AI in the Power Sector

The New York Public Service Commission has launched an inquiry into public utility use of artificial intelligence, noting that AI systems “create risk for organizations” but are also increasingly used to perform basic tasks.

The Sept. 17 order requires electric, gas and water utilities to respond within 60 days with “a written inventory report that describes all use cases of AI systems in their operations.”

Consolidated Edison, which serves New York City, said it uses AI for a range of uses, including customer service and system inspections. “As technology evolves, we will continue to evaluate opportunities,” a spokesperson said in an email to Utility Dive.

Public utilities are increasingly using AI systems for day-to-day operations, and New York regulators are concerned they are “susceptible to hallucinations, algorithmic biases, transparency issues, data privacy concerns, misconfiguration errors, cybersecurity attacks, and functional brittleness.”

“These potential risks may have negative consequences for the safety and reliability of New York’s critical infrastructure,” they said.

The PSC’s order directs utilities to file “a complete inventory” of all AI use cases in their operations and “disclose their policies, procedures, and protocols” around the burgeoning technology. The commission said it will “evaluate those procedures and protocols for adequacy and robustness against commonly accepted AI Governance frameworks.”

The commission’s order defines an “AI system” as a machine-based system that “can, for a given set of human-defined objectives, make predictions, recommendations, or decisions influencing real or virtual environments.”

New York utilities are already utilizing AI systems in various ways, including model outage predictions and electric usage. National Grid has deployed a system known as GridCARE to help free up interconnection capacity for large-load customers, regulators pointed out. The New York Power Authority has used AI to analyze drone-captured data for enhanced vegetation management.

Con Edison told Utility Dive the utility is using AI “to improve customer service, proactively identify potential equipment issues before they affect customers or public safety, and to strengthen our inspection and mapping capabilities.”

But more broadly, regulators noted “the actual extent and scope of AI systems in New York utility operations remains unclear.”

“The rapid evolution of AI is a double-edged sword for utilities,” PSC Chair Rory Christian said in a statement. “While AI can be a powerful tool that can result in cost efficiencies and improved operations, it also presents serious risks that must be evaluated and addressed appropriately.”

Exelon’s climate investment arm has become a minority investor in artificial intelligence-powered infrastructure and transmission planning company Continuum Industries, the companies announced.

Exelon, one of the largest investor-owned utilities in the U.S., has added the company to its Climate Change Investment Initiative portfolio and cohort, according to a Sept. 16 press release. The fund, also known as 2c2i, does not disclose its investment amounts, but typically invests at least $100,000 and up to $300,000 in equity investments to fill funding gaps for cohort companies, 2c2i Lead Faith Davis told ESG Dive.

Exelon launched the climate investment arm in 2019, which has a $20 million remit to support startups “that catalyze solutions to climate change challenges within Exelon’s major market areas,” according to the utility giant’s website. Companies backed are also expected to advance social equality and economic prosperity in Exelon’s markets.

Continuum launched in 2018 and has built out its “Optioneer” platform since, which helps companies plan grid connections; site plan; do automated environmental risk and land impact analysis; and map the risk, constraints and impacts of projects, according to its website. In addition to the investment, Continuum will also have its Optioneer platform deployed across Exelon’s family of companies, according to a Sept. 16 company blog.

Continuum and 2c2i first connected in December 2025. Davis said in an exclusive interview with ESG Dive earlier this month the fund was drawn to Continuum’s application by the “growth and opportunity” in the startup’s market, as well as the confidence and technical expertise of the founding team and Continuum’s customers and project pipeline.

“We're seeing a lot more focus around affordability and efficiency, and technologies that help the utility be more efficient in terms of building infrastructure or running our existing infrastructure more effectively is a space that we're looking at,” Davis said. “The transmission planning space is an area where — given all the demand that we're seeing for load and things of that nature — we're looking for ways to make the process and the planning quicker and more efficient.”

As a member of the 2c2i cohort, Continuum will have an Exelon liaison work with the company to help it scale, Davis said. The Optioneer platform is expected to be deployed across Exelon’s utility companies, which operate in the PJM interconnection transmission region across five states and the District of Columbia.

"As demand for electricity continues to rise, utilities and developers need a repeatable and consistent approach to planning and delivering infrastructure predictably and efficiently," Continuum CEO Grzegorz Marecki said in Exelon’s release.

Continuum said it has used its platform to screen more than 420,000 miles of infrastructure alternatives and helped developers find the “lowest-impact and fastest-to-permit solutions.” Continuum said it will work with Exelon to identify and prioritize where to deploy the platform across its system.

Companies 2c2i invested in have subsequently raised more than $555 million in follow-on funding, created more than 3,558 jobs, generated more than $548 million in revenue and led to 1.2 million metric tons in carbon dioxide efficient avoided or removed, Exelon said in its latest sustainability report released in August. 

Exelon announced last month that it added sustainable energy and grid solution companies Public Grid and Backstop to its 2c2i cohort. The climate investment fund’s current round of applications are due by Sept. 27.

Sponsored content

By Kraken Technologies

For decades, utility customer teams have had too much to deal with. Customer operations have been designed to manage demand and drive down contacts. But as service becomes more demanding, this model is creaking under the weight.

At the same time, leading utilities are transforming their operations with holistic, next-generation platforms, providing teams with powerful tools and dismantling siloes to unlock smarter ways of working. Now, AI is creating a genuine opportunity to rethink the role of customer operations altogether, proving that embracing positive touchpoints and better engagement doesn’t just drive better customer outcomes, but better business outcomes too.

Legacy operations models are under immense strain

For 30 years, customer operations have been designed to handle volume. When demand grows, hire more people; when costs rise, manage them down. Contact centers exist to handle interactions – not deliver outcomes. At the heart of this model was the phone-heavy ‘transfer model’. Customers were taken in by a contact center. If a query crossed the boundaries of a particular siloed back office team (which most queries do), the customer was bounced between teams, which often meant lost context and frustrating wait times. Siloed tech reflected and preserved this siloed operational structure, with team members struggling between precariously cobbled together legacy platforms.

The next stage came with the arrival of web portals and IVR. The industry celebrated digital transformation, but the underlying model didn't change – the goal was still to manage volume, not to serve customers better and support broader business outcomes. Customers called anyway, having spent twenty minutes failing first. Today in an era of increasing customer costs, customer contacts, burgeoning bad debt and complex DER customers, the transfer model is buckling.

Universal agents and the ownership era

Today, fortunately, we’re beginning to see a new model for engaging customers emerge, enabled by new utility tech. Next-generation software such as Kraken – is putting every aspect of utility operations – customer comms, billing, metering, field service and more – at team members’ fingertips, seamlessly connecting them in one intuitive platform. Well-formatted, silo-defying data and powerful tools are shifting ways of working, empowering ‘universal agents’, able to quickly and positively address a wide range of queries.

Yet tech is not a magic bullet. The real shift transforming customer experience is emerging as modern digital platforms are used to empower accountability and high-quality decision making. Someone who owns your problem until it's solved. Building this has lent itself to a different kind of organization: flatter, faster, more trusted, more efficient, creating teams with judgment. As utility operations enter a new era, this is essential.

New tech has given leading utilities the opportunity to reconceptualize the fundamental role and value in customer operations. Most organisations still measure customer operations in contacts deflected, calls contained, cost per contact driven down. But when you make every interaction easier, faster and higher quality, interactions actually tend to rise – you cannot deflect your way to value. The key question therefore shifts from "how do we reduce interactions?" to "how do we maximise the positive impact of every one — and create more?"

And just what outcomes can utilities deliver? Better service, lower costs, higher scores, yes. But consider that this operational revolution is also happening at a transformative moment for energy, with increasingly complex customer needs and rising grid balancing and infrastructure costs. Utilities increasingly need customers to do things: manage affordability, respond to personalized energy advice, shift consumption, choose products, stay engaged.

Take the adoption of consumer DERs. These customers don’t just tend to be more switched on with higher expectations. Properly catering for them with flexibility programs and bespoke products unlocks new revenue streams and opportunities to keep grid costs down – but only if these customers are properly engaged. This shifts operational transformation beyond serving customers, serving broader business aims.   

The era of impact: The next step for utility operations

Today, AI is creating unprecedented opportunities for deeper operational transformation. The rich, holistic data foundations of modern platforms, originally built to dismantle operational siloes and bring utility operations into the 21st century are also perfectly suited for integrating safe, versatile, impactful AI into utility operations. AI is already summarizing messages, pulling relevant information, guiding agents, automatically answering customer queries and, critically, anticipating customer needs.

Building on the ownership model remains vital. Without the ‘wisdom’ afforded by models that encourage accountability and good decisions, there is every chance that AI will simply enable utilities to do more that doesn’t work. Accountable team members must be there to guide and direct AI and address more complex queries with the powerful tools that AI provides.

In this paradigm, AI is not merely IT, it is operational capacity. It's not there to automate the contact center. Instead, it exists to fix problems before they even become calls, offer solutions before they're requested and identify opportunities before customers request them. The real measure is outcomes delivered. 

Rising to the challenge – embracing the era of impact – does require changes to the way that utilities have thought about their operations for decades. The benefits, however, are clear. Lower costs. Happier customers. But more than that – an opportunity to turn customer operations from a souped-up contact center into an integrated engine that creates value across the business.

South Carolina’s public power and water utility Santee Cooper announced that it is working with Google to modernize its financial planning and load forecasting with artificial intelligence tools. 

Google’s WeatherNext 2 is designed to provide more accurate medium-range weather forecasts, which is expected to improve Santee Cooper’s load forecasting, the utility’s CFO, Tami Wilson, said during a Sept. 2 press teleconference. The utility is also using Google’s Gemini Enterprise Agent Platform to speed up financial forecasting, she said.

Wilson said the utility is undergoing a “major digital transformation” by “shifting away from legacy, clunky, cumbersome processes,” and wants to “automate the routine” without automating accountability or risk management, which she said will remain with human operators.

Wilson said the utility plans to use different approaches with generative AI versus agentic AI. Generative AI will be used to perform “low-cost initial drafting,” while Santee Cooper will use agentic AI to “execute multi-step processes autonomously.”

Santee Cooper declined to provide details about what processes it intends to automate with agentic AI, or the cost of implementing this new software.

At the teleconference, Raiford Smith, global director of power and energy industry at Google Cloud, said Gemini Enterprise was designed “specifically to serve as that intelligent operating layer for modern utilities, delivering at least three distinct capabilities that are built for complex infrastructure.”

WeatherNext 2 is a weather modeling technology that delivers probabilistic forecasts across lead times of up to 15 days, according to a Google press release.

Wilson said Santee Cooper has previously relied on traditional meteorological models to predict power load, but WeatherNext 2 could “dramatically improve” the utility’s forecasting ability, helping it “save millions of dollars a year.” 

She noted that Google is helping the utility build a custom forecasting model that will “build in consideration for unique weather factors in our service footprint,” such as Lake Marion and Lake Moultrie — two man-made lakes large enough to create their own microclimate.

“A one-degree variance in the forecast can instantly swing our power to about 100 MW an hour, and that is about the equivalent of powering tens of thousands of homes,” Wilson said. “So if we're caught short on a freezing day in the winter or a scorching afternoon, it could cost us up to $100,000 an hour to go out and procure emergency power.”

Smith said on the call that the difference between WeatherNext 2 and traditional meteorological models is that the former is inference-based while the latter is physics-based. 

Physics-based forecasts do “a pretty good job looking out 24 to 48 hours,” he said, but after that, the math becomes so complicated that “the forecast accuracy just tails off significantly,” which can impede medium-range forecasts. 

Santee Cooper is also using Google AI tools to “enable rapid, scenario-based financial planning that directly feeds into the calculations for major projects, including Santee Cooper’s $10 billion grid-expansion budget,” Google said in a release.

Additionally, the utility is deploying “several hundred Gemini Enterprise licenses” across the organization in an attempt to modernize its workflow, Wilson said.

Wilson said the utility is taking care in its rollout of AI tools to ensure that participating employees receive proper training for the handling of confidential and restricted data.

With the U.S. facing an electricity affordability crisis and demand rising quickly for the first time in decades, experts say the Federal Energy Regulatory Commission’s June 18 order to system operators to provide transmission for flexible large loads underscores the urgent need to implement data center load flexibility.

Just a 1% to 2% reduction in data center peak demand can reduce electricity rates 0.5% to 2.8% and protect reliability, according to a 2026 Duke University Nicholas Institute study. And pilots and analysis led by the Electric Power Research Institute, or EPRI, show how flexibility is within reach that can also get data centers interconnected faster.

Electricity demand by artificial intelligence, or AI, data centers is driving a global urgency, Nat Bullard, chief strategy officer of research group Halcyon, reported in May. From Q1 2025 to Q1 2026, Amazon Web Services’ cloud business grew 28%, Microsoft Azure grew 40% and Google Cloud revenues increased 63%, he reported.

“Growth rates this high in already-mature businesses mean total revenue doubles in two years (or less),” Bullard wrote. “That revenue can only be serviced with compute, and that compute can only serve when energized.”

Experts told Utility Dive that public opposition to data centers, focused on local electricity costs, can be addressed by flexibility. Data center flexibility can reduce electricity demand, which reduces the costs to ratepayers of investments to protect reliability, the North American Electric Reliability Corp. acknowledged in its 2026 large load risk mitigation guidelines.

Flexibility can get a data center interconnected faster, make it a benefit to system reliability and increase system utilization which lowers rates, utilities, researchers and analysts agree. The challenge remaining is to resolve control issues between risk-averse utilities and impatient data center operators, they also agree.

The flexibility framework

Power demand from U.S. data centers will reach 66 GW in 2027, up from 31 GW in 2025, with summer peaks growing to 8.5% in 2027 from 4.1% in 2025, Goldman Sachs reported May 20. Though AI proliferation is uncertain and growth will vary widely by state, data centers could consume as much as 17% of U.S. electricity usage by 2030, EPRI found.

Flexibility adds “headroom” to a power system by allowing system operators to add large loads and maintain reliability while minimizing rate raising infrastructure investments, EPRI’s FlexMosaic framework concluded. That is why data centers willing to be flexible can also maximize their “speed to power,” the institute’s analysis found.

There are five flexibility “classes,” according to EPRI. Class A supports power systems during “infrequent extreme stresses” and Class B manages “daily or weekly” demand peaks. Class C meets long-lasting energy shortages, Class D protects in sudden supply or demand swings and Class E provides frequency stabilizing.

Key determinants of the flexibility class of a data center are the notification time it needs, the duration and frequency of its response, and the depth and quickness of its flexibility, FlexMosaic said.

FlexMosaic’s objective is to align incentives of data centers and utilities by linking faster and bigger data center interconnections with contractual agreements that protect utilities. Data centers with class D and E flexibilities that mitigate the local thermal overloads or voltage drops “unlock” the most system value, EPRI added.

To be flexible, data centers can combine three flexibility pillars — managed workloads, reduced AI plant energy consumption and back-up power, said Anuja Ratnayake, the emerging technologies executive leading EPRI’s DCFlex Initiative. FlexMosaic’s aim is to standardize data center designs and utility programs that include flexibility, she added.

EPRI’s work contradicts the assertion that data centers cannot be flexible, said Ann Rendahl, president of the National Association of Regulatory Utility Commissioners and a Washington state utilities commissioner.

The findings from EPRI’s pilots and analysis will allow state regulators to require data centers to take the potential for flexibility seriously, Rendahl added.

The flexibility breakthrough

Using onsite or stored power during system emergencies and reducing building electricity use at operator requests is not new for data centers and other large loads. But FlexMosaic’s third pillar, data center workload flexibility, is a new opportunity to protect reliability and affordability that is now within reach.

Today’s AI data centers can “be designed with flexibility as a core operational principle,” Boston University researchers reported in June. Training and inference workloads “can offer between 18% and 55% flexibility relative to their average power consumption” and still meet quality of service requirements, modeling found.

Emerald AI “operates as an orchestration and optimization layer between utilities and data centers” and “does not directly control utility or data center operations,” Emerald AI head of product Mansi Shah said. “Utilities retain full dispatch authority,” she added.

Emerald AI “enables utilities to issue curtailment or flexibility requests to participating data centers through a secure software interface,” Shah continued. The platform “translates grid requirements into operationally feasible dispatch targets and provides telemetry, verification, and event compliance reporting back to the utility operator,” she said.

In short, Emerald AI coordinates “approved flexibility actions,” Shah said. The actions address “compute, cooling, energy storage, backup generation, and power infrastructure,” but always respect “predefined operational guardrails,” she added.

Every utility-data center interaction “is governed by operational parameters agreed upon by both parties in advance,” Shah said.

The utility defines the parameters of an event requiring flexibility, including “maximum magnitude, minimum notice period, frequency limits, and event duration,” Shah continued. But data centers set the “hard floors that protect critical workloads and infrastructure under all circumstances,” she said.

These terms give utilities “confidence that contracted flexibility will perform reliably,” and assure data centers “that operational and [technical] boundaries will never be violated,” Shah added.

EPRI’s demonstrations have delivered flexibility of up to 40%, but future AI infrastructure with designed-in flexibility can “materially” expand that, Shah said.

EPRI, NVIDIA and utility demonstrations with Emerald AI software have shown that training workloads “can be paused or slowed,” and inference queries “could be redirected” to a data center on a less stressed system, NVIDIA said in a July 2025 blog post.

In a peer-reviewed test of data center workload flexibility, Emerald AI’s platform gradually ramped an AI workload at electric utility Salt River Project 25% during a three-hour peak, the release said.

Another Emerald AI test shifted a data center workload from Virginia to Illinois without a loss of service quality.

A UK test showed Emerald AI software using NVIDIA hardware could reduce an AI load over a third in under a minute while protecting critical compute.

Emerald AI sent Portland General Electric’s signals to an Oregon data center and reduced power 20% in simulated weather emergency scenarios, Emerald AI reported in March.

Emerald AI has now proven “temporal flexibility” by “slowing or pausing AI workloads” with “customer-designated flexibility on completion time,” the company reported in March. It has also proven “spatial flexibility” by “seamlessly rerouting latency-sensitive loads” from a power-constrained data center “halfway across the country” to where power is available, it added.

In late 2026, EPRI, NVIDIA, Emerald AI and their partners plan to bring the 96-MW Aurora AI Factory online in Manassas, Virginia. It will validate workload flexibility at scale in a data center “designed for flexibility,” EPRI’s Ratnayake said.

The needed agreement

The missing piece in EPRI's DCFlex plans is a standardized binding agreement between utilities and data centers, stakeholders said.

A key parameter of that agreement will be explored in the next EPRI-led pilot. Emerald AI will flex an NVIDIA data center served by Silicon Valley Power, or SVP, which serves 58 data centers in its 20 square mile territory, Chris Karwick, its chief operating officer, said.

For phase one of the pilot, Emerald AI is developing “a bidirectional communication platform with the data center,” Karwick said. SVP “will see the NVIDIA data center load in real time, send a signal to reduce the load in response to simulated events like a forecasted heat wave or a sudden lightening strike, and see the load reduced,” he added.

SVP must also have “the safety net of 100% control of the loadside breaker if the data center wants to have the faster interconnection and additional capacity,” Karwick said. “That is non-negotiable,” and would be part of the interconnection agreement SVP makes with the data center, Karwick said.

The Portland General Electric, or PGE, large load study process already identifies each new data center’s flexibility capabilities, said Isaac Barrow, the utility’s senior manager of data centers and growth. But traditional bill credit incentives are too small to guarantee AI data centers will reduce their lucrative workloads when the utility sends a signal through Emerald AI, he added.

An interconnection agreement offering the right incentive structure will, however, “unlock the next wave of flexibility technologies,” Barrow said. For PGE, an agreement providing accelerated interconnection for a data center must guarantee “visibility” and “dispatchability” of the data center’s load, he added.

New utility regulations could simplify the agreement process by requiring data centers to have flexibility capabilities, said Joe Reele, vice president, solution architects, for energy technology company Schneider Electric. But regulations could require consent to a single digital platform by the power and data center ecosystems, he acknowledged.

With 50 state regulatory jurisdictions and many other local jurisdictions, “it is not an easy solution” with so many proprietary software platforms in use, Reele said.

Will they agree?

The value of a definitive flexibility agreement between data centers and utilities is clear.

“Power availability has become the gating issue for AI,” said Mona Dajani, c-chair of multinational law firm Cooley’s infrastructure, energy and real estate group. In constrained power markets, projects that have “credible, measurable, and contractually defined flexibility may increasingly receive more favorable treatment,” she added.

But the “pain point” remains utilities’ need for control and data centers’ need to choose how to respond to a utility signal for a load reduction, said EPRI’s Ratnayake.

A control mechanism like the one SVP’s Karwick described “is a really hard concept for data centers to embrace because they lose control,” Ratnayake said. Data centers need a more gradual load reduction because their highly sensitive hardware would be at risk and it is probably worth much more than many power system assets, she added.

Utilities need control because their experience with demand response programs shows they cannot always depend on voluntarily load reductions, Ratnayake continued. And “if a GW scale data center goes offline instantaneously it would create a whole host of other system issues,” she said.

Ultimately, the data center and the utility need a program structure defined in a contractual interconnection agreement, Ratnayake said.

The April memorandum of understanding between EPRI and the Open Compute Project Foundation, or OCP, is a major step toward that agreement and those protocols, Ratnayake said. It creates a channel for EPRI utility members and OCP’s tech community to collaborate on standards and protocols that work for both, she added.

Validating technologies, regulatory structures and market certainty will take time, but “it will unlock the power needed to keep pace with needs of AI,” Ratnayake said

Meanwhile, “the reality today is if you want grid power at scale, you cannot have 100% of the hours,” said James Kacergis, senior vice president of corporate and business development for Terawulf, which is building the 750-MW Mariner Lake AI data center for anchor tenants Core42 and Google-backed Fluidstack.

“A hyperscaler’s primary incentive is getting the interconnection to scale the business, and if it is not willing to work with the utility or system operator, a competitor will,” Kacergis said. But “it is not yet clear what the amount of load reduction and duration is that will attract data center customers because it's an emerging market,” he added.

Data centers are capable of slowing, capping or shifting workloads, “but it's questionable whether they want to do it,” added Steven Carlini, chief advocate of AI and data centers for Schneider Electric. If, though, the choice is “having 80% of the data center working or none of it, they would take 80%,” he added.

“Data centers want power as soon as possible and they want that power to be reliable, and utilities want to bring them online as quickly as possible and protect reliability,” said NARUC’s Rendahl. “Flexibility is a way to do what both want.”

The rate of data breaches at companies that widely use AI tools is significantly higher than the rate at companies that don’t — 43% compared with 11% over the past 12 months — the identity security firm Netwrix said in a June report.

AI tools such as agents significantly increase organizations’ “identity footprint,” creating more gaps that hackers can exploit, Netwrix said.

At the same time, Netwrix found, the companies using AI the most widely are also the ones taking identity management the most seriously.

Netwrix’s report highlights the security risks of the sprawling web of user accounts and other identities that companies must create to use agents, copilots and other AI tools.

“AI agents are now acting on behalf of humans against sensitive data,” Netwrix researchers wrote. “Non-human identities need the same operational rigor long applied to privileged human access.”

And yet many companies aren’t taking identity management seriously, the report found. Roughly three-quarters lack “a single, unified view of sensitive data and which identities have access to it,” researchers said. More than half of organizations lack an up-to-date database of sensitive data, 71% can’t quickly determine which identities can access which data and 70% don’t have a security strategy linking data protection with identity governance.

Identity management is far from a new challenge for enterprises, but AI has magnified it, and companies are not always keeping pace. Three-quarters of organizations aren’t fully overseeing what AI identities are doing in their systems, even as 41% say they’re letting AI agents access sensitive data and perform vital tasks.

Netwrix’s report highlights how hackers have used identity security weaknesses as entry points in target networks. Three-quarters of incidents in which hackers access sensitive data involve compromises of identities or misconfigured account permissions. But despite widespread corporate awareness of this threat, most companies aren’t mitigating it.

Seventy-six percent of organizations can’t immediately revoke inactive accounts’ data access, according to the report, and 72% say their accounts have excessive permissions or they’re unsure which permissions their accounts have. Even more worrisome, roughly two-thirds of organizations said they believe at least some of their accounts have unnecessary access to vital data. Only one-quarter of companies said they were fully confident in their ability to detect potentially dangerous account access permissions.

The report — which also includes data about companies’ readiness to govern their AI systems and the frequency of unauthorized-identity-access incidents in different organization size segments — is based on a worldwide survey of 2,317 security professionals at 1,889 organizations in 60 industries.

The narrative by now is all too familiar: Deploy artificial intelligence, stabilize the grid and eliminate outages. Executives repeat it. Journalists amplify it. Vendors sell it. And that simplification is terribly wrong.

AI has rapidly evolved from a laboratory innovation to a vital component of decision-making in the modern energy industry. Electric utilities worldwide are turning to AI-driven forecasting, outage response and optimization systems to improve reliability and sustainability.

But the notion that simply deploying AI automatically guarantees better reliability is a misunderstanding that is costing utilities real money and real credibility. Reliable performance depends on how utilities manage data governance, workforce training and infrastructure modernization, not on which algorithm they purchase.

I have spent years in IT operations at Dominion Energy watching this play out in real time. The projects that succeeded were never the ones with the most sophisticated models. They were the ones embedded within enterprise strategy, supported by continuous feedback loops, and evaluated against operational metrics rather than abstract innovation goals — the ones that brought in frontline employees to understand the process on an intimate level. Projects that struggled often treated AI as a plug-and-play solution rather than a transformation requiring organizational commitment.

The evidence is real, but so are the conditions.

The empirical case for AI in energy is genuine. Utilities adopting AI-based forecasting tools have achieved notable accuracy gains and operational resilience. One Turkish utility replaced its conventional time-series forecasting model with a generative adversarial network approach, yielding sharper demand predictions, reduced overloads and fewer cascading failures.

But those improvements were not due solely to mathematics. They required managers who integrated model outputs into planning meetings and adjusted maintenance schedules accordingly.

On the outage management side, machine learning models using real-world utility data have outperformed baseline statistical methods by meaningful margins in predicting outage durations. More accurate estimates allow dispatch centers to prioritize repairs efficiently and communicate credible restoration times to customers.

In an era when reliability indices directly shape regulatory ratings, AI's predictive capability becomes a measurable operational asset, but only when people know how to use it.

Countries employing AI to coordinate renewable energy have achieved greater stability during peak fluctuations and extreme weather events. By learning from meteorological and consumption patterns, AI enables utilities to anticipate strain and reroute power before failure occurs.

Yet technical progress continues to outpace regulatory readiness. AI alone cannot resolve bottlenecks if policies, cybersecurity and human expertise lag behind. That lag is where most utilities currently find themselves.

3 obstacles between pilot and scale

Despite impressive results in controlled deployments, scaling AI remains difficult. Three persistent obstacles define the landscape: fragmented data governance, legacy hardware and cultural resistance to automation.

Many utilities still operate on supervisory control systems designed decades ago. These older platforms cannot handle the high-frequency data streams required for modern analytics. Upgrading them demands significant capital and cross-departmental coordination, neither of which moves quickly in a regulated industry built on long planning cycles.

The second obstacle is organizational, not technical. AI projects stall when utilities treat them as technology add-ons rather than as full organizational transformations. Executives may fund algorithms without restructuring workflows or retraining staff, producing isolated islands of automation that never connect to actual operations. The misunderstanding here lies in assuming that reliability stems from software alone. In truth, reliability emerges from integration.

The third obstacle is human. Employees accustomed to manual control sometimes distrust machine recommendations. Utilities that successfully implemented AI invested early in communication and change management, explaining what AI does, how it learns and when human intervention remains critical. And human intervention absolutely remains critical. This training reframes AI as a tool for empowerment rather than replacement, reducing internal resistance and increasing accountability. Without it, even technically sound deployments fail in the field.

Human oversight is not a limitation. It's the point.

While automation increases speed and efficiency, human oversight ensures accountability and ethical decision-making. Reliability depends on transparent human-AI collaboration. Supervisory engineers must interpret predictions, adjust thresholds and verify results. When operators understand model logic, they can intervene effectively during anomalies. My own experience in IT operations confirms this: Predictive analytics are only as reliable as the people maintaining them.

There is also a strong case for explainable AI frameworks that make algorithms interpretable. If engineers cannot see why a model forecasts a 45-minute restoration instead of 60, they may ignore the output, wasting AI's potential. Embedding transparency builds confidence and prevents blind reliance on opaque black boxes.

Cybersecurity adds another layer of oversight that the industry has been slow to take seriously. Adversarial attacks could manipulate AI inputs or corrupt training data, causing false predictions and system instability. Reliability now includes data integrity audits and model-drift detection. Ethical supervision requires vigilance not only over human bias but also over algorithmic vulnerabilities, a lesson that will become harder to ignore as AI systems take on greater operational authority.

Strengthening AI through data governance

Data governance forms the foundation of trustworthy AI, and it is consistently the piece utilities underinvest in.

Data interoperability, how information moves across departments, directly determines model accuracy. Many utilities maintain separate databases for operations, maintenance and customer service, creating silos that hinder holistic analysis. Integrating these datasets allows AI to contextualize predictions and improve optimization. Without integration, models are predicting in the dark.

Standardized architectures and common data taxonomies enhance both performance and accountability. Shared frameworks allow multiple utilities to train and validate algorithms collaboratively while protecting sensitive information through anonymization.

A common misunderstanding among policymakers is assuming that data sharing undermines competition. In reality, structured collaboration boosts collective resilience. By pooling non-proprietary operational data, utilities create an ecosystem where every participant benefits from improved reliability models — much like the electrical grid itself, interconnected and interdependent.

Policy is where technical optimism collides with bureaucracy.

The United States remains hindered by fragmented oversight among federal, state and local agencies. Without unified guidance, utilities face a patchwork of regional regulations that hinder adoption. Poorly regulated AI could create inequities in service reliability, favoring data-rich urban regions over rural areas.

The contrast with other countries is instructive. Among nations that aligned their energy, data and industrial strategies, Denmark and Norway are the clearest examples. They outperform others in grid stability and renewable integration. Clarifying jurisdictional boundaries and data standards would allow AI projects in the U.S. to scale efficiently. That clarification requires political will that has been largely absent.

Reframing the sustainability narrative

Beyond short-term reliability, AI is reshaping how utilities meet long-term sustainability goals.

Predictive algorithms help integrate renewable resources by forecasting variability and adjusting generation accordingly. Through digital twins, utilities can simulate thousands of energy-transition scenarios before implementing them, revealing trade-offs between cost, reliability and emissions that allow data-driven decisions in support of climate targets.

But the public narrative consistently oversimplifies the relationship between AI and sustainability. Many assume that any AI adoption automatically promotes clean energy, ignoring the carbon cost of data centers and model training. Understanding AI's environmental footprint is essential to evaluating its true role in decarbonization. The industry owes its customers, its regulators and itself a more honest accounting.

Looking forward, AI's role in energy reliability will expand from prediction to autonomous coordination. Self-healing grids that can isolate faults and reroute power autonomously are no longer speculative. Decentralized AI agents managing microgrids through peer-to-peer communication, each node learning from its neighbors to maintain stability, are in active development.

These advances promise faster recovery times but introduce new challenges: interoperability, cybersecurity and the need for regulatory frameworks that define accountability in automated decisions.

The misunderstanding most likely to persist is the belief that AI is an end point rather than an evolving partnership between technology and people. The future grid will not be fully automated; it will be collaborative. A system where algorithms augment rather than replace human decision making.

Genuine improvement arises from the synergy of data, infrastructure and human expertise. AI's true power lies not in autonomous prediction but in its integration into the social and organizational fabric of utility operations. Making AI work for utility companies requires more than powerful algorithms.

It requires a change in power itself, from technology as a master to technology as a partner.

National Grid is teaming up with GridCARE to “identify and unlock” interconnection capacity for large-load customers in New York, the companies said on March 25. 

GridCARE’s Energize platform uses AI modeling, grid simulations and “real-time system intelligence” to find spare capacity on the grid, potentially reducing interconnection headways from several years to “as little as six to 12 months,” the companies said.

The announcement did not mention specific customers that might benefit from the initiative. In an email, GridCARE CEO Amit Narayan said his company’s capabilities had drawn “significant interest from various sectors,” including large-scale manufacturers and data centers.

In a statement, National Grid President and Group Chief Strategy Officer Steve Smith said finding and utilizing spare grid capacity is the “fastest and least expensive way” to add new load.

“By responsibly unlocking latent capacity through modern, physics-based AItools, we can support economic growth, strengthen reliability, and protect affordability for the customers and communities we serve,” he said.

Data center construction costs rose from $7.7 million/MW in 2020 to $10.7 million/MW in 2025, commercial real estate firm JLL said in a January report on the state of the industry. Those figures exclude computing equipment but do encompass electrical infrastructure, which JLL said is a significant driver of project costs and siting decisions.

Still, speed to power remains “the primary criteria driving site selection,” JLL said.

GridCARE uses generative AI to sift through “quadrillions” of operating scenarios and identify conditions under which the grid could become constrained, according to last week’s announcement. Its platform then identifies potential capacity-boosting solutions, including flexibility assets like batteries and other forms of distributed energy.

The need to identify and unlock spare grid capacity is especially acute in regions with “new or phased expansions of large loads that would typically face multi-year connection delays due to required transmission upgrades,” Narayan said.

In October, GridCARE and Portland General Electric said the California-based company’s platform would free up 80 MW of capacity for data center interconnections this year. PGE already serves 800 MW of data center capacity in and around the Portland suburb of Hillsboro and anticipates energizing 400 MW more — much of it in the same area — by 2029, Larry Bekkedahl, PGE’s senior vice president of strategy and advanced energy delivery, told Utility Dive at the time. 

Narayan said GridCARE’s collaboration with National Grid would focus mostly on “concentrated areas with minimal supply resources” or locations near both dispatchable and renewable generation resources. National Grid’s upstate New York territory stretches from Lake Erie to the Adirondack Mountains and draws power from a mix of nuclear, gas, hydro and wind generation.

The region has recently seen a wave of interest from data center developers attracted by the relatively cheap, clean power and spare interconnection capacity at retired power plants and manufacturing sites.

High-tech manufacturing projects could also boost regional power demand in the years ahead. Micron broke ground earlier this year on a $100 billion foundry complex north of Syracuse that could host up to four chip production facilities within 20 years. To support the project, which Sustain CNY says will consume 7.15 TWh/year in 2032 and up to 16.17 TWh/year in 2043, National Grid will expand a nearby substation and lay eight “underground ultra-high-voltage lines,” according to a local media report.

In addition to reducing “time to power,” GridCARE’s solution can boost grid utilization, which the company says could help address customers’ and ratepayer advocates’ affordability concerns. It released a study in December showing that a hypothetical 1-GW data center in a mid-sized utility territory could reduce costs by 5% across all customer classes or free up more than $1.35 billion in capital for the utility by incorporating a “modest amount of flexibility.”

Portland General Electric has freed up more than 80 MW for data center interconnections next year using a new, artificial intelligence-enabled flexibility tool, the utility said Oct. 8.

PGE partnered with the California-based startup GridCARE, which uses AI, detailed hourly demand modeling and optimized flexible resources like batteries and onsite generators to find spare capacity. The added flexibility allows PGE “to interconnect multiple data center customers years earlier than initially expected,” the companies said.

“We are taking tools that have been in use for many years, that are well understood and well utilized … and moving them into the [grid] planning space,” GridCARE CEO Amit Narayan said in an interview. “In doing that, we realized we could open up quite a bit of capacity.”

PGE’s territory includes the Portland suburb of Hillsboro, long a global hub for chipmaker Intel. More recently, Hillsboro developed into a major data center hub thanks to its position at the mouth of an undersea fiber highway connecting North America and Asia, said Larry Bekkedahl, PGE’s senior vice president of strategy and advanced energy delivery.

Bekkedahl said Hillsboro alone has 800 MW of data center capacity operating today on a grid with a total system load of about 4.5 GW. PGE has around 3 GW of active data center load requests, more than 400 MW of which could energize by 2029, he added.

Bekkedahl said data centers eager to hook up to the transmission-constrained Pacific Northwest grid are coming to the realization that more onsite or nodal flexibility can speed up the process. That’s a big change from just a few years ago, when data centers presented as flat, round-the-clock loads that could push the system to its limits during the five or 10 days each year when demand peaks.

“We’d say to them, if that’s what you need, you’re going to have to wait until the transmission’s built… and so they realize they can flex themselves in our queue to get hooked up sooner,” he said.

The steady deployment of batteries, onsite generation and microgrids over the past decade has “created an opportunity that was not available to planners previously,” said Narayan, who founded and led virtual power plant platform AutoGrid before Uplight acquired it in 2023. Customers tend to use those resources for reliability and price hedging rather than mitigating strain on the grid, he said.

With electricity demand rising and speed to power top of mind for data center developers and tenants, Narayan said the moment for solutions like GridCARE’s has arrived. Every megawatt of additional capacity adds $30 million to $40 million in value for large data centers, he said.

And because the U.S. takes longer to build energy infrastructure than its principal geopolitical competitor, there’s a national security dimension to getting more out of the existing grid as well, he said.

“The U.S. is ahead in chips and algorithms, but the speed at which things are getting built in China and other places puts us at risk,” Narayan said.

Bekkedahl cautioned that grid flexibility can’t on its own address the challenges posed by headline-grabbing hyperscale AI data center proposals, which envision gigawatts of concentrated computing capacity. Former Energy Secretary Rick Perry’s Fermi America project, for example, has said it will deploy 11 GW of AI compute and energy infrastructure to support it in northwest Texas. Meta has more advanced plans for a data center in Louisiana that Entergy has said may require up to 2.2 GW of power.

“We do not solve the 2-GW data center problem,” he said.

But Bekkedahl said GridCARE’s tools gave PGE confidence to bring on data center loads ranging from 50 MW to 500 MW. It helped the utility and data centers work together to ramp up load and service in parallel and develop operational parameters for service agreements as new computing capacity came online, he said.

Narayan said it meant a lot that PGE took a chance on GridCARE. Famously risk-averse utilities like to see proof of concept before investing in non-wires alternatives, so this first real-world project is a big boost, he said. The grid tech company was developed out of the Stanford Sustainability Accelerator program.

“Once we show this can be done, it opens up a lot of opportunities in other places,” he said. “If we can get more from our existing infrastructure, that is helpful to all consumers using that infrastructure.”

More than seven of every 10 public companies on the S&P 500 now flag their use of artificial intelligence as a material risk in their public disclosures, according to a report released by The Conference Board. 

That figure represents a sharp increase from just 12% in 2023, reflecting the rapid implementation of AI use among major enterprises. 

“This is a powerful reflection of how quickly AI has developed from a niche topic to widely adopted and embedded in the organization,” Andrew Jones, a principal researcher at the Conference Board Governance & Sustainability Center, told Cybersecurity Dive via email. 

AI has moved beyond the experimentation phase at major enterprises and is embedded across core business systems, including product design, logistics, credit modeling and customer interfaces, Jones said. 

The report shows that corporate boards and C-suite leaders are addressing a range of risk factors in connection with AI deployment. 

Reputational risk is the most widely disclosed issue, at 38%, according to the report. This reflects the potential impact of losing trust in a brand in the case of a service breakdown, mishandling of consumer privacy or a customer-facing tool that fails to deliver. 

Cybersecurity risk is cited by 20% of firms. AI increases the attack surface, and companies are also at risk from third-party applications. 

Legal and regulatory risks are also a major issue, as state and federal governments have rapidly attempted to set up security guardrails to protect the public, while providing enough support for companies to continue innovation. 

While AI deployment is rapidly evolving in the enterprise, corporate leaders are still struggling to fully develop the governance structures to manage its use. 

The PwC “2025 Annual Corporate Director’s Survey” shows only 35% of corporate boards have formally integrated AI into their oversight responsibilities, an indication that companies are still working to develop stricter guardrails. 

“Directors recognize that AI brings both strategic opportunity and fiduciary risk, and many are starting to consider how to strengthen governance through regular education, clear oversight structures, and responsible-use frameworks," Ray Garcia, leader of PwC’s Governance Insights Center told Cybersecurity Dive. 

The United States has already experienced more wildfires this year than it has over the same period in any other year this decade, according to the National Interagency Fire Center.

With the risk of fire expected to grow due to climate change and other factors, utilities have increasingly turned to technology to help them keep up. And those efforts could get a boost following President Donald Trump’s June 12 executive order calling on federal agencies to deploy technology to address “a slow and inadequate response to wildfires.”

The order directed agencies to create a roadmap for using “artificial intelligence, data sharing, innovative modeling and mapping capabilities, and technology to identify wildland fire ignitions and weather forecasts to inform response and evacuation.” It also told federal authorities to declassify historical satellite datasets that could be used to improve wildfire prediction, and called for strengthening coordination among agencies and improving wildland and vegetation management.

Additionally, the order laid out a vision for consolidating federal wildfire prevention and suppression efforts that are currently spread across agencies. The White House’s proposed 2026 budget blueprint would create a new, unified federal wildland fire service under the Department of Interior.

So far, Trump’s directive has drawn a mixed response from wildfire experts. While some said it could empower local governments and save utilities money, others said the order’s impact will be limited.

“I think some people read into the order more than is there, and some people read less,” said Chet Wade, a spokesperson for the Partners in Wildfire Prevention coalition. “I don't know exactly what will come of it, but getting technology into the right hands could be very helpful.”

Fire prevention goes high tech

Since the 2018 Camp Fire that bankrupted PG&E and set a nationwide precedent for suing utilities that trigger large fires, energy companies around the U.S. have invested billions of dollars in grid hardening and undergrounding power lines. Public safety power shutoffs are now routine during high-risk weather for many utilities, especially in the West. Many utilities have invested in new monitoring equipment and artificial intelligence to better detect and prevent fires.

Abhishek Singh, CEO and co-founder of AiDash, which builds software for monitoring wildfire risk, said technology has allowed utilities to make significant strides in areas like tree management, where AI paired with satellite imagery can help them identify high-risk areas. Trump’s executive order, he said, could make this technology even cheaper by releasing satellite imagery that companies like AiDash can use to train their AI models.

That, in turn, could make AI monitoring more accessible to local governments and firefighting agencies with limited budgets, Singh said. He believes more government monitoring will allow utilities to relax their own surveillance efforts, freeing resources for other concerns like hardening grid infrastructure.

Former FERC commissioner and chairman Neil Chatterjee, who has accepted a role as an advisor to AiDash, shares Singh’s views on the potential of technology to mitigate wildfire risk. The executive order, he said, could “modernize wildfire prevention and bring federal policy in line with the technology that is available.”

Technology advances as risk grows

While Wade said he hoped the federal government improves coordination between agencies and increases access to public lands for vegetation management, he expressed skepticism the executive order would have much impact.

For one, he said, it calls for giving more authority to local governments, but it doesn’t say anything about ensuring those communities have the resources they need to address wildfires.

The president’s budget blueprint calls for funding the new federal fire service with $3.7 billion, plus $2.8 billion for a wildfire suppression reserve fund. But it would eliminate funding from other agencies and programs that have historically played a role in addressing wildfires, including grants to support local firefighting efforts and the management of state and privately-owned forests.

“While the budget provides robust support for Federal wildland fire management activities alongside States and local partners, these partners should be encouraged to fund their own community preparedness and risk mitigation activities,” the proposed budget states. 

U.S. Forest Service Chief Tom Schultz confirmed in testimony before lawmakers on the Senate Energy and Natural Resources Committee last week that the administration still has not released some firefighting funds to states for the current fiscal year, saying, “We’re still in discussion.” 

Trump has walked back his earlier calls to eliminate the Federal Emergency Management Agency in recent days. But the tax and spending bill he signed into law July 4 cuts funding for research and forecasting of climate-related threats. Partners in Wildfire Prevention said it was too early to determine what the cuts mean for wildfire mitigation.

Andrew Dressel, a power industry consultant and vice president at Charles River Associates, cautioned that even as utilities prepare and technology advances, the risk of fire continues to increase as well.

While the order focuses on prescribed burns and wildland management, he said recent urban fires like those that devastated Los Angeles have called attention to the need to harden homes and yards against fire. They have also raised concerns about the role of idle power lines in starting fires. But moving, upgrading or removing power lines is expensive and time-consuming, and utilities are facing these costs at a time of unprecedented projected load growth.

“An executive order can only do so much,” Dressel said. “We need legislation, federal legislation or state legislation or both, to really move the needle on these things.”