AI Empowers the Power Industry: Nvidia, IBM and Other Tech Giants Advance Applications, Regulatory and Data Privacy Challenges Remain

For a long time, digital technology suppliers have been frustrated by electric utilities' reliability-first approach that suppresses innovation. However, rapid load growth and the potential of AI solutions are driving the two sides to cooperate, especially in the use and sharing of power system data.

Electric utilities have seen thesignificant load growth expectationsbrought by electrification of transportation, manufacturing, buildings, and data centers. Executives at technology companies such as Nvidia, Microsoft, IBM, and Schneider Electric say they are gradually understanding the regulatory barriers that constrain utilities from transitioning to advanced AI computing strategies.

The dramatically increased computing power and hyperscale cloud resources of digital technology are upending previous assumptions about AI's potential to learn, execute, and optimize system operations, putting electric utilities on a so-called "technology transformation" path.

"The power industry is relatively conservative, but faces hard requirements for clean energy and emissions reduction," said Marc Spieler, senior managing director of energy business at Nvidia, the microprocessor market leader. Advanced computing's "real-time predictions can optimize" batch scheduling, maintenance, and other decisions, and Nvidia's "specialized modules will apply learnings from other industries to the energy transition."

Analysts point out that electric utilities have alreadyapplied advanced computing in wildfire prevention, vegetation management, and predictive maintenanceand are beginning to consider its potential in optimizing system dispatch.

"The tech industry's business model is 'move fast and break things,' and they don't always understand why regulated utilities can't move just as fast," said Emily Sanford Fisher, general counsel, corporate secretary, and executive vice president of clean energy at the Edison Electric Institute (EEI). "But there seems to be a new spirit of cooperation emerging in addressing utility challenges."

One obstacle could slow this technology transformation. Technology companies have seen how advanced computing innovation in other industries maximizes AI capabilities, but electric utilities still need to validate its potential and convince regulators that investing in advanced computing capabilities is justified.

Challenges and Potential

The rapid growth in load is evident.

"Over the past year, grid planners nearly doubled their 5-year load growth forecast, from 2.6% to 4.7%,"a December 2023 Grid Strategies research reportshows. The report adds that 2024 forecasts "may show even higher national growth rates," driven by new manufacturing, industrial, and data center investment.

The broad application of advanced computing is providing "decision support for bulk system operators" in managing new load, said Jeremy Renshaw, senior technical leader for AI, quantum, and innovation at the Electric Power Research Institute (EPRI). But fully optimizing distribution system operations and dispatch "could be a breakthrough, or it could still be years away," Renshaw added.

Many technology companies are investing in advanced computing to achieve this breakthrough.

Technology Companies Step Up

Advanced computing is beginning to serve electric utilities.

BrightNight's PowerAlpha platform helps design, operate, and optimize clean energy projects, significantly improving capacity factors and reducing costs, said Kiran Kumaraswamy, chief technology officer at BrightNight. Its machine learning and AI algorithms focus on utility-scale assets but do not yet have "the granularity to optimize the distribution system," he said.

Machine learning-based weather forecasting enables Amperon customers to purchase lower-priced electricity ahead of extreme events,rather than facing high scarcity pricing,said Sean Kelly, CEO and co-founder of Amperon. "Humans cannot calculate hourly updates of 15-day forecasts, or weekly long-range forecasts covering30,000 locations and 28 weather variables,"he added.

"Whether you call it machine learning or AI," thecomputerized replica of Southern California Edison's system built by Nearawill "apply more variables than humans can integrate,"said Rob Brook, senior vice president and managing director of the Americas at Neara. It "identifies ways to improve wildfire prevention" and "eliminates human error and cost," he added.

Although current AI applications in the power industry are mostly focused on bulk systems and maintenance, one advanced computing-based company appears poised to achieve a breakthrough at the distribution system level.

"Utilidata is deploying the first distribution system AI platform," using "customized Nvidia modules," said Jess Melanson, president and chief operating officer of Utilidata. The platform is being installed alongsideAclara meters,but "will ultimately be used in other system hardware such as transformers," he added.

Nvidia and Utilidata both see "tremendous opportunity" in power system applications, Melanson said. "Previously, Utilidata's analysis used incomplete, outdated, or poor-quality data," but world-standard Nvidia chips enable Utilidata'sKarman platform analysisto detail "what is happening in the system, what is likely to happen next, and the best response," he added.

This level of intelligence at the distribution system level could enable customer-owned resources to play a greater role in reliability and lower overall customer costs.

A key potential obstacle to realizing the benefits of advanced computing is the limited access to proprietary utility and customer data. Companies in the advanced computing space such as Nvidia and IBM place great importance on this.

Foundation Models and Federated Learning

Nvidia's software enables advanced computing to "predict data patterns and identify the next best action," said Nvidia's Spieler. For example, it can give utilities deeper insight into where outages are likely to occur and enable proactive maintenance, multiple analysts said.

The reluctance of utilities to share proprietary data needed to develop a more granular understanding of the grid is a real challenge, "but federated learning—used in healthcare to protect patient data—can be a solution," Spieler said. "With federated learning, collaborators can build their data models and share them in a central location," he added.

NVIDIA FLAREis a federated learning software tool that "builds additional synthetic data to address new problems," thereby addressing privacy concerns, Spieler said.

"Utilities value data security" and must ensure sharing is done in the "right way," said EEI's Fisher. "There must be protocols to protect critical energy infrastructure information," although "constructive conversations about how to cooperate on this are good news," she added.

Some argue that federated learning may be too limited for the complexity of the unique characteristics and resource mixes of different parts of the power system.

"Foundation models are emerging to scale advanced computing capabilities," said Bryan Sacks, global chief technology officer and leader of energy, environment, and utilities solutions at IBM.

Unlike applying AI components to individual problems, "an order of magnitude more data is pre-trained into foundation models to solve multiple problems," Sacks said. Foundation models can capture the diversity of the power system without disclosing any operator's proprietary details. But "the metering and monitoring system data utilities use for operational decisions is not in current large language models and is protected from external sharing," he added.

"For a foundation model to understand power system operations, it needs specific time and location data for every connected asset, but that data must be anonymized," Sacks said. "IBM has launched working groups to build foundation models trained on anonymized data for real-time, day-ahead operations, and long-term planning of power systems," he added.

Different power system stakeholders will be able to "fine-tune the foundation model to address their respective different problems," Sacks said. But "regulatory hurdles or restrictions on market participants' access to data needed to train models are real concerns," he added.

According to Sacks, IBM is inviting global stakeholders to contribute foundational research and will work with regulators "to help establish governance systems that promote data sharing," as well as build effective "guardrails" to protect systems and data.

IBM's awareness of regulatory issues aligns it with what utilities cite as their primary concern.

Utilities and Regulators

Utilities appear to view the regulatory obstacles to advanced computing more realistically than technology companies.

Using advanced computing to optimize distribution systems in real time requires utilities and regulators to have "sufficient confidence" in it, said Steve Smith, head of strategy, innovation, and market analysis at National Grid and president of National Grid Partners, the corporate venture capital fund. "We may get there in 10 or 15 years," he added.

"Technology companies and utilities have fundamentally different business models," and "tech companies don't understand why building a new transmission line takes 10 years," EEI's Fisher added. In fact, regulated utilities "can build a transmission line in just 18 to 24 months, but siting, permitting, and litigation take eight years," she said.

Introducing advanced computing requires providing "concrete evidence" to utilities and regulators that "necessary grid modernization spending will lower customer costs," Fisher said.

"Foundation models have enormous potential for the energy industry," and data federation is "an absolute necessity for extracting and coordinating data from different siloed power system data," said Scott Harden, global chief technology officer for innovation at Schneider Electric.

The ideal power system architecture should be built on a power industry foundation model that captures the key characteristics of its diversity and challenges, Harden said. It should also deploy more broadlyphasor measurement units—hardware devices that can record and transmit transmission and distribution system data—and "comprehensive deployment of smart meters at the edge of the system, with all data federated," he added.

Building that architecture could start with regulators supporting "new computing capabilities and the technologies needed to realize them," Harden said. The cost and time of deployment are not yet clear, "but the more important question is what the cost of not deploying will be," he added.

"The broad field of AI is still early for everyone, and the question now is how to navigate it," Federal Energy Regulatory Commission (FERC) Commissioner Allison Clements told Utility Dive. The advanced computing currently being applied "could create a positive feedback loop if federal and state regulators push for it," she added.

"Regulators must maintain a growth mindset at this moment of transformation, because this is the early stage of the 'messy middle' of grid modernization," Clements said. Utilities "are handling rate cases case by case to understand how to achieve this transition while protecting reliability and affordability," she added.

"Federal and state regulators need to be actively engaged because AI capabilities are coming," Clements added. "Whether it benefits society or creates problems depends on utilities, policymakers, legislators, and other leaders."