PNNL Teams Up with Fervo Energy and NVIDIA to Accelerate Geothermal Energy Development

RICHLAND, Wash. — All around the world, heat is key to generating electricity. By burning hydrocarbons, splitting atoms or tapping into Earth’s hot subsurface, humans generate much of the world’s electricity by heating water to create steam that turns massive generators. 

Of these resources, though, heat from Earth’s subsurface hasn’t contributed much to the nation’s electricity supply — in 2023, geothermal made up just 0.4% of all electricity generated in the United States. Energy experts see geothermal as a high-potential, untapped and underused resource. Especially as electricity demand is expected to rise over the next few decades, research institutions and industry alike are looking to increase geothermal energy production. 

To help accelerate this production, the Department of Energy’s Pacific Northwest National Laboratory has teamed up with geothermal company Fervo Energy and AI technology company NVIDIA to build a digital twin of enhanced geothermal reservoirs. A digital twin is a virtual platform that simulates a physical asset, like a battery, hydropower dam or, in this case, a geothermal reservoir. Through AI and computer simulations, digital twins mimic real physical processes. 

With Fervo Energy and NVIDIA, PNNL researchers will build Enhanced Geothermal System Twin, a virtual platform that will mimic the behavior of a real geothermal reservoir. Once launched, the platform would ultimately be available to any geothermal plant operator to help them make quick decisions to maximize electricity generation. 

“Current modeling capabilities for geothermal systems are too slow to fully incorporate and analyze production data, which can lead to an underutilized resource. A digital twin would allow EGS operators to understand, in real time, the dynamics of their reservoir and act quickly to maximize the power generation potential,” said Maruti Mudunuru, an Earth scientist at PNNL and principal investigator of the project.

“Fervo will provide us with their proprietary data for their geothermal sites in Nevada and Utah, and NVIDIA will provide their technical expertise on developing AI surrogates used in the digital twin for Earth’s subsurface,” he added.

In 2023, Fervo established Project Red in Nevada, which now generates 3 megawatts for the grid. Project Cape Station in Nevada is set to deliver 100 MW to the grid in 2026 and increase to 500 MW by 2028. The team will use currently available results to begin training the digital twin immediately and will continue developing it as additional production data comes online.

“We see digital twins as a critical step toward enabling data-driven geothermal operations,” said Sireesh Dadi, senior manager for data acquisition and advanced analytics at Fervo Energy. “Through this collaboration, we are contributing field data, operational context and validation use cases to ensure that the digital twin platform delivers actionable insights at the speed required for real-world decision-making.” 

What’s happening down there?

“Geothermal energy is all about extracting heat from Earth’s subsurface. You inject cold water into the system and then you get hot water out of the reservoir,” Mudunuru said.

By “the reservoir,” he means an enhanced geothermal system, consisting of a series of drilled wells and fractures up to 10,000 feet below ground (for reference, the Empire State Building is about 1,250 feet high). Some of the fractures are “enhanced” by hydraulic fracturing — or high-pressure water injection — to be wider, longer or to connect fracture networks to each other. Enhancing the fractures means more surface area and better heat recovery. 

Then, cold water is injected through the fractures, where it absorbs heat from the surrounding rock (which can reach up to 555 degrees Fahrenheit) and rises to the surface. At the surface, the water becomes steam, which then spins electricity-generating turbines. Fervo’s geothermal power generation process is designed so that the water can cool at the surface and none of it is lost to evaporation. The water is then injected back into the subsurface, creating a continuous electricity-producing process.

Although the process sounds straightforward, operators have a lot of factors to consider, Mudunuru said. 

“When you pump the water, you want the water to pass through the whole network of fractures because if it doesn’t, you’re not accessing all the potential heat,” Mudunuru said. “Plant operators need to answer questions like ‘How many monitoring wells does the system need? How do we design those wells? How much water should we inject?’”

Building a digital twin 

It’s difficult to answer these questions when operations occur 10,000 feet below ground. Fervo Energy deploys fiber-optic cables and uses acoustic technology to map and gather intelligence from the subsurface, but processing and analyzing all that data takes too much time for operators to act swiftly. Fluctuations in the operating data can indicate issues in the wells, reservoir or pipelines that may require attention. Current models that help represent the dynamics of a geothermal system can take weeks to run. 

EGS-Twin would run in real time and allow operators to respond quickly to any underground problems that may arise.

To build the digital twin, PNNL researchers will train scalable AI models on NVIDIA AI infrastructure to learn and process field data from Fervo’s EGS asset. The team will then incorporate those trained AI models into the NVIDIA Omniverse libraries, which will be able to show a physical model of the geothermal system. 

NVIDIA tools would be able to apply enormous amounts of records from Fervo’s field production to create a simulation of the entire fracture system, showing operators whether injected water is successfully flowing through that network to collect as much heat as possible. The final EGS-Twin will contain anonymized data so that other geothermal plant operators can adapt it to their own operations.

“Geothermal has the potential to be a reliable, always-on source of energy, but unlocking it at scale will require advanced computing to better understand complex reservoirs thousands of feet below the surface,” said John Josephakis, global vice president of high-performance computing and supercomputing at NVIDIA. “PNNL and Fervo Energy are using NVIDIA accelerated computing to build EGS-Twin, applying AI and simulation to help improve reservoir modeling, planning and operations.”

The EGS-Twin platform should be ready to deploy by 2029. The project is funded by DOE’s Hydrocarbons and Geothermal Energy Office.