Geothermal Is in "Heat", and this time, buyer is moving in

the new buyer has the deepest pockets

My love story with geothermal started in university. It fascinated me with its 24/7 production, closer than the sun, literally right under our footsteps. And I was lucky that my professor, Fusun Tut Haklidir, came from the corporate world. So the class was never just about engineering but also included feasibility, profits, and the corporate reality. Which hooked the entrepreneur in me at first sight.

That is why I wrote my thesis on Geothermal energy in Izmir, my hometown, with an investment analysis rather than just a literature review. I flew all the way to the Stanford Geothermal Workshop just to listen. Later, I tracked down Gad Shoshan, Ormat’s director for Turkey, because I refused to model the field on assumed numbers. Ormat had the real production data even the local government did not have. To my surprise, all I did was ask, and the rest was an instagram post I guess, with Mr.Shoshan, me holding the proprietary data and a permission to publish for the first time.

Following a master’s at Columbia, energy audits in Los Angeles, factory efficiency analyses in Istanbul, and then I drifted from soul-crushingly slow energy into tech, my next love. Chaotic and dynamic, then founded companies of my own.

the villain was never the geology. it was the geography.

Geothermal energy has always been chained to whatever carried value away from the field. First the railroads, then the wires. More than a century ago, when the first aluminum smelters followed cheap hydropower, the constraint became the railroad as the metal had to be carried away from the field. Nobody built a railroad to a powerhouse. The powerhouse had to be born next to one. Then, with electricity production, the constraint just changed clothes. The transmission line became the new railroad.

The question was never where the richest resource was, but what stood next to it. Distance from the wellhead decided what got discovered, and we just kept our eyes closed to further lands. After all, that was the corporate reality. No one would pay Columbus to search for “hot” resources just for fun.

Today I am having a full-circle moment. AI data centers showed up with wire cutters to free geothermal from its chains. US data centers already consume about 4 to 5 percent of US electricity, on a path that could double that share this decade, and they are the first buyer at this scale that can pick themselves up and move to the heat. That unlocks geothermal’s biggest constraint. Distance from the wellhead. Going off-grid with data centers is now real.

These are exciting times for geothermal.

when geography kept beating geology

A drilled field sat idle for fifteen more years.

MILES OF GEN-TIE / MW OF PLANT ≈ 1

A 25 MW field justifies roughly 25 miles of wire. Past that, it usually “strands.” The ratio only holds while the buyer stays on the far side of the wire. In the end, the long-distance relationship was not worth the hassle.

A 2005 Western Governors’ workshop walked through roughly 140 known sites across the West. Most near-market entries remain unbuilt. The rare field that beat the screen did it with brute force. Dixie Valley exists because two ventures merged into one 67 MW plant large enough to justify a dedicated 220-mile line. The “line” is still the point.

McGee Mountain, Nevada, is the field that never beat it. Drilled wells. Measured temperatures around 116°C. Geothermometry arguing for something above 150°C. GeothermEx put the heat-in-place at 25 MW on a P90 and 52 MW on a P50. A DOE-funded report closed the case in 2014 in one sentence. Transmission is “the biggest problem hindering development.” The nearest line was a 115 kV co-op circuit. Interconnection math ran from about $181,000 to $485,000 per megawatt depending on the upgrade.

A drilled field sat idle for fifteen more years. Not for lack of heat, but for lack of a wire.

Then Ormat pulled Nevada well permits in September 2025 and again in March 2026. Permits are not a plant but for sure they are a signal. The company whose field data started my career is looking again at a field the old transmission math left behind. McGee is visible because somebody paid to drill. The larger failure is the field nobody sampled, because a perfect remote well with no path to a customer is just an expensive science project.

Off-grid used to be off the map for a good reason.

Today, the buyer is not hypothetical. Project Red has sent carbon-free power onto the Nevada grid serving Google since 2023, a 3.5 MW pilot. Corsac’s 115 MW supply agreement won PUCN approval in 2025. Cape Station targets 500 MW. Fervo’s S-1 discloses a 3 GW Google framework that does not obligate Google to buy. Ormat signed up to 150 MW for Google through NV Energy. Meta signed 150 MW with Sage Geosystems and 150 MW with XGS Energy into the PNM grid.

Those projects are not off-grid. What they prove is that the buyer can move to the source. Of course a remote campus still needs fiber, crews, permits, and redundancy. But the tradeoff that once kept off-grid off the map no longer holds.

the new buyer is pricing time, not the kWh

The deepest pockets in the economy just became the most impatient buyer in it.

The hyperscaler’s real bill is not kilowatt-hours. It is time to compute. The electron is an input. Compute is the product. Waiting three years in a queue for interconnection is part of the bill. When time is the constraint, power stops being priced as a commodity and starts being priced as a calendar. That changes everything.

The receipt for that already exists. Anthropic’s 20-year campus lease with TeraWulf covers about 401 MW of IT load for roughly $19 billion. The arithmetic comes back near $271 per MWh. That is a lease for buildings and cooling, not a power-only PPA, but nobody signs a number like that to save on electrons. They sign it to buy the calendar.

Look at what the old buyer tolerates. More than 2,060 GW sat in US interconnection queues at the end of 2025, and historically only about 13 percent of queued capacity ever reaches operation. Those gigawatts are not steel waiting for a switch. They are applications waiting for a bill. Withdrawn projects in PJM faced about $599/kW in interconnection costs against $84/kW for the ones that finished. In the non-ISO West, where much of American geothermal actually lives, withdrawn projects averaged $671/kW. Do the arithmetic on 100 MW and a connection budgeted near $8 million comes back near $60 million. Nothing about the plant changed. The wire did.

Rhodium counted 1.7 GW of geothermal waiting for an interconnection agreement in early 2025, enough to raise US installed geothermal by about 40 percent if it all got built. LBNL’s later tally had about 4.8 GW of geothermal active in the queues. Those are only the fields that got close enough to apply. The ones the mile-per-MW screen “stranded” never enter any queue at all. You cannot wait in a line you cannot afford to reach.

Rhodium also ran the counterfactual. Put the data center at the resource instead of clustering it where the fiber already is, and the LCOE of new geothermal falls 31 to 45 percent. Behind-the-meter EGS then covers projected hyperscale growth, whereas clustered siting covers a little more than half. That is a model, not a signed campus, but it is still a real number on moving the load.

A data center ships its product over glass at the speed of light. It can park a hundred megawatts at the wellhead with nothing to haul back out.

the deepest pockets are willing to take risks

The balance sheet scales too. At Krafla, IDDP-1 hit magma at about 2,100 meters, brought up hydrochloric acid, and was later quenched after a master-valve failure. That hole can end a developer. But it is just a rounding error for a hyperscaler. And they are comfortable with economics of scale.

Fervo already showed what repetition does. The first Project Red horizontal well took about 70 days, the fastest Cape well took 21, and the cost on the first four Cape wells fell from $9.4 million to $4.8 million. Impatience favors the resource you can start drilling tomorrow.

being “hot” has its own challenges

Data centers powered by off-grid geothermal energy face distinctive thermal challenges. A geothermal plant has to reject waste heat, either through evaporative water cooling or massive air-cooling fans. Water cooling stabilizes power output, especially in hot ambient conditions, but securing substantial consumptive water rights is an uphill battle getting steeper each day. Air cooling preserves water, but then a binary plant can lose 20 to 30 percent of its net output on a hot afternoon, and even more in extreme heat.

Now pair that plant with a data center, thermal penalty compounds. The same ambient heat derating the generation also forces the data center to draw more power just to cool its silicon at the exact same summer afternoon. Generation falls when demand rises, you just double the hit.

Navigating this tradeoff requires evaluation not just from an engineering and financial standpoint, but also under growing public scrutiny over local resources. Fortunately, solutions like hybrid cooling systems are advancing quickly.

data centers can flex

Historically on-grid systems treats the grid as a battery that absorbs that double hit alongside maintenance cycles, failed units, and power spikes. Off-grid campus must size around them or use a load that can bend.

Today’s interconnection deals with hyperscalers prove the load can bend. To get power on the timeline they need, operators are already agreeing to demand response. Google has signed those deals. Duke’s Nicholas Institute found that roughly 76 GW of new large load could fit on the existing grid with about 0.25 percent annual curtailment. At the end, If I have already agreed to flex my load when the utility calls, I would rather flex it around my own schedule. Skip the line, go off-grid, and control my own calendar.

Not everyone can make that trade. Alcoa found that out the hard way in Iceland. The Bakki smelter was one “frozen” block of demand that needed 400 MW. When those fields came up 140 MW short, the whole project walked away in 2011. Traditional industry is rigid like that. AI training centers share that same stubborn streak. But inference is different. It is modular, distributed, and it can throttle.

it’s time for columbus to redraw the heat map

Remote basins are not empty of resource. They are empty of accidents.

The exploration is starting again, and this time without the chains of geography.

The known geothermal map inherits the transmission bias. The USGS in 2008 counted 9,057 MW identified in the US and 30,033 MW of undiscovered conventional resource. The 2025 Great Basin update is a different object. 135 GW of EGS technical potential in that province if the technology works, not a raised national ceiling. Most of it remains unproven. Blind systems show nothing at the surface. Roughly half of utility-scale discoveries were water wells, mining holes, and other lucky mistakes.

Remote basins are not empty of resource. They are empty of accidents. Under the old screen, even a perfect remote well could be unsellable. Finding heat paid nothing, so nobody paid to find it.

Then DOE looked on purpose. Gabbs Valley is a blind Nevada system. Gradient holes at 124 to 125°C, geothermometry pointing to 130 to 140°C, no hot spring, no steam. Steptoe models 170 to 230°C at about three kilometers. Granite Springs confirmed a thermal anomaly with a gradient well. All undeveloped, because a prospect with no buyer is expensive science.

A buyer at the wellhead flips that. Movable demand gives blind hydrothermal fields a reason to be found, and it pays for the horizontal drilling EGS needs. Basel and Pohang showed what happens when induced seismicity meets cities. An empty basin can run DOE traffic-light protocols without betting the project on a sleeping neighbor. Last October, Mazama hit 331°C at Newberry. The hottest EGS well on record.

Heat was never the scarce input. A reason to go get it was.

The middle of nowhere used to be geothermal’s curse. Now it is where the customer can reach the resource without waiting in everybody else’s line. Ten years ago, I had to charm a regional director to see one field’s data. Ten years from now, the fields worth seeing may have their buyers camped on top of them. Geothermal is about to get the shine it has deserved all along.

Now the map is finally open the way an explorer wants it. Resource first, no wires drawn on it yet.

For the first time in a century, geothermal has a buyer that can break its chains of geography. Today, the companies with the deepest pockets can cut the cord and move straight to the heat. They are pricing time over power. Their balance sheets can absorb uncertainty below ground, and their timelines leave them no choice but to bend the load above it. They are ready to redraw the map around the geology and pay for the fast track to get there.

What did I miss? Let’s discuss.

receipts

• DOE-funded McGee Mountain final report, DE-EE0002830 (2014). Transmission is “the biggest problem hindering development.” ZGlobal interconnection assessment in the same docket. $215,400/MW at 30 MW, $180,600/MW at 80 MW, and $485,000/MW for the 230 kV upgrade case. Measured ~116°C. GeothermEx P90 25 MWe / P50 52 MWe. Nearest line. 115 kV Harney Electric, ~26 miles.
• Nevada Division of Minerals well permits. GEO 1670–1673 (3 Sep 2025). GEO 1730 McGee 84-27 (23 Mar 2026).
• Gen-Tie Screen. The DOE GeoVision barriers report, NREL/TP-6A20-71641 (2019), Section 4.4, records roughly 1 mile per megawatt as a developer screening rule of thumb. It is a screening heuristic, not a universal design formula.
• Western Governors’ Association / 2005 Reno workshop. Roughly 140 known sites evaluated.
• Dixie Valley. 67 MW plant built to justify a ~220-mile, 230 kV line (1988). Ormat more recently reports ~56 MW net at Dixie.
• LBNL Queued Up 2026. 2,061 GW active at end-2025. 13 percent of 2000–2020 queued capacity reached COD by end-2025. ~75 percent withdrawn. Geothermal active in that later tally. ~4.8 GW.
• LBNL Interconnection Costs. PJM withdrawn $599/kW vs complete $84/kW (2020–2022 means). Non-ISO withdrawn $671/kW vs complete $194/kW (PAC/BPA/Duke, 2018–2024). The $8 million to $60 million example is arithmetic on 100 MW, not a cited invoice.
• Rhodium Group (March 2025). 1.7 GW waiting for an interconnection agreement, a potential 40 percent increase in US installed geothermal. 31 to 45 percent is the drop in national capacity-weighted LCOE of new geothermal when sited at data centers versus clustered siting, in a modeled behind-the-meter EGS case. Clustered siting. BTM EGS meets 55–64 percent of hyperscale growth (15–17 GW). Power-first siting. 100 percent of projected growth.
• Data Center Power Consumption. LBNL 2023 electricity. 176 TWh, ~4.4 percent of US use. IEA 2024. 183 TWh, >4 percent. LBNL projection. 6.7 to 12 percent of US electricity by 2028.
• Duke Nicholas Institute. Norris et al., Rethinking Load Growth (February 2025). 76 GW of new flexible load at 0.25 percent annual curtailment.
• Fervo Energy. Project Red operational for Google as grid CFE, Nov 2023, ~3.5 MW. Corsac 115 MW. PUCN approval of NV Energy Clean Transition Tariff, 13 May 2025. Cape Station 500 MW target. S-1 (2026). 3 GW Google framework dated 19 Mar 2026, non-binding (“does not obligate Google to purchase power from us”). Drilling. first Project Red horizontal well ~70 days. fastest Cape well 21 days. Cost. first four Cape horizontal wells, $9.4M to $4.8M.
• Hyperscaler Deals. Ormat / NV Energy / Google up to 150 MW, announced 17 Feb 2026, COD 2028–2030. Meta + Sage Geosystems 150 MW (26 Aug 2024). Meta + XGS Energy 150 MW into the PNM grid (12 Jun 2025).
• Alcoa Bakki Withdrawal (2011). 400+ MW smelter. Iceland National Planning Agency comment on the joint EIA, ~140 MW shortfall. Planning spend reported between about $9 million and $17 million.
• Iceland Deep Drilling Project (IDDP-1 at Krafla). Rhyolitic magma at ~2,100 m. HCl in superheated steam. completed and flow-tested 2010–2012. quenched after master-valve failure.
• USGS Assessments. 2008 national assessment. 9,057 MWe identified. 30,033 MWe undiscovered conventional. USGS 2025 Great Basin fact sheet. 135 GWe EGS technical potential in that province, upper 6 km, if sufficient technological advances occur. Not a national hydrothermal-plus-EGS ceiling.
• Seismicity Protocols. DOE Induced Seismicity Mitigation Protocol and traffic-light implementations. Basel (2006) and Pohang (2017) as cautionary cases. Two-thirds of Great Basin Quaternary faults have no well within a kilometer.
• DOE Play-Fairway. Gabbs Valley (TG holes 124–125°C at ~152 m. geothermometry 130–140°C indicated). Steptoe Valley (modeled 170–230°C at ~3 km). Granite Springs Valley (gradient well ~96°C at ~250 m).
• Mazama Energy, Newberry Volcano, Oregon (28 Oct 2025). 331°C bottomhole in an EGS demonstration well. 15 MW pilot targeted 2026. 200 MW planned build-out.
• Anthropic–TeraWulf Campus Lease (TeraWulf 8-K). ~401 MW IT load, ~$19B contracted lease revenue over 20 years, implied ~$271/MWh. Buildings and cooling, not a power-only PPA. A 2026 campus lease that wanted the term that long is still a number on pricing time, not electrons.
• Summer Stacking. Binary geothermal in hot dry basins typically loses about 20 to 30 percent net output on peak afternoons because air-cooled condensers work harder. data-center cooling demand can also rise at the same hour. NREL ARRA hybrid-cooling report, NREL/TP-5500-58024. above 30°C, air-cooled binary output may decrease by more than 50 percent on the hottest days.
• Google Demand Response. Google announced demand-response agreements with Indiana Michigan Power and TVA in August 2025, but the notice periods are not public. Google’s earlier Northern Wasco County PUD pilot in Oregon demonstrated day-ahead demand response.