PNNL is supporting five Genesis Mission projects that advance AI for scientific research, including energy reliability work.

Oct. 12, 2026 — Keeping the nation's energy reliable and affordable means solving challenging scientific problems faster than ever before. Researchers at Pacific Northwest National Laboratory (PNNL) are bringing their expertise in grid operations, geosciences, and computing to five new Genesis Mission-funded projects that use Artificial Intelligence (AI) to meet these challenges.

The five projects aim to tap new energy sources deep underground and understand how fluids flow through those subsurface rocks, integrate how uncertainty about weather impacts grid operations planning, allow scientists to plan complex physics experiments using plain-language instructions, and create a reusable framework that uses AI-coding agents to keep computers running smoothly.

In each case, AI is not replacing scientists. It is becoming a powerful partner that helps them work faster and make better decisions.

"These five projects reflect the spirit of the Genesis Mission and its emphasis on partnering across government, industry, and academia to accelerate progress toward solutions in science, energy, and national security that were previously out of reach," said PNNL Director Deb Gracio. "Together, we are mobilizing the talent, facilities, and data these challenges require to ensure our energy future and strengthen national leadership in discovery science."

**Reducing Uncertainty in the Power Grid**

Grid planners choose sites for new transmission lines and substations years in advance without being able to predict weather patterns or electricity demand. This kind of guesswork could leave the grid unprepared for extreme changes in energy demand, such as a severe cold snap.

A team led by MIT, with PNNL's Nathalie Voisin, is using AI to help planners prepare for many possible futures. The researchers are training AI models to produce realistic weather scenarios about 1,000 times faster than currently possible. Other parts of the project include generating electricity demand scenarios and economic forecasts. In the past, this many scenarios would have overwhelmed available computing power. But AI and smart model tailoring make the plan tractable. The approach will be tested on grids in New England, which relies heavily on natural gas, and in Texas, where wind, solar, and demand are all growing fast.

**Fluid Flow Models That Get Closer to the Real Thing**

Recovering critical minerals and tapping geothermal heat depend on predicting how fluids move through rock. Simulations can help, but the complexity of the systems studied can quickly overwhelm traditional methods. A team led by Oregon State University, with PNNL's Tim Scheibe, aims to fix that with AI.

The researchers plan to train AI on detailed simulations of flow through individual pores and on 3D X-ray scans of actual rock or sediment samples. The AI evaluates shapes and interprets how those shapes affect physical properties, such as how easily fluid flows through different kinds of porous formations. The team is incorporating physics rules into the AI to keep its answers realistic and reliable.

**AI Coding Agents to Close the Computing Language Gap**

Much of the nation's scientific infrastructure runs on computer programs built over decades using different programming languages than we use today. Each switch to a new program can create a translation problem. Programmers have been patching these gaps manually, but now a team led by Argonne National Laboratory, with PNNL's Andrew Lumsdaine, is building an AI system to close those gaps.

Using AI coding "assistants," they will create a reusable framework that allows programmers to quickly test AI-generated coding translations. When AI proposes a code change, the code will be run, tested for scientific accuracy, and refined. Every step will be recorded, and successful solutions will be shared widely. The team will test the framework on real scientific programs used for chemistry, wind turbines, and land modeling. Results will be measured in time, cost, and human effort savings.

**AI to Speed Access to Energy Trapped Underground**

Enhanced geothermal systems have the potential to supply clean power around the clock. But building them can be risky because so much about the deep underground is still not well understood.

A team led by the University of California, Irvine, with PNNL's Kevin Rosso, is building a team of AI agents to bring together existing knowledge and build upon it. The agents will have access to data from 20 monitored geothermal sites, supercomputer simulations, and published research. They will update predictions as new data arrive, forecast how fractures will grow, and assess earthquake risk. Operators will be able to ask the system questions through a chat interface. Built-in physics checks will keep answers realistic. The team aims to cut development time and financial risk by up to 35 percent.

**A Shared Resource for Advanced Physics**

In fall 2025, seven national laboratories began building a shared AI platform that connects AI to the software that runs nearly a dozen advanced particle accelerators operated by the Department of Energy. These complex instruments smash particles together at near light-speed to reveal the fundamental laws, forces, and building blocks of the physical universe. Now a team led by Lawrence Berkeley National Laboratory, with PNNL's Malachi Schram, plans to expand it. Scientists will be able to use a plain-language request, while behind the scenes, AI translates the request into precise machine commands, drawing on knowledge from many facilities and disseminating improvement at one lab to benefit them all.