Enhanced geothermal: what Project Red actually proved

Definition

Enhanced geothermal system (EGS): a power plant that circulates fluid through hot rock that lacks the natural water or permeability of conventional geothermal fields. Wells are drilled horizontally and stimulated with the shale industry's plug-and-perf method, creating an engineered reservoir between an injection well and a production well. Because the heat source is rock, not water, EGS is location-flexible — and it is the leading candidate for clean-firm power: generation that runs around the clock regardless of weather.

The pilot record

Fervo Energy's Project Red near Blue Mountain, Nevada, is the reference pilot. A 30-day well test in spring 2023 flowed 63 L/s at up to 336°F from 191°C rock, supporting 3.5 MW gross — the most productive EGS well pair ever tested. The plant grid-connected on November 1, 2023, selling power to Google.

The honest steady-state picture, per Fervo's own two-year reporting, is smaller: average gross output of 2.1 MW, roughly 1.4 MW net after downhole pumping loads, at 347°F and ~36 kg/s circulation, across 614 days with zero workovers (Fervo Energy). Temperatures held stable for over 500 days, then slipped about 2.5°F — expected for a deliberately tight reservoir. An independent critique notes the caveat: if that decline continued it would imply ~9°F per year, versus ~1.5°F per year typical for Nevada wells, and ~70% water recapture means a 100 MW project would need on the order of 30,000–40,000 gpm of makeup water (GeoExpro). Both observations are fair. Neither negates the core result: the reservoir has run continuously for two years.

From pilot to power plant

Cape Station in Utah is the commercial test. The first 33 MW block synchronized in September 2026 — the world's first utility-scale EGS — with roughly 100 MW due in early 2027, 500 MW total by 2028, and ~900 MW contracted, including a 396 MW corporate power purchase agreement. The drilling learning curve is the strongest evidence yet: the fastest well reached target depth in 21 days versus 70 for the first Project Red well, with per-well drilling cost falling from $9.4M to $4.8M and a realized learning rate near 35% (Fervo Energy).

Project	Capacity	Indicative cost	Status
Project Red, NV	3.5 MW gross peak; ~2.1 MW avg	Pilot	Grid-connected Nov 2023
Cape Station Phase 1, UT	~100 MW	~$7,000/kW	First 33 MW online 2026
Cape Station full build	500 MW by 2028	$5,500/kW; $3,000/kW goal	Under construction
DOE 2035 target	—	~$45/MWh (~$3,700/kW)	Goal

For context, an average new gas plant runs near $2,157/kW — so the gap is real but closing fast.

Bottleneck one: hotter rock, tougher tools

Raising megawatts per well means drilling hotter. Fervo's design trajectory moves from 350°F toward 460°F, but conventional oilfield downhole electronics are typically rated 150–175°C. Hotter wells mean tool failures and non-productive cooling time. The intervention exists: DOE has funded 300°C-rated systems, and high-temperature electronics are an engineering problem with a known price, not a physics wall.

Bottleneck two: reservoir life

Time to thermal breakthrough scales with the square of well spacing. A 30-year commercial project needs roughly 278 m between wells; Project Red used ~111 m. Whether stimulation can reliably connect wells at commercial spacing remained an open question at the 2025 Stanford Geothermal Workshop. Induced seismicity, by contrast, has been managed well: all monitored events have stayed below magnitude 2 under the DOE protocol.

What the numbers support

DOE's Liftoff analysis puts the prize at 90 GW of clean-firm capacity by 2050 — up to 300+ GW — conditional on a demonstrator scale of 2–5 GW across 4–6 states costing $20–25B (DOE). Cape Station alone is roughly a tenth of that liftoff threshold.

My assessment, as an opinion: the pilot claims are quantified and largely self-critical, which raises my confidence. What remains genuinely unproven is 30-year reservoir life at commercial spacing. Watch temperature-decline curves from Cape Station's first years — they will answer the question pilot data cannot.