Demand is climbing.
Firm power is retiring.

01

Demand

Data centers are the largest driver. The breadth underneath them is what makes the demand durable.

Strip out data centers and roughly 76 GW of U.S. peak load growth still arrives by 2030. That figure is onshored manufacturing, building electrification, EV charging, and industrial electrification, none of which tracks the AI capex cycle.

Coal retirements pull from the other side. Announced retirements across North America exceed 105 GW over the next decade, and what is leaving is dispatchable.

The presumed replacement has its own problem. Natural gas supplied roughly 41% of U.S. utility-scale generation in 2025 and sets the marginal power price across most of the country, so replacing retiring coal with gas is the default plan. But gas is being claimed from three directions at once.

LNG export capacity is rising from roughly 17 Bcf/d to somewhere near 24–29 Bcf/d by 2029 on projects already under construction, sold largely under long-term contracts that financed them. Storage has not kept pace: working capacity stood at 4,683 Bcf in November 2025, up just 26 Bcf on the year, against demand that has grown far faster. And one major interstate pipeline has been completed in the past decade.

The structural point does not depend on a shortfall. A minority of analysts read this as an outright shortfall late this decade; that forecast is not settled and we do not rely on it. The fuel expected to backfill retiring coal is simultaneously being committed to export and to AI load, with a shrinking cushion and no new pipe. If gas sets the marginal price and gas gets tighter, every firm clean alternative re-rates against it.

The tighter constraint is seasonal. NERC's ten-year winter peak growth (+245 GW) now runs ahead of summer (+224 GW). Winter peaks land in the dark hours at either end of the day, when firm capacity counts for most and intermittent output counts for least.

Data centers remain the sharpest edge of it. Data-center electricity demand more than doubles from ~415 TWh in 2024 to ~945 TWh by 2030 in the IEA's base case, with the U.S. taking by far the largest share of the increase. Hold that loosely: the IEA's own 2035 range spans 700–1,700 TWh.

Geothermal is the only clean energy solution that is firm, virtually carbon-free, and sitable near load. The operating fleet runs at empirical capacity factors of 70–90%. That's three to four times solar PV's ~25% and roughly twice onshore wind's ~35%.

Binding offtake scaled from a 3 MW pilot to 658 MW of signed PPAs in about five years, which Fervo values at ~$7.2B of potential revenue backlog (its own wording, not contracted revenue). The headline number is larger — roughly 3.7 GW once Google's 3 GW framework is counted — but that framework is explicitly non-binding, and the gap between the two figures matters.

The cost curve collapsed because the technology is borrowed, not new.

Horizontal drilling, multi-stage fracturing, and fiber-optic reservoir management ported directly into hot rock. That is the shale playbook, and it is why the cost curve moved. Drilling time per well fell ~70%, from 68 days on Fervo's first horizontal EGS well in 2022 to 21 days at Cape Station in 2024. Well costs fell 49% across the first four horizontal wells, from $9.4M to $4.8M.

Next-generation geothermal changes drilling site requirements for economic power generation. A conventional hydrothermal resource needs heat, permeability and fluid together in one reservoir, which makes development a question of finding the right place. NGG engineers whatever nature did not provide: EGS fractures an artificial reservoir into hot rock that has too little permeability, and closed-loop seals a circuit that never exchanges fluid with the formation.

What is left is depth economics. The geothermal gradient runs ~25–30°C/km in stable continental crust and varies widely elsewhere, so the question becomes how deep a well has to go to reach usable heat and whether that depth pays. Across much of North America it does, which is why the addressable resource is continental rather than confined to a handful of volcanic zones. The IEA puts the full technical potential of next-generation geothermal at ~140x global electricity demand, second only to solar PV among renewables.

What is proven is the learning curve, not the endpoint. Build costs per kW remain modelled rather than disclosed. The IEA's medium case puts construction at ~$7,000/kW in 2035; its low case reaches ~$3,000/kW and ~$50/MWh.

03

Capital

The bankability line has been crossed — by exactly one company.

Fervo closed $421M in non-recourse project debt in March 2026 — the sector's first EGS project financing. Fervo IPO'd on Nasdaq (FRVO) in May, raising $1.89B at a ~$7.66B valuation and closing its first session above $10.2B in market value. The shift from venture equity into project finance turns a technology into an infrastructure asset class. No other next-generation geothermal developer has crossed that line yet: Sage, XGS, Eavor, and Mazama still finance on equity, grants, and strategic PPAs.

The equity has not held that debut. FRVO closed at $22.54 on 31 July 2026, 16.5% below the $27 IPO price, for a market value near $6.6B. The project-finance milestone is a fact about the capital structure and it stands. The share price is a separate question, and it is going the other way.

Policy favors geothermal, and the credit phase-down is already scheduled.

The 2025 OBBBA curtailed wind and solar tax credits but preserved 45Y/48E eligibility for geothermal. That is a genuine relative advantage, and it comes with a hard phase-down: full credit for construction starts through 2033, then 75% / 50% / 0% by 2036.

BLM's IM2026-004 now directs annual competitive lease sales "where appropriate". It also removes required public input absent an EIS, which trades durability for speed.

The lease-price gap is a first-mover cost advantage, not a market re-rating. BLM sales in premium basins now average $111–117/acre, Utah and California state records, while Fervo assembled 595,900 acres at ~$4/acre before competition arrived. Utah's per-acre record was already $170 back in 2007, so the market was never at $4/acre. Peak parcels now reach $344–412.

Geothermal also draws backing from both the climate coalition and the energy-dominance coalition, which lowers the reversal risk that dogs single-coalition technologies.

July gave that claim a test. The House passed H.R. 8790, the Next-Generation Geothermal Research and Development Act, on 20 July 2026 by voice vote under suspension of the rules. Suspension requires a two-thirds majority, so the breadth of support shows in the procedure rather than the sponsor list, which runs to one Republican and one Democrat.

Read it as confirmation that the dual-coalition argument holds, not as a new incentive. The bill directs Department of Energy research and development, it attaches no money we can confirm, and it is not law. It went to the Senate Energy and Natural Resources Committee on 21 July, which is where most bills stop.