The Lens · Firm Power for the AI Buildout

Geothermal Went Public. Twelve Days Later Its Flagship Well Blew Out.

Q: What is enhanced geothermal?

Enhanced geothermal drills into hot, dry rock, fractures it, and circulates water through the fractures to bring heat to the surface, so geothermal power works far beyond natural hot springs. Fervo Energy cut drilling time about 70 percent by applying oil and gas drilling techniques.

Q: What happened at Fervo's Cape Station in May 2026?

On May 25, 2026, drilling rods broke while being removed from a well and pressurized fluid surged to the surface. Fervo and Utah state officials reported no injuries, non-toxic stimulation water, no environmental damage and no impact to the project schedule. The incident came twelve days after Fervo's 1.9 billion dollar IPO, about 2.2 billion with the over-allotment.

Q: What is a sand battery?

A sand battery is a large insulated mass of sand or crushed stone that is heated with cheap electricity and discharged as heat later. Finland's 1 MW, 100 MWh unit in Pornainen runs a town's district heating. Heat batteries are cheapest where the end use is heat; converting stored heat back into electricity is lossy.

Q: Who benefits from a geothermal buildout?

Drillers and oilfield services firms whose rigs and crews are needed regardless of which developer wins, developers with signed power purchase agreements, turbine and heat exchanger makers, and, for industrial heat, heat battery makers such as Polar Night Energy, Rondo and Antora.

On May 13, 2026, Fervo Energy raised $1.9 billion (about $2.2 billion once the over-allotment was exercised) in what its bankers called the largest primary clean-energy equity offering on record. On May 25, a well at its flagship Utah project blew out. Both things are true, and the honest story about next-generation geothermal — 24/7 power from the heat under our feet, the kind AI data centers are desperate for — lives in the space between them. Here is the promise, the setback, the heat batteries quietly scaling beside it, and who gets paid. The one-line version: $7.2 billion of contracted paper, $60–80 million of expected 2027 revenue, and a well that unloaded twelve days after the bell.

Dragonfly Lens · 27 August 2026 · Reviewed against a second independent read before publication · Every figure sourced and dated. Targets are labelled as targets.

The short version

What’s real: Fervo borrowed the oil industry’s horizontal-drilling playbook and, on its first Cape Station wells (reported 2024), cut drilling time 70% and cost per well from $9.4M to $4.8M — and in July 2026 drilled a deeper, hotter, longer well (19,448 ft, 460°F, 7,500 ft lateral) in 21 days. Its 500 MW Cape Station in Utah is under construction: first power targeted late 2026, ~100 MW by early 2027, phase two 2028 — with 658 MW of binding contracts (Southern California Edison, Shell and others) worth ~$7.2B of backlog. Meta separately agreed to buy up to 150 MW from Sage Geosystems at an undisclosed site east of the Rockies — and Sage’s first plant, a 3 MW unit in Texas, was placed in service in August 2026.
What went wrong: on May 25, during work on a Cape Station well, pressurized fluid surged to the surface (reports describe drill rods breaking as they were pulled). No injuries, non-toxic stimulation water, no environmental damage or schedule impact, per Fervo and Utah officials. Real, remedied, and far less dangerous than an oil-and-gas blowout (no hydrocarbons) — but a reminder that this is drilling, with drilling’s risks.
The sleeper beside it — heat batteries: a 1 MW / 100 MWh “sand battery” runs a Finnish town’s district heat on 2,000 tons of crushed soapstone; Rondo switched on a 100 MWh industrial heat battery in 2025; Antora raised $550M in July 2026 to build carbon-block heat batteries aimed at data centers. Cheap, boring, proven physics — with one honest catch below.
The number under the headline: against that $7.2B backlog, Fervo guides 2027 revenue of just $60–80 million; Phase I is three 33 MW blocks (~100 MW), Phase II targets $5,500/kW all-in on the way to a long-run $3,000/kW, and second-half 2026 capex is $850–900 million. The backlog is twenty-year paper; the plant is not built yet.
The government’s bet: the U.S. DOE’s Enhanced Geothermal Shot targets 5 GW by 2030, 90 GW by 2050 and a 90% cost cut by 2035. Ambition, not a forecast.
The Lens angle: if 24/7 heat from the ground becomes firm power for AI, the value pools with the drillers and oilfield-services firms (same rigs, new hole), the developers with signed contracts, the turbine and heat-exchanger makers — and in industrial heat, the battery makers. Own the shovel, not the story.

Why geothermal, and why now

Data centers need power that runs all day, every day. Solar and wind are cheap but intermittent; gas is firm but carbon-heavy and turbine-constrained; nuclear is firm but a decade away (see our nuclear piece). Geothermal is the one clean source that is firm and can be built in years rather than decades — if you can reach hot rock cheaply. Traditional geothermal needed rare natural hot springs. Enhanced geothermal makes its own reservoir: drill deep, crack the rock, pump water through, bring the heat up. The insight that changed the economics was simply using oil-and-gas drilling technology — horizontal wells, modern drill bits, mud coolers — on geothermal formations. The headline numbers (70% less drilling time, $9.4M to $4.8M per well) come from Fervo’s first Cape Station wells, reported in 2024; the learning curve has kept going — in July 2026 its Sawtooth 7 well reached 19,448 ft in 460°F rock with a 7,500 ft lateral in 21 days, matching that 70% gain on a far harder hole.

Zoom out: how big is geothermal, really?

Before the IPO story, the scale. The entire world’s geothermal power fleet is about 16 GW (IRENA, end of 2025 — up from 15.4 GW a year earlier, a crawl). The United States leads with 4 GW, a quarter of the global total and roughly 1% of U.S. electricity; 53 of its 99 plants sit in California. So a single 500 MW Cape Station would be about 3% of all the geothermal power on Earth, and the incumbent leader, Ormat, operates 932 MW worldwide after decades in the business. This is a small industry — which is exactly why the next-generation claims are so large relative to it.

The trajectory being priced is the IEA’s: if the cost-downs land, next-generation geothermal could reach 120 GW by 2035 and over 800 GW by 2050 — about 8% of world electricity, roughly fifty times today’s entire fleet — with costs falling ~80% to around $50/MWh and cumulative investment of $1 trillion by 2035. That curve has barely started. Here is who is on it, measured against claimed:

Player	Technique	Measured (Aug 2026)	Claimed / contracted
Ormat (incumbent)	Conventional hydrothermal	932 MW operating worldwide (647 MW U.S.) — the largest operator.	Up to 150 MW new for Google via NV Energy (signed Feb 2026, online 2028–30, regulator approval pending); 13 MW for Switch.
Fervo	Enhanced geothermal (EGS), horizontal wells	Nevada pilot online; Cape Station Phase I under construction, first power targeted Q4 2026; $2.2B IPO.	~100 MW early 2027; 400 MW Phase II 2028; 658 MW binding PPAs; 3 GW Google framework (non-binding).
Eavor	Closed-loop (no fracturing, no water withdrawal)	First closed-loop grid power, Geretsried, Germany, 4 Dec 2025 — a first-of-a-kind that works.	Scaling the Bavarian plant; “geothermal anywhere” thesis.
Sage Geosystems	Geopressured (sedimentary basins)	3 MW Texas plant placed in service Aug 2026 after 120 days of monitored operation.	Up to 150 MW for Meta, site undisclosed, date soft.
XGS Energy	Closed-loop, water-independent	3,000-hour demonstration in an idle California well.	150 MW for Meta in New Mexico, two phases, full operation by 2030.
Quaise	Millimeter-wave drilling for superhot rock	100 meters drilled in granite — research stage.	1 km next; the “superhot” prize is decades out.

The honest read of the table: next-generation geothermal has delivered a few tens of megawatts of real electricity worldwide — Eavor’s German plant, Sage’s 3 MW, Fervo’s Nevada pilot — against a conventional fleet of 16,000 MW and an IEA target of 800,000 MW. Every hyperscaler deal on the list (Google–Fervo, Google–Ormat, Meta–Sage, Meta–XGS) is for power that arrives 2027–2030. The technology has crossed from “does it work” (yes, at pilot scale) to “does it scale on cost” — and that question is being answered one Cape Station at a time.

What actually happened (spring–summer 2026)

The IPO. Fervo priced on May 12 and began trading on Nasdaq as FRVO on May 13 at $27 a share: 70 million shares for $1.89 billion, lifted to about $2.2 billion gross when the underwriters exercised their full 10.5-million-share option. That valued it at ~$7.7 billion at the offer price and over $10 billion at the first close. Its bankers marketed it as “the largest primary clean-energy public equity deal of all time” — a fair claim for that category (X-energy’s nuclear IPO raised $1.02 billion a month earlier). Behind it: 658 MW of binding power-purchase agreements (~$7.2B potential revenue) and a non-binding 3 GW framework with Google that contemplates data-center offtake but obligates Google to nothing.

The blowout. Twelve days later, on May 25, a well at Cape Station in Milford, Utah suffered a blowout: pressurized geothermal fluid breached the surface during work on the well (Energy Intelligence and Axios reported drilling rods breaking apart as they were removed; Utah regulators confirmed the incident). Fervo and Utah state officials reported no injuries, non-hazardous stimulation water, no environmental damage, and no impact to the project or schedule; cleanup followed standard incident protocol. Energy analysts noted geothermal blowouts are “less catastrophic” than oil-and-gas ones because there are no hydrocarbons and far lower toxic-gas concentrations. The drilling industry’s own body (IADC) published geothermal well-control guidelines the same month — already in the works, and the tell that the industry treats this as a class of risk to be engineered, not a one-off.

The build. Cape Station’s first 500 MW is under construction with first power targeted for late 2026 and roughly 100 MW operating by early 2027; phase two targets 2028. Meta’s agreement for up to 150 MW with Sage Geosystems (a different technique — geopressured geothermal) was announced in August 2024 with a 2027 first-phase aim at a still-undisclosed site east of the Rockies; that date is soft. What is real: Sage’s first commercial plant, a 3 MW unit in Christine, Texas, sold its first electricity in Q2 2026 and was placed in service in August after 120 days of monitored operation. (Meta has a separate New Mexico geothermal project with XGS Energy.)

Targets are targets. “First power late 2026,” “100 MW by early 2027,” and the DOE’s 90 GW are company and agency goals, not delivered megawatts. Fervo’s track record on drilling cost is real and measured; its track record on operating a 500 MW plant does not exist yet. That is what the IPO priced — and what the blowout reminded everyone of.

The sleeper: heat batteries

Beside the geothermal headlines, a quieter technology is scaling on plain physics: store heat, not electricity. Heat a pile of sand, crushed stone or carbon blocks with cheap surplus power, insulate it, draw the heat out when needed. In Pornainen, Finland, Polar Night Energy’s 1 MW / 100 MWh sand battery — 13 meters tall, 2,000 tons of crushed soapstone — now runs the town’s district heating, cutting its heating emissions about 70% and retiring oil from the network entirely; it survived its first full winter. In California, Rondo Energy switched on a 100 MWh industrial heat battery in 2025, the largest of its kind, and is opening a large factory. And in July 2026, Antora Energy raised $550 million to mass-produce carbon-block heat batteries explicitly pitched at the AI data-center power crunch.

Why it matters: industrial heat is a quarter of global energy use and most of it still comes from burning things. A heat battery charged on cheap midday solar and discharged as 24/7 process heat is close to the cheapest storage that exists per kilowatt-hour — because sand is cheap and the physics is centuries old.

The honest catch: heat batteries store heat. Heat-to-heat, they are superb (Rondo reports over 97% round-trip). Turning stored heat back into electricity is another matter: binary geothermal-style turbines convert only a fraction of heat to power — often well under a third at these temperatures — so for a data center that needs electrons, the fit is debated — one trade outlet’s headline literally asked “do heat batteries make sense for data centers?” They shine where the end use is heat: factories, district heating, steam. Treat “heat batteries will power AI” as a claim to watch, and “heat batteries will decarbonize industrial heat” as the one already happening.

Who gets paid if this works

The layer	Who	Why they win either way
The drillers	Oilfield-services firms and rig contractors	Next-gen geothermal is oil-and-gas drilling aimed at a different target. The same rigs, bits and crews get paid whether Fervo or a rival wins — the picks-and-shovels position.
The developers with contracts	Fervo (658 MW binding PPAs), Sage (Meta 150 MW)	Signed offtake is the asset. Non-binding frameworks (Google’s 3 GW) are options, not revenue.
The power-block makers	Turbine, heat-exchanger and organic-Rankine-cycle equipment suppliers	Every geothermal megawatt needs a turbine and heat exchangers regardless of whose well it is — and these makers also serve nuclear and gas.
Industrial heat storage	Polar Night Energy, Rondo, Antora	Cheapest storage per kWh where the end use is heat; the proven, unglamorous layer. The data-center pitch is the speculative extension.

One layer deeper: the backlog is paper, the plant is the test

The IPO headline is $7.2 billion of “potential revenue backlog.” The number that matters is smaller and slower: Fervo guides 2027 revenue of $60–80 million (it flagged transmission-related curtailment as a risk), against second-half-2026 capital spending of $850–900 million and a Q2 net loss of $55.9 million on $113,000 of revenue. Cape Station Phase I is three 33 MW GeoBlocks — about 100 MW — and no 500 MW enhanced-geothermal plant has ever run. Phase II (400 MW, 2028) targets an all-in cost of $5,500 per kW on the way to a long-run $3,000/kW; that cost-down is the underwriting, not the drilling record. And the scarce input is shifting from hot rock to long-lead surface equipment and grid connection: Baker Hughes is already contracted for five 60 MW organic-Rankine-cycle units for Phase II. Twenty-year contracts, a learning curve that has to keep compounding after the easy wells, and a first plant that hasn’t delivered its first megawatt — that stack, not the IPO, is what a reader should underwrite.

Claims we have seen but could not confirm (as of 27 Aug 2026)

Our rule is that nothing goes in the article body without a primary source. These circulated during our review and are plausible, but we could not trace them to one. We list them here so you know they exist, dated — and we will move each into the body (or strike it) as evidence arrives, with a timestamp.

Blowout mechanism (unconfirmed): that the May 25 event occurred during a workover on well 6-IB — newly created fractures connecting to a pressurized adjacent well and overwhelming the wellhead safety equipment — per a state filing. Published reports we could verify say only that rods broke during removal and pressurized fluid surged. If the regulator’s account surfaces, it is the more important version, because inter-well communication is a design question, not a one-off accident.
Phase I all-in cost of ~$7,000/kW (unconfirmed): consistent with the disclosed Phase II target of $5,500/kW and long-run $3,000/kW, but we found no filing stating the Phase I figure.
Heat-battery-to-electricity efficiency of 20–40% (unconfirmed range): the literature we checked puts standard Carnot-battery round trips below ~70%, with optimized designs at 76–83%; binary geothermal turbines convert roughly 10–27% of heat depending on temperature. We used “well under half” in the body as the defensible statement.

The risks — named, not buried

Drilling risk is real and now demonstrated. Well control, casing failures, blowouts. Less dangerous than oil-and-gas wells, but not nothing — and each incident costs schedule credibility.
Induced seismicity. Fracturing hot rock and injecting water can trigger small earthquakes; a Swiss EGS project was shut down in 2006 after tremors in Basel. Modern projects monitor and manage this, and it remains the technology’s reputational tail risk.
Water and geology. EGS needs water and the right rock. The U.S. West has both; much of the world does not. “Anywhere” geothermal is still a research goal.
The cost curve has to keep falling. Fervo’s 70% drilling-time improvement is the whole thesis; the DOE’s 90% cost target by 2035 assumes it continues. If learning stalls, geothermal stays a niche.
IPO valuation vs. delivered megawatts. A ~$8–10 billion valuation on a company guiding $60–80 million of 2027 revenue, whose flagship plant has not yet delivered first power, is a bet on execution. The backlog is real; the plant is not, yet. If Phase I slips past early 2027 or costs stall above $5,500/kW, the market was pricing a learning curve that did not compound.
Heat batteries: the electricity conversion. Heat-to-heat is ~97% efficient; heat-to-power at these temperatures returns well under half. That is physics, not engineering. For data centers, watch for actual deployments, not fundraising.
Not advice. This is how we read the chain, not a recommendation. Company names illustrate a structural shift, not buy calls.

The viral take and the true take are rarely the same trade

The IPO is the story. The drill bit is the trade.

Dragonfly Lens maps the AI buildout as one connected chain — and keeps finding the layer where the value actually pools. Plain English, every claim sourced and flagged, promises and setbacks in the same piece. When we're wrong, we say so.

Join the Lens →

Not ready to subscribe? Get the free weekly →

More: Big Tech goes nuclear · The green data center · All explainers

Quick answers

How big is geothermal today? About 16 GW worldwide (IRENA, end-2025); the U.S. leads with 4 GW, roughly 1% of its electricity. The IEA sees next-generation geothermal reaching 800 GW by 2050 if costs fall as hoped — fifty times today’s fleet.

What is enhanced geothermal? Drilling into hot, dry rock, fracturing it, and circulating water through the fractures to bring heat to the surface — so geothermal works far beyond natural hot springs. Fervo cut drilling time 70% by using oil-and-gas techniques.

What happened at Fervo’s Cape Station? On May 25, 2026, drill rods broke during removal and pressurized fluid surged to the surface. Fervo and Utah officials reported no injuries, non-toxic stimulation water, no environmental damage and no schedule impact. It came twelve days after Fervo’s $1.9B IPO (~$2.2B with the over-allotment).

What is a sand battery? A large insulated mass of sand or crushed stone heated with cheap electricity and discharged as heat later. Finland’s 1 MW / 100 MWh unit runs a town’s district heating. Heat batteries are cheapest where the end use is heat (over 97% efficient heat-to-heat); converting back to electricity returns well under half.

Who benefits from a geothermal buildout? The drillers and oilfield-services firms (same rigs, new target), developers with signed power contracts, turbine and heat-exchanger makers, and — for industrial heat — the heat-battery makers.

Sources: Global geothermal ~16 GW end-2025 (15.4 GW 2024); U.S. 4 GW / 24% / ~1% of U.S. electricity; 53 of 99 plants in California — IRENA 2026, NREL 2025 U.S. Geothermal Market Report. IEA: next-gen 120 GW by 2035, 800+ GW by 2050 (~8% of electricity), costs −80% to ~$50/MWh, $1T cumulative by 2035 — IEA, The Future of Geothermal Energy. Ormat 932 MW; Google/NV Energy up to 150 MW (Feb 2026, 2028–30) — Ormat. Eavor closed-loop grid power, Geretsried, 4 Dec 2025 — ThinkGeoEnergy. XGS–Meta 150 MW New Mexico by 2030; 3,000-hour demo — POWER. Quaise 100 m milestone — ThinkGeoEnergy. Fervo IPO May 13 2026, $27/sh, 70M shares $1.89B + 10.5M over-allotment = ~$2.2B gross, ~$7.7B at offer / >$10B first close; “largest primary clean energy public equity deal” (bankers); X-energy $1.02B Apr 24 2026; Cape Station 500 MW, first power late 2026, ~100 MW early 2027, phase II 2028; 658 MW binding PPAs (~$7.2B), Google 3 GW non-binding framework — Fortune, Canary Media, Utah Money Watch. Blowout May 25 2026: rods broke during removal, no injuries, non-toxic water, no schedule impact; IADC well-control guidelines — Axios, Energy Intelligence, JPT / IADC. 70% drilling-time cut, $9.4M → $4.8M per well (first Cape wells, 2024) — POWER Magazine; Sawtooth 7: 19,448 ft, 460°F, 7,500 ft lateral, 21 days (Jul 2026) — Fervo. Q2 2026: 2027 revenue guide $60–80M, H2-2026 capex $850–900M, net loss $55.9M, Phase I = 3×33 MW GeoBlocks, Phase II $5,500/kW → $3,000/kW — Fervo Q2 2026 results. Baker Hughes: five 60 MW ORC units for Phase II — Baker Hughes. Fervo IPO over-allotment ($2.2B) — Axios, TechCrunch (X-energy). Sage 3 MW Texas plant placed in service Aug 2026; Meta “up to 150 MW” — ThinkGeoEnergy, Canary Media. Rondo >97% round-trip (heat) — Rondo. DOE Enhanced Geothermal Shot: 5 GW by 2030, 90 GW by 2050, 90% cost cut by 2035 — US DOE. Meta / Sage Geosystems 150 MW by 2027, first next-gen geothermal east of the Rockies — Utility Dive, Meta. Polar Night Energy 1 MW / 100 MWh sand battery, 2,000 t soapstone, ~70% heating-emissions cut — pv magazine, CNBC. Rondo 100 MWh heat battery (2025) — MIT Technology Review, Energy-Storage.News. Antora $550M (Jul 2026) for data-center heat batteries; “do heat batteries make sense for data centers?” — Canary Media, Latitude Media.

Educational research, not personalized investment advice. Dragonfly Lens is not a registered investment advisor. Figures are as reported by the sources above at the dates given; project timelines are company and agency targets and carry execution risk. Company names illustrate a structural shift, not buy recommendations.