The AI Power Problem, Part II: The Clean Energy Beneath Our Feet

In my previous article, I argued that the next major bottleneck for artificial intelligence may not be model intelligence, chip supply, or investor enthusiasm. It may be something far more physical: electricity.

AI is becoming a power-hungry industrial system. Data centres are expanding rapidly, and the International Energy Agency expects global data-centre electricity consumption to roughly double by 2030, reaching about 945 terawatt-hours. That is not just a technology trend; it is an energy-infrastructure challenge. The future may be intelligent, but as I wrote before, we still need somewhere to plug it in.

This raises the next question: if AI needs enormous amounts of electricity, where should that electricity come from?

The Search for Firm Clean Power

Solar and wind are essential, and their growth has been extraordinary. In 2025, global renewable power capacity reached a record 5,149 GW, with solar alone adding 511 GW and reaching 2,392 GW worldwide. Renewables now account for nearly half of global installed electricity capacity. But AI data centres do not only need cheap electricity. They need reliable electricity every hour of the day, every day of the year. That is where the conversation becomes more difficult.

The missing piece is firm, clean power: electricity that is low-carbon, available around the clock, and not dependent on whether the sun is shining or the wind is blowing. Nuclear power can play that role, but new projects are slow, expensive, and politically complex. Batteries help, but long-duration storage at hyperscale remains a work in progress. Hydropower is valuable but geographically limited.

This is why geothermal energy deserves much more attention.

Why Geothermal Energy Matters

Geothermal power is not new. Countries such as the United States, Indonesia, the Philippines, Türkiye, Kenya, Iceland, New Zealand, Italy, Mexico, and Japan have used it for decades. The problem is that traditional geothermal has been limited to areas where hot underground resources are naturally close to the surface. That made it useful, but niche.

Enhanced geothermal systems may change that.

The idea is simple in concept, though complex in execution. Instead of waiting for nature to provide an ideal geothermal reservoir, engineers drill deep into hot rock, create or enhance underground fractures, inject water, allow the rock to heat it, and bring the hot fluid back to the surface to generate electricity. The system can then reinject the fluid and repeat the cycle. In effect, it turns deep hot rock into a controllable underground heat exchanger.

Google, Fervo Energy, and the Proof of Concept

This is what made Google’s partnership with Fervo Energy so important. In 2021, Google partnered with Fervo to develop an enhanced geothermal project in Nevada. Fervo drilled horizontal wells, applied techniques borrowed from the oil and gas industry, and used fiber-optic sensing to monitor the underground system in real time. The project began supplying carbon-free electricity to the grid serving Google’s Nevada data centres in 2023.

At first, the scale was modest: about 3.5 MW. But the significance was not the size. The significance was proof. It showed that enhanced geothermal could move from theory and pilot projects into commercial operation.

Google and Fervo then expanded the model. In 2024, Google, NV Energy, and Fervo Energy announced a deal to increase advanced geothermal supply in Nevada from 3.5 MW to 115 MW over roughly six years. The arrangement used a new Clean Transition Tariff model, designed to allow large electricity users to support new clean, firm capacity through the local utility system rather than simply buying power directly from a developer.

In 2025, Nevada regulators approved the model, and Google described it as a replicable pathway for large users and utilities to accelerate clean, reliable electricity.

Then came the bigger step. In September 2026, Fervo announced a 396 MW power purchase agreement with Google from its Cape Station enhanced geothermal project in Utah, expected to come online in 2028. The agreement also gives Google an option to expand its offtake by around 600 MW, bringing the potential total close to 1 GW by 2030.

This matters because AI infrastructure is increasingly looking for exactly this kind of power: clean, large-scale, firm, and available 24/7.

The Global Potential of Geothermal Energy

Globally, geothermal remains underused. It supplies only a small share of world electricity today, and most capacity is concentrated in a handful of countries. Geothermal currently supplies around 2% of global electricity, while the top 10 geothermal producers account for about 94% of existing capacity.

The International Energy Agency has estimated that, with modern drilling and technology transfer from oil and gas, next-generation geothermal could become much larger, potentially supplying up to 8% of global electricity by 2050 if costs fall and investment scales.

The U.S. Department of Energy has also identified major upside. Its geothermal assessments suggest that geothermal power could provide around 120 GW of firm electricity capacity in the United States by 2050. That would be enough to make geothermal a serious part of the national electricity system, not just a regional niche.

Why Geothermal Fits the AI Era

The attraction is clear. Geothermal has several advantages that directly match the AI power problem.

First, it is always on. Unlike solar and wind, geothermal does not depend on weather or daylight. That makes it valuable for data centres, industrial loads, and grids with rising demand.

Second, it has a small surface footprint compared with many other energy sources. Much of the “plant” is underground.

Third, it can use skills, equipment, and supply chains from the oil and gas industry: drilling, subsurface mapping, reservoir engineering, horizontal wells, and project operations. This is politically and economically important because it provides a bridge from fossil-fuel expertise to clean-energy infrastructure.

Fourth, it can support the grid rather than simply add intermittent generation. AI data centres need firm capacity, and grids need resources that can provide stability as electrification grows.

Geothermal Is Not Magic

But geothermal is not magic, and it is not truly “infinite energy.” The Earth’s internal heat is vast and long-lived, but each project must manage reservoir performance, drilling cost, seismic risk, water use, permitting, financing, and local geology. Enhanced geothermal systems also need to prove that they can scale economically, not just technically.

The Google-Fervo story is therefore not that Google discovered infinite energy. The better story is that Google helped prove a new commercial pathway for firm clean energy.

Lessons from the Google-Fervo Model

That pathway has several lessons.

The first lesson is that large technology companies should not only consume power; they should help create new clean power. Buying renewable credits is no longer enough. The next generation of corporate energy strategy must bring new, additional, dispatchable, carbon-free capacity onto the grid.

The second lesson is that utility regulation matters. Google’s Nevada model is important because many data centres operate in regulated electricity markets where they cannot simply buy power directly from any generator they choose. The Clean Transition Tariff gives utilities, customers, and developers a mechanism to finance new clean capacity while protecting ordinary customers from bearing the full cost.

The third lesson is that AI companies need energy portfolios, not energy slogans. Solar, wind, batteries, nuclear, geothermal, long-duration storage, grid upgrades, demand flexibility, and efficiency all have roles. The future will not be powered by one perfect source. It will be powered by a portfolio designed around reliability, cost, carbon intensity, and speed.

What This Means for Saudi Arabia and the Region

The fourth lesson is especially relevant for countries such as Saudi Arabia and the wider region. Geothermal is often associated with Iceland, Kenya, Indonesia, or the western United States, but geothermal potential also exists in parts of the Middle East.

Saudi Arabia has geothermal prospects along the Red Sea coast, including areas near Al-Lith, Jizan, NEOM, and other western regions. KAUST has also been studying geothermal potential as part of broader clean-energy research. A 2024 study highlighted potential geothermal reservoirs in the Riyadh region, with estimated temperatures of 80°C to 120°C at depths of 2 to 3 km.

For Saudi Arabia, this should not be seen only as a power-generation question. It could become part of a broader industrial strategy: geothermal mapping, drilling innovation, clean power for data centres, cooling applications, district energy, desalination support, and the reuse of subsurface capabilities already familiar to the Kingdom’s energy sector.

The AI Race Is Becoming an Energy Race

The AI race is becoming an energy race. Countries that can provide clean, reliable, scalable electricity will have an advantage not only in climate policy, but also in cloud computing, AI infrastructure, digital sovereignty, and industrial competitiveness.

The previous article ended with a warning: the future may be bright, but we have nowhere to plug it in.

This follow-up should end with a more constructive thought: perhaps part of the plug has been beneath our feet all along.

The challenge now is not to romanticize geothermal as a miracle. It is to treat it as serious infrastructure: map the resource, finance early projects, modernize regulations, connect it to data-centre demand, and use large buyers like Google to pull the technology down the cost curve.

AI needs power. The climate needs clean power. Grids need firm power.

Enhanced geothermal may not be infinite energy. But if developed properly, it could become one of the most important clean-energy answers to the AI era.

#ArtificialIntelligence #AIInfrastructure #CleanEnergy #GeothermalEnergy #EnhancedGeothermal #DataCenters #EnergyTransition #SustainableAI #DigitalInfrastructure #PowerGrid #CarbonFreeEnergy #FutureOfAI

---

Originally published in the VerbionPulse newsletter on LinkedIn: The AI Power Problem, Part II: The Clean Energy Beneath Our Feet.