AI applied to seismic, geophysical, and historical drilling data — the first new blind-system geothermal discovery in three decades.
Zanskar Geothermal
venture active confidence: Medium status: Draft updated 2026-07-14
Summary
Zanskar Geothermal uses machine learning and geoscience to predict hidden heat, then tests the prediction by drilling. If its blind discoveries generalize, geothermal exploration could shift from expensive geological luck to repeatable inference—an enabling breakthrough with credible global, S-scale upside.
Impact
Geothermal is one of the few clean power sources that is firm and dispatchable: it is on at 3 a.m. in February as readily as at noon in July. The catch has always been site discovery — most subsurface heat is invisible from the surface, and traditional exploration is slow, expensive, and high-risk. The available evidence suggests Zanskar's bet is that breaking the exploration bottleneck shifts the problem for the entire industry from "can we find sites?" to "can we drill them fast enough?" — a far more tractable problem with established playbooks.
If the method generalizes, the second-order impact is enabling-infrastructure scale: every other geothermal developer (Fervo, Cyrq, Ormat, and emerging entrants) becomes more capital-efficient, and previously uneconomic regions across the western United States and beyond become drillable.
What They Are Building
Zanskar is building an exploration tool that ingests seismic, geophysical, geologic, and historical drilling data and outputs site-level predictions of heat and permeability. The product surface for talent is unusual: it sits at the intersection of applied earth science, machine learning under genuine uncertainty, and field validation. According to public coverage, founders Carl Hoiland (CEO) and Joel Edwards (CTO) come from geoscience and data science respectively, which suggests a culture that rewards technical depth.
The 2025 blind-system discovery is currently the strongest demonstration that the inference is doing real work, not just picking up signals visible to the naked eye.
What They Need Now
Hiring (Salt Lake City hybrid / field) spans project development, mechanical and systems engineering, equipment and geologic technicians, and platform/AI engineering, with seismology and software roles rotating through the board. That matches the stack: project development and field ops alongside geoscience and ML platform work. For talent matching, Zanskar remains a strong candidate for AI engineers who want predictions falsified by drilling, and for subsurface operators who can navigate land access and federal leases.
Who Could Help
Useful helpers include subsurface researchers, federal grant writers (DOE has been an active backer of geothermal innovation), project finance advisors, utility interconnection specialists, lease and permitting counsel for federal lands, and operators who have run drill campaigns. The company is also a useful reference point for founders trying to argue that AI's most consequential applications may not be in software but at the interface between models and physical systems.
Utah Context
Zanskar is headquartered in Salt Lake City and sits inside the same Utah geothermal cluster as Fervo Energy's Cape Station project in Beaver County and the federally-funded Utah FORGE research site. Together, they make the case that Utah is one of the most concentrated places in the world for the entire enhanced-geothermal stack — exploration, drilling, reservoir engineering, and policy.
Evidence
- Official Website: Zanskar Geothermal · https://zanskar.com
- Source: MIT Technology Review on AI Geothermal Discovery · https://www.technologyreview.com/2025/12/04/1128763/ai-geothermal-zanskar/
See Also
- Fervo Energy — Utah EGS developer deploying the firm-power plants that Zanskar's exploration could de-risk for future developers
- Rodatherm Energy — closed-loop geothermal pilot in Utah's Great Basin; a different approach to the same resource constraint
- Torus — Utah grid-edge storage; complementary firm-power infrastructure for the same baseload clean-energy stack
Open Questions
- One confirmed blind-system discovery is a strong signal but not yet proof that the method generalizes; future verifications are the central evidence to watch.
- The revenue model — licensing exploration software, co-developing projects, or operating wholly-owned sites — has direct implications for which talent profiles fit best.