Industries / Energy, water and waste / Power generation / Geothermal
Drilling
Drilling makes the wells that bring hot water or steam up and send cooled water back down, through rock that is often harder, hotter and more broken than the rock oil and gas wells pass through.
6
Milestones achieved, from 1922 to 2023.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1922
Steam from a shallow well
At The Geysers in northern California, an area of steam vents, John Grant's first well in 1921 blew out. His second, in 1922, only 200 feet (61 m) deep, was the first successful geothermal well there, and in 1923 it ran a 35-kilowatt plant at the Geysers resort. The American Society of Mechanical Engineers says pipes and turbines of the day could not stand the grit and corrosion in the steam, and the project was abandoned; Magma Power drilled the first modern well at The Geysers in 1955. Sources differ on his middle initial: Calpine gives John C. Grant, the ASME John D. Grant.
Lessonreaching the steam was not enough; the materials that carry it had to last.
Credit
John Grant and family (The Geysers Development Company)
1983
Tools that survive heat
Logging tools are instruments lowered into a well to measure it, and most electronics fail well below geothermal temperatures. Sandia National Laboratories built a prototype acoustic borehole televiewer around 1983, a tool that maps fractures on the well wall by timing sound echoes, with military-grade electronics inside a vacuum flask so it could work at 275 °C. The maker it partnered with moved on to other products, only two copies were built, and one was lost down a well in Indonesia. In 2003 Sandia backed Mount Sopris and Advanced Logic Technology to build a new televiewer for the Coso field rated for ten hours at 275 °C.
Lessongeothermal is a small market for tool makers, so hot-well tools have depended on public labs to exist at all.
Credit
Sandia National Laboratories; Mount Sopris; Advanced Logic Technology
1984
Diamond bits fall short
Sandia began work in 1975 on drag bits with polycrystalline diamond compact (PDC) cutters, which scrape rock instead of crushing it, to replace the roller bits that wore out fast in hot, hard rock. By 1984 Sandia's David Glowka reported mixed results: one bit drilled up to 170 feet per hour at Valles Caldera, New Mexico, but tests at The Geysers failed when braze joints let go, and cutters wore faster as they heated. He concluded a bit must also outlast roller bits to cut costs.
Lessona fast bit saves money only if it stays in the hole long enough.
Credit
Sandia National Laboratories (David A. Glowka)
2009
A well hits magma
The Iceland Deep Drilling Project (IDDP) set out to reach supercritical water, so hot and pressurized that liquid and steam become one fluid. Its first well, IDDP-1 in the Krafla caldera, ran into 900 °C rhyolite magma at only 2,100 m and had to stop. It still produced superheated steam at 452 °C, enough flow for about 35 MWe, but after two years of tests its master valves failed, and cooling the well with water to repair them collapsed its casing.
Lessona well can find more heat than its steel, cement and valves can handle.
Credit
Iceland Deep Drilling Project consortium
2017
Supercritical conditions reached
The second IDDP well deepened an existing production well at Reykjanes and reached 4,659 m along its slanted path, about 4.5 km straight down, in January 2017. After six days of heating the bottom read 426 °C at 34 megapascals, above the critical point of seawater, which feeds this field. Below 2.5 km all drilling fluid was lost into the rock and could not be stopped with plugging materials or cement, so drilling went on without returns. The authors call it "the first well that successfully encountered supercritical hydrothermal conditions."
Credit
HS Orka and the IDDP consortium (Guðmundur Ó. Friðleifsson, Wilfred Elders and others), with EU Horizon 2020 funding through DEEPEGS
2023
Hard rock, drilled faster
At Utah FORGE, a US Department of Energy test site, drilling practices adapted from oil and gas cut on-bottom drilling time from 310 hours for a well in 2020 to 110 hours for one in 2023, as reported by the National Laboratory of the Rockies (NLR, formerly NREL). Fervo Energy reported cost per foot falling from about $1,050 at its Project Red wells in Nevada to under about $450 for its first six Cape Station wells in Utah; those are Fervo's own figures, relayed by NLR. Speed did not mean cheaper wells at first: NREL's 2022 review found FORGE's actual well costs at or above the old baseline, and FORGE well 16B(78)-32 used 16 PDC bits that averaged 6.5 hours on bottom each.
Lessonfaster drilling lowers costs only when bits last longer too.
Credit
Utah FORGE (University of Utah, US Department of Energy); Fred Dupriest and Sam Noynaert (Texas A&M); Fervo Energy
Watch for
Deeper well in 21 days
Fervo says it drilled Sawtooth 7, a Cape Station Phase II well in Utah, from spud (the start of drilling) to total depth in 21 days, reaching 19,448 feet (5,928 m) measured depth with a 7,500-foot (2,286 m) horizontal section in rock at 460 °F (238 °C). It says this matches the 70% cut in drilling time it first claimed in February 2024 against its 70-day first well at Project Red, now on a deeper, hotter and longer design.
Achieved whenthe well's spud and total-depth dates and depth are confirmed by a regulator's well file or by an outside party such as a national laboratory.
Claimed by
Fervo Energy, July 2026
Drilling with millimeter waves
Quaise Energy says it drilled 100 m into granite at its Central Texas field site by vaporizing rock with a gyrotron, a tube that beams high-power millimeter waves, with no drill bit in the hole; before 2025 the method had only drilled a few centimeters in an MIT laboratory. In July 2026 it said the system was "approaching nearly 0.6 miles" (about 1 km) of depth.
Achieved whendepth, rate and hole quality are measured by someone outside Quaise, or published with logs in a reviewed paper.
Claimed by
Quaise Energy, July 2025 and July 2026
Open
01
Log every hard-rock bit
A public table of every drill-bit run in geothermal wells whose records are public, one row per run: bit type and size, rock, depth in and out, footage, hours on bottom, rate of penetration, condition when pulled and why it was pulled. Add a simple model of cost per foot against bit life.
The briefCredit · unclaimed02
Where the rig days go
A public ledger that splits the rig time of geothermal wells from several fields and operators into drilling, tripping (pulling pipe out and running it back), casing and cementing, lost circulation (drilling fluid escaping into cracks in the rock) and waiting, with every hour linked to the line of the daily drilling report it came from; then a fresh measure of lost circulation's share of well time and cost beyond a single test site.
The briefCredit · unclaimed
Open 01 · the brief
Log every hard-rock bit
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public table of every drill-bit run in geothermal wells whose records are public, one row per run: bit type and size, rock, depth in and out, footage, hours on bottom, rate of penetration, condition when pulled and why it was pulled. Add a simple model of cost per foot against bit life.
Why it is still open
NREL's 2025 drilling cost update says "There is a paucity of public information about the actual costs of PDC bits," built its new cost curves on averages from three projects, and assumed a PDC bit costs twice a roller bit from limited data at one site. It found average bit life of 6.5 hours on bottom at FORGE well 16B(78)-32 and 34 hours at Cape Station, against the 50 hours assumed in 2017. The raw records are public: the Geothermal Data Repository holds drilling data, daily reports and bit-sensor data for the Utah FORGE wells and drilling data from other projects, though The Geysers bit-design dataset is still awaiting release. Single projects publish their own bit records, such as Sandia's 2012 field test of two PDC bits in granite at the Chocolate Mountains in California, but no one has pooled them. GitHub searches in November 2026 found rate-of-penetration models built on FORGE well 58-32 data, not a pooled table of bit runs.
Built with what exists
Geothermal Data Repository datasets (FORGE wells 58-32, 56-32, 78B-32, 16A(78)-32 and 16B(78)-32; the GLADE wells in Colorado; McGinness Hills 27A-10); a low-cost AI model to pull bit records out of PDF reports (see the Note "The jobs agents repeat"); Python and pandas.
First step
Download the FORGE 16B(78)-32 drilling data from the Geothermal Data Repository and tabulate its bit runs.
Done looks like
At least 300 bit runs from at least ten wells, each row linked to its source file; median footage and hours on bottom for PDC and roller bits by rock type; a cost-per-foot curve, set beside NREL's published averages of 70, 102 and 65 feet per hour.
Open 02 · the brief
Where the rig days go
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public ledger that splits the rig time of geothermal wells from several fields and operators into drilling, tripping (pulling pipe out and running it back), casing and cementing, lost circulation (drilling fluid escaping into cracks in the rock) and waiting, with every hour linked to the line of the daily drilling report it came from; then a fresh measure of lost circulation's share of well time and cost beyond a single test site.
Why it is still open
The common figure, that lost circulation costs about 10% of a well in mature fields and over 20% in exploration wells, comes from Sandia's handbook, which cites a 1982 paper. Newer studies used private records: Lawrence Berkeley and Sandia researchers studied 61 wells from Ormat's proprietary files in 2023, and a 2015 NREL and Colorado School of Mines comparison of 21 geothermal and 21 oil and gas wells counted 3,474 hours lost to lost circulation, fluid influx and cement plugs, and said "Geothermal drilling data is scarce." Daily reports for FORGE wells, McGinness Hills 27A-10, the GLADE wells and Lanai in Hawaii are now public on the Geothermal Data Repository. The nearest work covers one site: Al Qais and colleagues (SPE Journal, June 2026) coded more than 550 lost-time events from more than 330 daily drilling reports at seven Utah FORGE wells and found that five causes, led by rig equipment failures, made up more than 60% of the lost time; neither their abstract nor Dang Ton's 2021 Stanford report, which used a language model to read drillers' remarks, mentions releasing the coded records. The oil and gas file's Volve item labels drilling events to train models on sensor data; this one is a time and cost account across operators.
osti.govescholarship.orgenergy.govgdr.openei.orgapi.crossref.orgpurl.stanford.edu
Built with what exists
Daily drilling reports from the Geothermal Data Repository; a language model that tags each report line with an activity and quotes the line (see the Note "Findings that cite notes"); two people checking a sample; Python.
First step
Code the daily reports of McGinness Hills well 27A-10 by hand into the five activity groups to make a checked reference set, and ask the FORGE study's authors whether their coded events can be shared.
Done looks like
At least 15 wells coded, at least eight of them outside Utah FORGE and from at least two operators; agreement between the model and two people on 300 sampled report lines published as Cohen's kappa; lost circulation's share of rig hours for each well, set against the 10% and 20% figures.