Fervo Energy: Big Tech’s Clean Energy Bet Is Unraveling

E² review · 2026-10-06

Evidence audit

Regulatory balances and catalogue counts reproduced · commercial models unvalidated

The report documents material operating and delivery risks. Its calculations support several narrow findings. They do not establish deliberate concealment, seismic damage or inevitable commercial failure.

Findings

Project Red water balance

192.87 million gallons of cumulative net imbalance. June 2026: 34.68%; July: 22.68%. C17The June 2026 pilot injection–production imbalance was approximately 35%.Type: Calculated outputSupport: Not applicableReview: Replication completed · C18The cumulative pilot injection–production imbalance is about 193 million gallons.Type: Calculated outputSupport: Not applicableReview: Replication completed

Temperature interval

344°F to 332°F across six calendar months annualizes to 24°F/year. Its long-term interpretation remains open. C10November 2025 to May 2026 temperatures imply a 24°F annualized decline.Type: Calculated outputSupport: Not applicableReview: Replication completed

Seismic records

58 company-catalogue yellow events reproduce. The USGS prose window yields 37; the linked window yields 39. C38The published Cape catalogue contains 58 yellow-level events through April 2026.Type: Calculated outputSupport: Not applicableReview: Replication completed · C41The August 26–September 25 USGS window contains 39 magnitude-2-or-higher events.Type: Calculated outputSupport: Not applicableReview: Unresolved conflict

Claims requiring correction

The ten-month water maximum, disclosure chronology, gross/net power comparison and seismic selection boundaries need correction or reconciliation. C14Net pilot output below 50% of its stated capacity predicts difficulty meeting a roughly 70% PPA threshold.Type: InferenceSupport: ConjecturalReview: Unresolved interpretation · C20June 2026 water imbalance reached a ten-month high.Type: Reported measurementSupport: ConjecturalReview: Contradicted as labeled · C39March and April account for 29 yellow-level events, exceeding the earlier period.Type: Calculated comparisonSupport: Not applicableReview: Unresolved conflict · C40The blog acknowledging 36 events was published weeks after Latimer’s July interview.Type: Reported factSupport: ConjecturalReview: Contradicted as labeled · C41The August 26–September 25 USGS window contains 39 magnitude-2-or-higher events.Type: Calculated outputSupport: Not applicableReview: Unresolved conflict

Runnable audit · Frozen inputs · Results · Plot source · Source record

Temperature

Source mapping

Nevada’s workbook identifies injection permit 1505, API 27-013-90146, and production permit 1506, API 27-013-90147. It reproduces the report’s 90°F changes in March 2024 and June 2026. These are plant-operator submissions hosted by a regulator. The report’s page-19 caption calls permit 1506 “34A-22”; the inventory identifies 34A as injection permit 1505. Permit/API mapping is more reliable than the inconsistent names. C04Fervo’s 2024 temperature chart conflicts with monthly Nevada submissions for the same production well.Type: InferenceSupport: ReportedReview: Unresolved interpretation · C08The reported monthly production temperature fell 90°F in March 2024.Type: Reported measurementSupport: SupportedReview: Source checked

The 2024 paper plots a steadier operating series. A monthly average can differ from a selected flowing-temperature series because of startup, shutdown, measurement location or weighting. Neither side supplies the full bridge here. The difference warrants disclosure; it does not establish falsification or intent. C05Fervo deliberately altered temperature data to hide underperformance.Type: InferenceSupport: ConjecturalReview: Not established

Decline and calibration

d = \frac{344-332}{6}\times12 = 24\;{}^\circ\mathrm{F}/\mathrm{year}

Linear regression across the seven November–May entries gives 24.43°F/year. This is a description of one interval, with no inference of independent monthly errors. June falls to 242°F; July rebounds to 316°F. The cause and the appropriate operating-state exclusions remain unresolved. A forecast must specify those exclusions before looking at the outcomes. C10November 2025 to May 2026 temperatures imply a 24°F annualized decline.Type: Calculated outputSupport: Not applicableReview: Replication completed

The 16× comparison to 1.5°F/year reproduces arithmetically. An economic threshold needs resource temperature, power conversion, flow, useful life and costs. Fervo’s 2026 model describes a tighter, shorter pilot designed to reveal decline sooner. Its fit to observed data is not evidence of prospective prediction without a frozen earlier model and held-out period. C11The annualized decline is about 16× a 1.5°F/year viability benchmark.Type: Calculated comparisonSupport: Not applicableReview: Arithmetic reproduced · C12A decline of about 1.5°F/year is the economic limit for Fervo’s EGS designs.Type: InferenceSupport: ReportedReview: Not validated

Water

Monthly balance

L = \frac{I-P}{I}\times100\%,\quad I=\text{injected volume},\ P=\text{produced volume}

March’s monthly injection increases 33.42%. The workbook’s reported daily injection increases 29.11%. A monthly-volume comparison must not silently become a flow-rate comparison. February’s reporting date is February 28 in a leap year; the raw daily and monthly fields are retained. The matched imbalance rises from 21.17% to 56.86%. C06Water injection increased about 33% in March 2024.Type: Calculated outputSupport: Not applicableReview: Replication completed · C07The injection–production imbalance rose from about 21% to 57% in March 2024.Type: Calculated outputSupport: Not applicableReview: Replication completed

The signed cumulative imbalance is 192,872,191 gallons. Positive deficits alone total 206,737,068; ignoring negative months changes the question. June’s 34.68% is below September 2025’s 37.82%, so the “ten-month high” does not reproduce. July recovers to 22.68%. C17The June 2026 pilot injection–production imbalance was approximately 35%.Type: Calculated outputSupport: Not applicableReview: Replication completed · C18The cumulative pilot injection–production imbalance is about 193 million gallons.Type: Calculated outputSupport: Not applicableReview: Replication completed · C20June 2026 water imbalance reached a ten-month high.Type: Reported measurementSupport: ConjecturalReview: Contradicted as labeled

Retention and water rights

Fervo acknowledges approximately 70% pilot recapture. ResFrac’s multi-well model explains how outer producers and pumping could reduce imbalance; it supplies a testable mechanism. It has not validated Cape Station’s claimed commercial recovery here. “Not yet demonstrated” and “cannot work” require different evidence. C15Project Red recaptures about 70% of injected water under steady-state conditions.Type: Reported measurementSupport: SupportedReview: Source checked · C16Commercial multi-well configurations can approach 99% water recapture.Type: Model assumptionSupport: Not applicableReview: Source checked · C19Persistently high imbalance proves reservoir pressurization has not worked.Type: InferenceSupport: ReportedReview: Unresolved interpretation · C21Former employees report worse-than-expected water loss and weak Cape Station test results.Type: Alleged leakSupport: ReportedReview: Not independently verified

Water injected into rock but not immediately recovered may be stored, migrate elsewhere or interact with surrounding fluids. Its eventual location and timing require a hydrogeologic balance. Legal non-consumptive use additionally depends on the approved water-right conditions. General Utah policy does not authenticate the specific grant. The report’s application links were inaccessible, so its exact conditions remain unverified. C24Cape Station’s water access depends on a non-consumptive designation.Type: Reported factSupport: ReportedReview: Not independently verified · C25Leak-off cannot recharge the source aquifer on a commercially useful timescale.Type: InferenceSupport: ReportedReview: Not validated

Power and capacity

The 2023 paper’s 5.1 MW estimate assumes 35% thermal recovery, 15% conversion efficiency and ten years of useful life. The later 2.1 MW gross average is 41.18% of that modeled value. A capacity factor requires calendar energy divided by comparable nameplate capacity and calendar hours. The pilot’s average, selected on-production period and different gross/net figures cannot substitute for that calculation. C09Latimer’s statement that the pilot met all technical objectives was false.Type: InferenceSupport: ReportedReview: Unresolved interpretation · C13Average gross pilot output is about 41% of the 2023 modeled 5.1 MW figure.Type: Calculated comparisonSupport: Not applicableReview: Arithmetic reproduced · C14Net pilot output below 50% of its stated capacity predicts difficulty meeting a roughly 70% PPA threshold.Type: InferenceSupport: ConjecturalReview: Unresolved interpretation

The 2024 Frisco paper estimates electric output from thermal flow and anticipated plant performance; it publishes initial power data from a thirty-day test. This is not thirty days of measured net generation. Meanwhile, Fervo’s October 1 announcement reports 33 MW net and September 30 commercial operation. That is a company attribution, with no sustained metered series authenticated here. Date the report’s “zero megawatts” interview before treating it as current evidence. C22Fervo’s Frisco presentation publishes only initial power data from a thirty-day test.Type: Reported factSupport: SupportedReview: Source checked · C23A Bearskin test was completed by October 2025 and its results were withheld.Type: InferenceSupport: ReportedReview: Unresolved interpretation · C46Fervo had “zero megawatts in production,” as an employee quoted in the report says.Type: Alleged leakSupport: ReportedReview: Unresolved interpretation

Costs and contracts

Lazard’s 2026 analysis identifies $107/MWh as an illustrative new-geothermal point using $7,000/kW installed capex and $160/kW-year fixed O&M. Matching two inputs does not produce Fervo’s project-specific cost. The source also supplies a tax-credit sensitivity. Financing, conversion, capacity factor, price escalation and credit monetization require a common cash-flow model. Levelized total cost is different from marginal operating cost. The $99 price remains an analyst attribution. C26A $99/MWh PPA against $107/MWh levelized cost means loss on every MWh sold.Type: InferenceSupport: ConjecturalReview: Unresolved interpretation

The long-term $3,000/kW target requires a 57.14% reduction from $7,000/kW. The Phase II $5,500/kW target requires 21.43%. These are conditional targets. Conference estimates, actual installed cost and changed designs need a bridge before an upward estimate proves a worsening learning curve. A change in chart presentation does not settle concealment. Anonymous completion costs remain unverified. C27Reported installed capex rose from $5,000/kW to $7,000/kW.Type: Calculated comparisonSupport: Not applicableReview: Not reproduced · C28Reducing installed capex from $7,000/kW to $3,000/kW requires a 57% reduction.Type: Calculated outputSupport: Not applicableReview: Arithmetic reproduced · C29The changed drilling-cost chart conceals rising costs.Type: InferenceSupport: ReportedReview: Not established · C30Completions cost roughly $3,000/kW and are a hidden obstacle to the target.Type: Alleged leakSupport: ReportedReview: Not independently verified

The screenshot’s arithmetic checks out. The full contract link returned 403. Actual default requires the executed measurement period, excused-energy provisions, deadlines and cure rights. The report’s footnote acknowledges excused product and a two-year period. Its prospectus supports material cancellation risk, not an already-triggered default. C31Late-operation damages can exceed half a million dollars per week.Type: Calculated outputSupport: Not applicableReview: Arithmetic reproduced · C32872,140 MWh/year from 115 MW implies an 86.6% capacity factor.Type: Calculated outputSupport: Not applicableReview: Arithmetic reproduced · C3380% of the stated supply amount corresponds to roughly 69.3% of nameplate annual energy.Type: Calculated outputSupport: Not applicableReview: Arithmetic reproduced · C34Failure to commence operation within six months can permit PPA termination.Type: Reported factSupport: SupportedReview: Source checked

Transmission

The prospectus discloses 290 MW of rights, enough for original SCE/CPA commitments but insufficient for the expanded 384 MW. It explicitly warns of additional-capacity risk. Fervo’s August update confirms a Utah delivery point retiring in 2027. Its CAISO comment conditions a pre-2032 pathway on timely new planning. These substantiate delivery risk. Rights held, usable delivery capacity, plant completion and amended customer dates still need separate schedules. C35The prospectus discloses about 290 MW of transmission rights, with additional capacity still needed.Type: Reported factSupport: SupportedReview: Source checked · C36California delivery faces a retiring Utah path and contingent new transmission.Type: Reported factSupport: SupportedReview: Source checked · C37Guidance for California power delivery has moved to “2030+”.Type: Reported factSupport: ReportedReview: Not independently verified

Seismicity

Catalogue comparisons

The company series reproduces 58 yellow events and the published six-month count of 36. Exactly one April event has magnitude 2.00. Excluding it gives 29 March–April events but 57 total. This boundary must be consistent. The June 29 blog predates the interview described as July; “weeks after” needs correction or an archived disclosure history. C38The published Cape catalogue contains 58 yellow-level events through April 2026.Type: Calculated outputSupport: Not applicableReview: Replication completed · C39March and April account for 29 yellow-level events, exceeding the earlier period.Type: Calculated comparisonSupport: Not applicableReview: Unresolved conflict · C40The blog acknowledging 36 events was published weeks after Latimer’s July interview.Type: Reported factSupport: ConjecturalReview: Contradicted as labeled

The report’s linked USGS search begins August 25, one day before its prose window. Those two additional events explain 39 versus 37. The linked window contains 38 events below magnitude 3 and 1 above it, so “39 yellow” also needs qualification. A regional box is not a list of events proven to be caused by Fervo. Network sensitivity, magnitude methods, observation windows and injection exposure differ between catalogues. Comparing their raw counts does not calibrate a future earthquake probability. C41The August 26–September 25 USGS window contains 39 magnitude-2-or-higher events.Type: Calculated outputSupport: Not applicableReview: Unresolved conflict

Attribution and damage

USGS event uu80156586 records ML 3.13, 2026-09-12T22:52:29.720Z, and depth 3.65 km. It supports the reported nearby event. A traffic-light threshold supports a conditional pause requirement, not proof of an actual shutdown. No shutdown or casing-failure log was obtained. C42USGS records a magnitude-3.1 event near Cape Station on September 12, 2026.Type: Reported measurementSupport: SupportedReview: Source checked · C43The September event likely triggered at least a 24-hour operational pause.Type: InferenceSupport: ReportedReview: Not independently verified · C44Seismic events have damaged Fervo well casing and increased leak-off.Type: Alleged leakSupport: ReportedReview: Not independently verified

Higher activity is a legitimate monitoring concern. Frequency, magnitude, ground motion, fault interaction and casing damage are separate variables. Matching failure times and locations to events, and testing pressure, construction and corrosion alternatives, is necessary before attributing damage. The company’s low reported surface ground motion does not independently rule out subsurface casing damage.

Revision tests

1. Reconcile daily and monthly temperatures, meter locations, operating-state filters, zero-volume periods and the anomalous July 2024 volumes.
2. Freeze reservoir forecasts before new outcomes. Compare matched net generation, temperatures, flow, pressure and water balance on held-out operating intervals.
3. Publish full Cape test records, failure/repair logs and dated seismic response records. Keep source identities authenticated without exposing private witnesses.
4. Release approved water-right conditions and the field-scale recovery model. Define timescales and failure thresholds for the pressurization and recharge hypotheses.
5. Reconcile all-in installed costs, tax credits, PPA terms, liquidity and delivery pathways. Test lower output, higher water use, delay and replacement-transmission cases.
6. Set an explicit horizon and observable commercial-failure criteria. Record what would weaken the report’s thesis as well as what would strengthen it.

The present evidence substantiates operational imbalances and transmission exposure. Commercial longevity, profitability, intentional concealment and claimed seismic damage remain unresolved. C45Fervo’s commercial buildout will fail and consume its remaining capital.Type: ForecastSupport: ConjecturalReview: Not yet resolvable