RICHLAND, Wash. - The Department of Energy's Pacific Northwest National Laboratory has partnered with Amazon Web Services to explore and validate some of the most advanced artificial intelligence tools that could help operators of the nation's electric grid respond more effectively to severe weather, rising electricity demand and cyber or physical threats.
PNNL grid experts will develop and evaluate advanced AI systems in collaboration with AWS and test different scenarios in PNNL's Electricity Infrastructure Operating Center with a secure, operations-grade hybrid-cloud Amazon environment.
The work will address challenges critical for utilities, grid planners, technology developers and government stakeholders alike: improving situational awareness, reducing operator burden, strengthening grid resilience and helping the grid adapt to faster and more variable supply and demand situations.
"With rapidly increasing demands for electricity, grid control centers and the operations of the grid will need to evolve significantly. By applying new technologies like AI in ways that are rigorous, practical and human-centered, we can advance the grid while maintaining reliable, secure and affordable power for the nation," said Jim Ogle, PNNL's director of electricity infrastructure.
The work supports PNNL's Enhanced Visibility & Event Response (EVE@PNNL) capability, launched earlier this spring as an effort to develop real-time AI-driven analytics, realistic simulation environments and novel sensing and monitoring tools to address emerging challenges facing the nation's grid operators.
Next-generation AI systems
PNNL has decades of experience studying and developing advanced computing tools to support grid operations. In the wake of the largest blackout in U.S. history in 2003, the Laboratory led a DOE effort then called the North American SynchroPhasor Initiative to deploy precise measurement tools across the electrical grid. These tools helped operators better understand system health and dynamics at scale.
And with the sudden influx of data points from thousands of sensing instruments across the grid, PNNL researchers embraced machine learning, a type of AI, to study all that data. Over the past two decades, PNNL has continued to lead the engineering and analysis of the electric grid focused on improving reliability and addressing emerging challenges.
In May 2025, the Lab published a report detailing opportunities for grid operators to incorporate AI-specifically in the form of large language models-into their work.
However, LLMs can produce errors when answering user queries. Known as "hallucinations," these errors occur because LLMs are trained to predict what letters or words follow the previous letters or words based on probability. LLMs don't actually "know" what a given word or letter means.
To combat that challenge, researchers have increasingly looked into an advanced type of AI model called neuro-symbolic AI that combines the pattern recognition of machine learning and generative AI with topic-specific knowledge and reasoning-for the grid, that means understanding physics.
Electricity on today's grid is largely generated by rotating machines that are designed to maintain a frequency of 60 hertz. A useful AI tool would need to be trained with knowledge about how electricity is generated and how it flows.
To that end, AWS's neuro-symbolic AI tools could help operators make decisions during fast-evolving disruptions, such as how to optimize power shutoffs during wildfires, how to keep power flowing during a hurricane or how to protect the grid from a cyber or physical attack.
"We want to augment human oversight across these systems and data without overwhelming the operators with too much information," said Todd Hay, a data and software engineering researcher at PNNL. "That means developing tools that can help triage information and strengthen mission assurance in face of disruptions. For national security missions, this is critical as we want to ensure that the planes will fly, that the ships can launch and that the military can operate regardless of whether there's a wildfire or an attack on the grid."
Expanding, accelerating grid
The partnership with AWS comes at a time when large-scale computing facilities, including data centers, are contributing to rapid growth in electricity demand. In some cases, these facilities can require as much electricity as a small city. And it's not just amount of electricity demand; it's the speed at which that need for electricity can rise and fall. Data centers can create sharp changes in electricity demand within tens of milliseconds.
"AWS is proud to partner with PNNL on critical grid modernization initiatives that will strengthen America's energy infrastructure for the AI era," said Jamie Baker, director of federal civilian health, science and policy at AWS. "Through our collaboration with leading national labs like PNNL, we're leveraging advanced AI technology to help modernize the grid, ensure reliable power delivery and support the communities where we operate."
The protection systems that ensure the reliability of the overall grid are not designed to withstand the seemingly instantaneous changes driven by highly concentrated loads at data centers.
And neither are the humans in the control room.
"Humans can't operate at the sheer speed at which data centers can operate, and so we need automated tools and decision-support, including AI systems, to handle those dynamics," Hay said. "We need tools that can handle data triage and help turn complex information into actionable insight that the human can then use to make an informed decision."
The new partnership supports DOE's Genesis Mission, an integrated ecosystem to accelerate discovery and strengthen U.S. leadership in AI-powered science. More broadly, the effort reflects growing national interest in developing trustworthy, operationally relevant AI tools that can help strengthen grid resilience, secure critical infrastructure and prepare the energy sector for a more dynamic and demanding future.
