Next-generation geothermal energy, which for decades was considered promising but commercially unviable, is finally scaling up — and artificial intelligence is the engine driving this comeback. Fervo Energy, a Houston-based company specializing in enhanced geothermal systems (EGS), announced in its third-quarter earnings report that it now has $11.9 billion in contracted revenue, a massive jump from the $7.2 billion reported in the previous quarter. This number is not just a financial indicator: it is concrete proof that the business model Fervo built around its Cape Station project in Utah is gaining conviction among investors and energy buyers.
What made this moment possible is a convergence of two trends that, two years ago, seemed misaligned. On one side, hyperscalers — Google, Meta, Amazon, Microsoft — are in an unprecedented pace of data center expansion, driven by the infrastructure race for AI. On the other, power constraints in the regions where these data centers can be built have become a critical bottleneck. The existing grid simply cannot keep pace with the speed at which tech companies want to add compute capacity, and the alternative — waiting for new transmission lines — can take over a decade. In this context, next-generation geothermal emerged as a solution combining constant availability, low emissions impact, and increasingly, a competitive cost per kilowatt.
Fervo is not the only company in this space. Ormat Technologies, a traditional geothermal operator that has run conventional plants in the Western United States for 60 years, has also made an aggressive pivot toward the AI market. But Fervo stands out in the current narrative because its Cape Station project is the largest example of enhanced geothermal power ever built at commercial scale. The site, in Beaver County, Utah, has an estimated potential of 8.4 gigawatts of electricity — double what experts initially projected. The project consists of multiple generation units called "GeoBlocks," each capable of producing between 33 and 50 megawatts. They are designed to be delivered in phases, meaning a data center can start operating with one block and expand as more GeoBlocks come online. It is, essentially, the energy version of the "infrastructure as code" concept: build what you need, scale incrementally.
The relationship between Fervo and Google is the core of this model. The power purchase agreement (PPA) signed in September 2026 for 396 megawatts — internally dubbed "Cape Google 400" — is the largest geothermal PPA ever executed. The project is scheduled to enter commercial operation in three stages starting in the third quarter of 2028. The contract allows Google not only to consume the generated power but also to resell excess energy to other customers, turning the operation into a shared energy infrastructure asset. As part of the agreement, Google has the option to expand its share by up to 600 MW additional, bringing the total to nearly 1 gigawatt by June 2030.
What stands out is not just the scale, but the cost evolution. The first phase of Cape Station cost US$ 7,000 per kilowatt of capacity. For the Google project, Fervo has already reduced this to US$ 5,500 per kilowatt. The company sets a long-term target of reaching US$ 3,000 per kilowatt — a level at which, with existing tax credits, it would be equivalent to the cost of a natural gas power plant fueled by gas at US$ 6 per million BTUs. If the target is achieved, enhanced geothermal would have a structural advantage over natural gas in a market where fossil fuel price volatility is a chronic variable.
For context, geothermal received only US$ 22 million in global financing in 2018, according to the International Energy Agency. Today, companies like Fervo are operating with budgets already exceeding US$ 200 million per quarter in capital expenditure. Ormat, for instance, reported capital expenditures of US$ 226.5 million in Q2 2026, more than double the US$ 108 million in the same period of the prior year. This is exponential growth in a sector that, for decades, operated on a modest and predictable scale.
But the transition is not without risks. Fervo ended the quarter with an operating loss of US$ 43.7 million, which obliges it to maintain rigorous financial discipline while scaling. The company holds US$ 1.9 billion in cash and US$ 301.6 million in long-term debt — enough to provide a solid push, but not enough to cover project missteps. Enhanced geothermal relies on drilling at depths and temperatures that, in many cases, represent frontiers of technical knowledge: each new well requires learning about local conditions, and the success rate is not yet perfect. Fervo openly admits that early drilling hit lower temperatures than expected, which forced design adjustments on the GeoBlocks.
Beyond technical challenges, there is a quiet but fierce competition with other baseload energy sources. Nuclear energy, for example, is being revisited as a clean and constant generation alternative — OpenAI announced the construction of a dedicated nuclear plant for its data centers, and energy company TeraWulf expanded its compute capacity with nuclear power. Natural gas, although with significant emissions, remains the fastest source to deploy, with turbines available in months rather than years. Anthropic itself, one of the leading AI companies, recently signed a US$ 3.4 billion deal to purchase natural gas turbines instead of waiting for the equipment supply queue, showing that in the short term, gas is still the fastest path. Geothermal needs to convince investors that its heavier upfront capital investment is worth it for decades of zero-emission, stable output.
What Fervo demonstrates is that the AI energy transition will not be powered by a single source. It will be a complex mix of nuclear, natural gas, solar, wind, and geothermal — each occupying its place in hyperscalers' energy strategy. Geothermal positions itself as the constant baseload energy, the one that does not depend on weather conditions or fuel cycles. If the cost per kilowatt continues to fall at the pace Fervo projects, this sector could transition from a niche for climate-focused investors to a pillar of the global AI infrastructure.
The question that remains is whether this cost reduction curve holds up when real-scale construction begins, or whether the engineering challenges already visible in the first GeoBlocks will multiply as the project expands to 8.4 gigawatts. What can be stated with certainty is that geothermal energy — for decades an "almost there" energy option — has finally found its moment. What remains to be seen is whether it can sustain the speed needed to arrive before the 2030 horizon, when most planned data centers need to be operational.
Sources: TechCrunch, Fortune, Energy Monitor
✓ Independent sources cross-checked and verified before publishing
