervo’s blog gives $9.4M → $4.8M across the first four Cape horizontals, DOE Liftoff gives $9.5M → $4.8M “over six wells in six months” and separately “$13M → under $5M between the first two pilots,” and ITIF attributes a $13M → $5M drop to Utah FORGE. Read them as one ~50% reduction described three ways, with sub-$3M the stated goal.

Plant-level: DOE’s waterfall runs ~$27,800/kW (2021 ATB) → ~$14,700 (2023 FOAK) → ~$4,700–5,000 (2030 liftoff, $60–70/MWh unsubsidized) → ~$3,700 (2035 Enhanced Geothermal Shot, $45/MWh), against conventional geothermal at $61–102/MWh and PPAs at $70–100/MWh ($20–50 above solar). Fervo quotes $7,000/kW for Cape Phase I and $5,500/kW for Phase II against a $3,000 fossil-parity target. The IEA sees FOAK EGS at ~$14,000/kW today falling to $3,000–7,000 by 2035 and ~$50/MWh, if ~$1T is invested by then. Rhodium’s independent model gives $78–85/MWh nationally ($49–111 regionally), notes that dry-cooled plants need 15–30% thermal overbuild to hold flat output, and finds hyperscaler adoption (55–64% of data-center growth by the early 2030s) depends on a ~20% green premium, with federal-land permitting at 7–10 years and only 1.7 GW in queues. Barnard (CleanTechnica, early 2025, before the Cape data) puts EGS at $0.10–0.25/kWh, argues that 20–40 wells per 100 MW at ~200 °C makes “every single well its own mini-project,” and notes that bits degrade above 250 °C and electronics fail at 175–200 °C.

Resource, targets and policy

The resource claims are large and mostly modelled. Fervo cites a 2016 NREL study that heat within 5 km could meet all US electricity demand (7 km: five times over); DOE puts conventional hydrothermal at ~40 GW (3.7–4 GW on grid) against a next-generation resource of >5 TW, with 90+ GW deployable by 2050 toward a need of 700–900 GW of clean firm capacity. DOE’s liftoff pathway is 2–5 GW and $20–25B by ~2030, then 88–125 GW and $225–250B; Ricks et al. (2024) argue that ≥500 MW installed by 2030 lets learning make EGS out-compete other firm sources by 2050. The IEA sees geothermal supplying ~15% of global electricity-demand growth to 2050 and stresses that ~80% of components and skills already exist in oil and gas. The 2025 US policy turn favored geothermal specifically: inclusion in the “energy emergency” declaration (wind and solar excluded), a BLM categorical exclusion for resource confirmation, ITC/PTC retained through 2033, and roughly $400M of DOE program and field-test money.

The failure record — why EGS took fifty years

Pollack, Horne and Mukerji’s review of 64 EGS sites is the ledger every EGS claim should be read against (the authors call their counts minimums):

37% (24 sites) had drilling, completion or well-integrity problems that raised cost, delayed or killed the project. Habanero-3 failed by stress-corrosion cracking and blew out uncontrolled for 28 days; its replacement cost USD 34M. Newberry’s stimulation fluid escaped through casing leaks; Coso abandoned a 327 °C well over a stuck liner.
≥18 sites had reservoir failures: no connectivity, water loss (Ogachi lost 97% of injected water; Fenton Hill’s 7–12% loss “set the standard for reasonable”), or thermal short-circuit — Fenton Hill fell from 175 to 85 °C in a 75-day test because the “reservoir” was one joint; Hijiori fell 163 → ~100 °C; Bacman had to cut injection 187 → 70 kg/s. Soultz drilled from 1,700 to 5,000 m for a gain of only 140 → 200 °C, and one zone of GPK-3 carries 70% of its flow.
Pressure-dependent permeability meant injection pumps ran continuously at Rosemanowes and Fenton Hill, “significantly lowering the energy efficiency potential”; Bad Urach’s 2.9× transmissivity gain reverted to baseline within a year. Parasitic load is structural in unpropped reservoirs — the reason Fervo’s self-flowing Cape producer matters. Even a 30× injectivity improvement at Bradys left the well 7× short of the ~10 gpm/psi commercial threshold.
Scaling and corrosion ended or interrupted projects on their own: halite clogged Genesys Hannover’s tubing, CaCO₃ scaled Hijiori after thermal breakthrough, chloride corrosion caused a steam leak at the Northwest Geysers, and Soultz needed radiation-protection procedures for radionuclide-bearing scale on the reinjection line.
6 sites were shut by induced seismicity. Basel 2006 ML 3.4 (US$7M claims, criminal prosecution); Landau restarted only with EUR 50M/yr insurance and curtailed flow; Pohang 2017 — Mw 5.4 in Pollack’s tally, Mw 5.5 in Ellsworth’s — injured 90, displaced 1,700, did US$52M damage and ended Korea’s EGS program, from only ~12,800 m³ injected, which refutes the idea that small volumes guarantee small quakes; the main shock came ~2 months after the last stimulation on a previously unmapped fault. Ellsworth’s critique: traffic-light systems are tuned to felt quakes, not the rare damaging one. Habanero’s operators concluded seismicity “is unavoidable in a successful stimulation,” since first events came only 2 MPa above baseline pressure on a critically stressed fault.
Most abandoned projects ultimately failed on finance, having survived the engineering: Paralana lost AUD 37.5M of grants for want of AUD 5M in equity, and Australian EGS was squeezed by shale-era rig costs and remote transmission.

Habanero is the sharpest lesson: 230–264 °C granite, six deep wells, a binary plant that ran 160 days in 2013 generating only enough to run its own reinjection pump and camp — and abandonment “despite the technical success of the project” because circulation peaked at 19 kg/s against a 50–100 kg/s commercial need, since the only permeable structure was one pre-existing critically stressed fault (Jolokia-1, held above minimum principal stress for 14 days, took 1 L/s). Fervo’s propped multi-stage completions are best understood as the direct answer to that failure mode. They are not an answer to seismic risk or to finance, which remain open.

What EGS means for a cooling wiki

It is a power technology. Every design choice — 199 °C production, ORC optimization, PPA structure — is tuned to sell electrons. The natural cooling pathway from EGS is therefore EGS electricity → electric chiller at COP 3–6 (Vapor-Compression Cooling), not heat-driven cooling. The case for burning 200 °C brine in a COP-0.7 absorption chiller has to be made against that default; see Geothermal-Driven Sorption Cooling. Rhodium’s own note is that geothermal heat pumps and absorption chillers for data-center cooling are “still nascent.”
Its reject heat is probably too cold. Binary plants are designed to extract as much exergy as possible; the reinjection temperature is unpublished, but the only documented power-plant-bottomed chiller (El Haj Assad) assumes 100 °C reject from a flash plant. A sorption stage on an EGS plant most likely sits parallel to the turbine, taking brine before the ORC and giving up electricity.
The thermal product exists elsewhere. Closed-loop Geretsried sells 64 MWth of heat next to 8 MWe of power. If geothermal heat is ever piped to a chiller at scale, the district-heating-first architectures (Geretsried, Szeged) are where it will happen, not at a Fervo GeoBlock.
The plant’s own heat rejection is a cooling problem. Fervo advertises “extremely low water use” and Rhodium models dry-cooled plants with 15–30% thermal overbuild; in the arid basins where EGS sites are, the condenser sink — not the drive heat — is the scarce resource, the same lesson the shallow-geothermal article draws for chillers.
The firm-power argument still helps sorption indirectly. 24/7 geothermal electricity removes the intermittency problem that drives interest in thermal storage for solar-driven cooling (Solar Cooling: PV vs Thermal), and in-reservoir storage (round-trip 59–93%) makes EGS dispatchable.

Sources

Fervo Energy — Technology page — vendor framing: horizontal multi-well pads, fiber-optic sensing, binary-cycle 24/7 firm power, low water use, 2016 NREL resource claim, in-reservoir storage and DAS seismicity paper list
Norbeck, Gradl, Latimer 2024 — Cape Station cost reductions — Project Red and Cape well tests, plug-and-perf stage data, ORC design point, brine effectiveness, power density, drilling learning curve; companions: Project Red 2023 paper, two-year production data, URTeC reservoir design (well spacing, fluid loss)
Akindipe & Witter 2025 — NREL drilling cost curves — independent per-well days and $/ft, ROP and bit life, GeoVision assumption update, speed-vs-cost caution; companion Fervo drilling blog ($9.4M → $4.8M)
DOE Liftoff — Next-Generation Geothermal Power 2024 — EGS vs closed-loop taxonomy and depths, resource size, $/kW waterfall, Earthshot targets, liftoff phases, storage round-trip efficiency
Next-gen geothermal market status 2025–2026 — Cape Station status and GeoBlocks, PPAs, financing/IPO, Eavor Geretsried, Meta/Sage/XGS, Quaise, IEA/ITIF outlook and 2025 policy, Barnard and Rhodium critiques
Pollack, Horne, Mukerji 2021 — 64 EGS sites — the failure ledger: seismicity, water loss, short-circuits, well integrity, scaling, finance; companion Pohang summary
Hogarth & Holl 2017 — Habanero lessons — the technically successful, economically dead counter-case: overpressure, single-fault reservoir, pilot-scale flow, intrinsic seismicity