WASHINGTON ― The biggest roadblock to the clean energy transition now underway in the U.S. is not technology-related or even the anticipated spike in power demand from data centers, according to speakers at a conference on the energy transition June 12. 

It is the engrained, slow and risk-averse culture of U.S. utilities and other private-sector players, they said. 

The technologies are ready, said Jigar Shah, director of the Department of Energy’s Loan Programs Office, at the Clean Energy Transition Conference, held at the National Press Club by Tech for Climate Action, a UK-based event organizer.  

“Now we need to get utilities to act like private-sector companies and actually take risk. You see that in the stock market. … The utilities’ stock prices have [gone up] in the anticipation that they’re going to turn from dividend companies into growth companies,” Shah said during an on-stage conversation with Mary de Wysocki, chief sustainability officer for Cisco. “So, figuring out how that cultural and norm thing occurs is really fascinating to watch.” 

DOE is providing technical assistance “helping a lot of those folks through that change,” Shah said. 

Marissa Hummon, chief technology officer of Utilidata, a company developing grid-edge artificial intelligence applications, agreed that a major obstacle for her company is “getting the distribution utilities to act very differently than they have in the past.” 

The energy transition “is going to happen whether or not utilities decide to step up,” Hummon said during a panel on the role of AI in the energy industry. “There will be new energy demands on the system, but the distribution utilities could really be encouraged to take that proactive step to deploy a platform that allows them to actually respond to the changes.” 

The Biden administration’s position has been that the energy transition will be private sector-led and government-enabled with the billions of federal tax credits, loans and other incentives from the Infrastructure Investment and Jobs Act and Inflation Reduction Act. But the message that emerged from the conference is that at least some parts of the private sector have “been caught flat-footed” when asked to lead, Shah said, especially in the face of rising electricity demand from data centers and AI. 

While the private sector is supposed to be the most efficient allocator of risk, “that process has been messy,” he said. “But I do think it’s a little bit unreasonable to believe the entire ecosystem has figured this whole thing out [in] less than two years” since the IRA was passed. 

Faced with rising demand from data centers, Shah said, the focus has been on the AI chips and servers, but “much of the rest of the data center actually uses the electricity, so figuring out how we make the system more efficient is the more difficult thing to do. … 

“We can’t actually decarbonize our processes by thinking the same way we thought about things 10 years ago. This is not just [about] buying carbon credits, figuring out direct air capture and doing everything exactly the same. This is about us reimagining how we actually still live a modern lifestyle but doing things with materials that are more sustainable; doing things with a more thoughtful approach.” 

The electric power system must be able to “flex” load with the “same level of dexterity that we currently only flex supply,” he said. 

“When you think about what it’s going to take to really meet this moment, it was something we actually needed to do in 2000, but people weren’t forced,” he said. “When you’re a monopoly, obviously, you have a tendency not to deploy innovation as fast as a more vibrant capital system, and so we’re doing it now because the pressures are just so great from weather and load growth.” 

A similar sense of urgency should be used to create new narratives about AI, said Charles Yang, policy adviser at DOE’s Office of Critical and Emerging Technologies. 

The challenge of load growth from data centers can be converted, not into new natural gas plants, but “into building an order book for the next generation of clean, firm, advanced technologies,” Yang said, pointing to Microsoft, Google and Nucor’s recently announced plan for aggregating their demand and contracting for clean power. 

“We need better stories about what AI can do,” he said. “How it can help us discover more abundant, affordable batteries; how it can help us coordinate our EV charging and lower costs for ratepayers. These are the stories that we haven’t really told; they’re not the future we’ve been told about.” 

Moving AI to Grid Edge

Since ChatGPT was introduced in November 2022, AI has exploded in the public consciousness, but, Hummon said, “Utilidata has been running AI models to operate the grid for more than 12 years. … We’ve been using data-driven, real-time methods to create outcomes of a more efficient, powerful grid, very reliably.” 

What’s changed is the emergence of “generative AI” and the creation of large language models (LLMs) that allow users to ask questions or “prompt” the software in plain language.  

Utilidata is deploying these advanced technologies to move AI to the grid edge, improving system visibility and opportunities for more efficient operations for distribution system operators, Hummon said. Such systems could not only get “the right information back to a central system to make a better decision, but also … interface with the customer using natural language about their energy use, about their choices, about just what sort of resources they want to purchase,” she said. 

AI can also support better use of unused capacity on the grid to increase the power that can be sent down distribution lines without having to build new substations or feeders, she said. Optimizing the operation of a substation with traditional, physics-based calculations can take 12 to 18 months, Hummon said. 

“If you’re using data-driven, machine-learning methods, you can be up and running in two weeks,” she said. 

Claus Daniel, associate laboratory director at DOE’s Argonne National Laboratory, said his researchers and scientists want to push the use of AI in the electric power system further “to figure out how we can use that technology to help us in research and development to find better ways of utilizing energy; better ways of generating energy. … 

“Artificial intelligence is particularly well suited to figure out what are the tradeoffs … and what are the connections. It’s particularly well suited for handling complexity and recognizing patterns that we currently cannot fully resolve when we just use high-performance computing and physics-based models.” 

DOE and the National Labs are currently working with their Frontier and Aurora supercomputers ― the largest computers in the world ― to create “reliable and safe large language models,” Daniel said, noting that most publicly available LLMs often answer questions with convincing but completely wrong information. 

Eelco de Jong, head of AI-enabled utility service at McKinsey & Co., zeroed in on how AI can be used to “more precisely allocate our capital towards the investments that have the highest return for [grid] reliability.” 

Instead of replacing equipment based on age or zip code, “we’re seeing companies using granular data to forecast, for example, which households are most likely to adopt electric vehicles or heat pumps or switch from gas to electric,” de Jong said. “And based on that forecast data, we know exactly which neighborhoods or even which feeders are going to first run out of capacity, and we can channel … our capital dollars to that.” 

Similarly, AI can help with stressed supply chains by routing equipment “to the places where [it has] the biggest impact on customer reliability,” he said. 

DOE’s recent AI for Energy report, released in April, focuses on advancing the intelligence of the grid, Daniel said. (See AI Critical to US Clean Energy, Grid Modernization Goals.) 

“This is something that will fundamentally change how we operate the grid” and help solve the problem of non-dispatchable wind and solar, Daniel said. “If I manage through building controls, through heating and cooling needs … [to understand] what’s happening on the edge, I can control my demand in a better way. I can live with a higher percentage of non-dispatchable generation.” 

AI integrated into thousands of devices on the grid edge could also make the system more resistant to cyberattacks, Hummon said. 

“If the edge is intelligent in and of itself, then every individual endpoint can make its own separate decision,” she said. “You’d have to hack all those separate decisions in order to create the same type of risk that with pure central decision-making.”