Fervo Energy, PNNL and NVIDIA build AI digital twin for geothermal
Fervo Energy (Nasdaq: FRVO), Pacific Northwest National Laboratory (PNNL) and NVIDIA have announced a joint project to develop EGS-Twin, a digital twin platform for enhanced geothermal systems (EGS). The platform is scheduled for implementation by 2029 and will draw on field data from Fervo's operational sites in Nevada and Utah.
The agreement combines three distinct capabilities: Fervo's proprietary subsurface field data, PNNL's high-performance computing workflows and data pipelines, and NVIDIA's accelerated computing infrastructure including its Omniverse libraries. PNNL researchers will train AI models on NVIDIA hardware, using US Department of Energy supercomputing resources to run large-scale simulations, then integrate those models into NVIDIA Omniverse to produce real-time operational forecasts.
Jack Norbeck, Fervo's chief technology officer and co-founder, said the platform has the potential to "reshape reservoir management, improve heat recovery, and enhance system reliability" by combining physics-based models with AI-driven forecasting.
What EGS-Twin is designed to do
EGS-Twin targets one of the persistent challenges in geothermal development: understanding subsurface behaviour without continuous, expensive physical intervention. By integrating high-resolution sensor data from live sites with physics simulations, the platform is intended to give operators faster visibility into reservoir changes, help optimise power generation output, and reduce the operational risk that comes with uncertainty about rock and fluid dynamics underground.
PNNL will begin training the model immediately using data already available from Fervo's sites, with the platform to be refined continuously as more production data comes online. No interim milestones or accuracy benchmarks were disclosed in the announcement, and Fervo did not name any customers who will access the platform once operational.
Market context and competitive read-across
Geothermal has historically attracted less technology investment than solar or wind, but the sector is gaining momentum as AI hyperscalers and utilities seek firm, dispatchable low-carbon power to back up intermittent renewables. Fervo has positioned itself explicitly around that demand, and its GeoBlock product line is framed as a modular, repeatable approach to utility-scale EGS deployment.
Digital twin technology is well established in oil and gas reservoir management and in semiconductor manufacturing, but its application to EGS is still early-stage. The use of NVIDIA Omniverse as a physics simulation backbone is consistent with NVIDIA's broader push into industrial simulation, where it competes with Ansys, Siemens and Dassault Systèmes for enterprise physics workloads. PNNL's involvement provides access to federally funded supercomputing capacity that most commercial geothermal developers could not procure independently, which is a meaningful cost advantage during the training phase.
The 2029 implementation target gives the partnership a three-year runway. That timeline aligns broadly with the period in which several US geothermal developers, including Fervo, are expected to bring additional EGS capacity online under power purchase agreements with data-centre operators. A validated digital twin could lower the cost of drilling decisions and accelerate the learning curve for new site development, though the release does not quantify either benefit. Investors will watch for named customer deployments and published performance benchmarks as the platform moves from training to production use.