EPRI’s Summer Seminar tackles the historic challenges and opportunities of AI and the Future Energy Landscape
Few, if any, people realized it at the time, but the future course of the electric power industry was a topic at a technical conference held in Long Beach, California, at the end of 2017. The Neural Information Processing Systems (NIPS) conference was the venue for eight Google researchers to discuss their paper, “Attention Is All You Need.”
What could a vaguely named paper discussed at a highly technical conference possibly have to do with the future of utilities and electricity? The short answer: a lot. The paper outlined the next step beyond the Google founders’ original idea to index every word on the World Wide Web and create an online yellow book. “What that paper said is you really can’t just look at every word. You have to look at how every word is used in a sentence,” said Arshad Mansoor, president and CEO of EPRI, at the kickoff to this year’s EPRI Summer Seminar, held in Coronado, California, on August 11 and 12. “What’s the context? How does that word relate to the paragraph, how does that word relate to the page, how does that word relate to the whole book?”
Put more simply, the paper described how large language models (LLMs) work. These models are a key subset of artificial intelligence (AI), trained on massive amounts of text to generate and interpret natural language. Today, the connection between AI and electricity is no mystery. “It all takes compute time. In the U.S., we are hearing 50-gigawatt, 100-gigawatt numbers thrown around. I was in the UK, and they say they have 60 gigawatts in the queue,” Mansoor said. “Knowledge is power. Now, power is creating knowledge, but that power has a different meaning. It’s electricity. It’s energy.”
Forecasts abound about the hefty power demands of data centers needed to train large language models, enable AI applications, and run cloud computing operations. For example, EPRI recently teamed up with the research institute Epoch AI to investigate the power needed to train large-scale AI models. Their findings: the power demands for training models have more than doubled over the past decade, and by 2030, it could take more than four gigawatts to train leading models—the equivalent of the power demand of millions of U.S. homes.
AI’s Rising Power Demands Transform Energy Planning
These are big numbers with short timelines. But utilities can take some solace in knowing this is not the first time the industry has had to meet skyrocketing demand for electricity. “This is what Con Edison’s transmission and distribution substation build looked like over the last 70 years,” Mansoor told the Summer Seminar audience. “100 gigawatts seems like a lot. But we have done it.”
How the industry meets rapid load growth today and over the next few years will obviously differ from the past and will include AI-driven power solutions. One reason is that continuous advances in the specialized computer chips inside data centers will require ever more power. “The chips are changing every three to five years. A rack [of chips] that used to be 30 kilowatts is now going to 1.2 megawatts,” Mansoor said. “If you designed your distribution substation for 400 megawatts, did you think you might need 600 after five years? That’s the type of different thinking we will need to do.”
AI and Onshoring Drive Electric Sector Power Growth
This year’s Summer Seminar fostered the fresh thinking and innovation utilities needed to surmount the pressing challenges posed by AI-driven load growth and seize the many opportunities it presents. Since the 1970s, EPRI’s annual Summer Seminar has provided an opportunity for company executives, policymakers, and leaders from academia, NGOs, government, and industry to meet and discuss pressing issues facing the sector. Summer Seminar is a critical part of EPRI’s role as an industry convener, helping to foster collaboration and knowledge sharing about near-term actions and to set a longer-term research and development (R&D) agenda.
Given the sprawling and rapidly changing nature of AI and the critical nexus between the AI and electric power industries, this year’s Summer Seminar discussions were necessarily wide-ranging. For example, after Mansoor’s opening remarks, he also moderated a conversation between the CEOs of Constellation, Nebraska Public Power District, and PPL Corporation. The session, “From Insight to Impact: Setting the Stage for Action,” provided a C-suite perspective on opportunities utilities have to fully harness AI to create a more resilient and cost-effective AI-enabled energy system.
Valuable AI tools can support electricity companies in a range of scenarios, including predictive maintenance, real-time monitoring, and optimized dispatch. The CEOs also offered their views about how EPRI’s global network can drive the innovation crucial to meet soaring demand from AI-enabled infrastructure. They also outlined tangible steps EPRI, its members, and industry stakeholders can take to quickly deliver scalable and affordable solutions. After the panel discussion, Gil Quiniones, the president and CEO of Commonwealth Edison (ComEd), offered a company-specific perspective about how ComEd is planning, investing, and innovating to meet demand from not only AI but also electrification and a resurgence of industrial activity.
This was a theme Mansoor emphasized to the audience. While data centers and AI grab daily headlines, a series of recent announcements underscores a trend in onshoring that will also drive load growth. Just a few days before Summer Seminar began, for instance, Apple announced an additional $100 billion commitment to domestic manufacturing, bringing its total planned investment to $600 billion over the next four years. One day after the Summer Seminar, GE Appliances announced a five-year, $3 billion-plus investment in U.S. operations, including plants in Kentucky, South Carolina, Georgia, Alabama, and Tennessee.
“There are 50 gigawatts of power that is not considered in most of our IRPs (integrated resource plans) in the U.S. that will come from reindustrialization,” Mansoor said. “Globally, other countries are also seeing this. So, data centers are huge, but they’re not alone.”
Innovation and Flexibility are the New Imperatives
Speed and innovation were also prevalent themes in Summer Seminar panels. Though the electricity industry has indeed met soaring demand before, load growth is driven by data centers, reindustrialization, and electrification, and it is on a much quicker timeline. Moving quickly depends on robust supply chains. The session, “Navigating Bottlenecks: Solutions for Data Center and Power System Supply Chains,” dug into the effects of growing global supply chain pressures. CEOs from Omaha Public Power District, Vistra, GE Vernova, and Quanta Services addressed bottlenecks and resilience strategies, pointing to the need for innovation and forward-looking procurement.
Innovation must infuse everything utilities and their partners do to meet surging demand. Pacific Gas & Electric (PG&E) CEO Patti Poppe led a session titled “Meeting the Moment: Planning for the Current Wave of Load Growth,” which underscored the industry-wide innovation imperative from grid modernization to transmission expansion to public-private partnerships.
One area that connects innovation and the need to move quickly is grid flexibility. Simply put, maximizing the existing grid helps contain costs for electricity customers while also providing time for the critical capacity additions needed to meet growing demand. The session, “From Concept to Reality: Demonstrating Data Center Flexibility in Production Environments,” shared lessons from EPRI’s DCFlex initiative, which is exploring how data centers can function as flexible grid resources. Already, results from DCFlex show potential for computational flexibility, ancillary services, and the use of renewable fuels in backup generation.
“We have to build smart—and fast. The need for speed in every country means asking what you can do in the next 12 to 18 months,” Mansoor said. “Much of that will come from unlocking headroom in the existing grid, through flexibility and load curtailment,” he said.
AI is Part of the Power Solution
Understandably, utilities, data center developers, hyperscalers, and others are all focused on how to meet AI-driven demand. However, AI itself can also be a tool to help meet data center demand. In the session, “Empowering Power: Time for AI to Return the Favor,” Jeremy Renshaw, EPRI’s executive director of AI and Quantum, explored how AI can be a tool for utilities to unlock new levels of system performance.
It’s a theme Mansoor also emphasized. “If you look at what the one holy grail that AI can really help, it’s the orchestration of flexibility,” he said. “We can get flexibility from markets, EV charging, batteries, and demand response.” Where EPRI can play a role, he said, is in standardizing how devices communicate.
Summer Seminar concluded with the session, “From Insight to Impact: Bringing it All Together,” a plenary featuring Mansoor alongside leaders from Southern Company, the Nuclear Energy Institute (NEI), and the New York Power Authority (NYPA). The discussion underscored how accelerating innovation, convening global collaboration, and balancing investment priorities are all essential to managing AI-driven load growth while keeping electricity reliable and affordable.
If a single research paper could reshape how machines understand language, the worthwhile outcome of the Summer Seminar is to spark meaningful research, collaboration, and—most importantly—action to accelerate progress in meeting the growing demand for electricity. These conversations will continue and evolve, all moving towards the same goal. “We have to be ready to power this demand,” Mansoor said.
