Nearly 500 data center, artificial intelligence, and utility leaders came together to examine the impact of AI on the grid at GridFWD 2025 last month.
Given the constant hype about AI, I was expecting to hear that the energy industry is undergoing a dramatic transformation. What I witnessed instead looked more like incremental steps forward — steps which, taken together, could have a profound impact on how we plan and run the electric grid.
In the opening session I asked the audience: “Where do you see the most potential for AI at your organization?” No one answer rose to the top. We saw strong support across all the fields:
Let’s unpack each area briefly.
Customer engagement and information
One of the more promising aspects of AI is how large language models can search vast amounts of information and usually find helpful insights. LLMs have the potential to improve the customer interface of electric energy users; however, we are not yet seeing much uptake of these capabilities.
For example, we already have systems that examine the signature of electric use at a home or business to determine what’s working, and what may be out of the ordinary. These disaggregation tools are already enabling companies to automate assessments and improve demand-side capabilities. Fully automating this trusted relationship is an area where operators will tread carefully, though I can’t imagine this area not advancing in the near term.
Reliability and resilience
A few months ago, I wrote a piece asking “what is the point in advancing the grid if the system is less reliable and more costly?” This is one of the central questions informing how the energy transition interacts with the rise of AI — which is contributing to the load growth currently straining the grid.
I see the use of the new technology to address reliability issues as a rich opportunity area. For instance, AI is well-suited to quickly identify outages by absorbing a wide range of data, assessing what caused issues, and quickly suggesting solutions.
Asset optimization
This may be the most fruitful category for better managing the grid to date. In recent years the industry has started to deploy low-cost drones, sensors, and other advanced imaging information-gathering capabilities. Combined with image recognition software, they quickly pinpoint asset issues, at times before they even occur.
But friction remains where these information gathering systems feed into the overall systems of operators. Next-generation asset management, advanced distribution management systems (ADMS), fault location systems (FLISR), and other systems tend to be costly, complex and time-consuming deployments. When eventually used effectively at scale, though, asset optimization has the potential to really benefit the grid’s resilience.
Predictive analytics
Arguably the oldest of the AI and machine learning uses on the grid is to monitor critical assets. For decades, we have monitored large generation units and large grid infrastructure. These systems fed into SCADA and other data monitoring capabilities, enabling grid operators to better understand when issues may arise — often before they do — and have visibility into more parts of the grid. AI can deepen this situational awareness, and improve visibility into what is happening and about to happen.
Advanced forecasting
Grid operators have been forecasting both generation and demand for decades. These processes have been digitized in recent years but still rely heavily on intuition and human expertise. That said, computers, software systems, and now AI tools are getting better at understanding correlation, and in some instances even causation — for instance, between the weather and grid dynamics. Turning this data into actionable insights is an ideal use case for AI, though it’s still early days.
Back office and productivity tools
Integrated AI tools like CoPilot and Gemini are showing promise for helping the back office of grid operators run more smoothly. The fewer resources needed for administration, the more resources available to help the energy system run better.
However, I don’t foresee dramatic efficiency savings in this category of improvements. Even with AI automation, we will want human experts to make sure critical capabilities operate as needed. These experts — for instance, those well-versed in cybersecurity best practices — are both hard to train and hard to find. So even if AI might generate a great preliminary engineering spec, an experienced senior engineer will be needed to confirm and complete the project documentation.
Lots of promise but not a panacea
At GridFWD 2025, participants demonstrated their use of AI across functions including:
- Streamlined interconnection and permitting processes;
- Integrated grid planning;
- Orchestrated resource dispatch;
- Preliminary technical documentation;
- Grid simulation and modeling;
- Summarizing complex processes;
- Beginning of control room optimization; and
- Improving inspection and asset monitoring.
Collectively, these solutions may indeed remake the way that energy systems are managed. Indeed, they can also be seen as natural next steps in the trajectory of digitizing grid operating functions.
However, we want to get to truly transformational AI use, we need to address some real challenges that are holding back wider deployments at scale.
First, the industry has a lack of standards and controls for AI.
NERC-CIP and other grid-related standards ensure that significant resources are managed in ways that protect critical information and maintain base levels of operational best practices. But what do these areas mean with new AI automation? The market does not yet know.
There are no existing standards or frameworks for controls on AI use that align with how grid operators are used to working. And it’s not clear how AI guidelines could come at a speed that ensures sufficient security.
Meanwhile, there are no safety mandates or guardrails, which contributes to a lack of trust in the technology. Given the all-out race to win AI platforms, chips and other critical components, governments around the world are taking a laissez-faire approach to safety. Will this lead to an outcome where platforms will responsibility ensure the safe use of their products? History would suggest no.
Users of AI for the grid cannot be expected to self-regulate; NGO organizations are coming together and creating best practices but this is not enough. To get to wider use, AI providers need to actively demonstrate that they can be trusted.
There’s also a need for better accuracy. LLMs and other AI platforms are great at serving up “mostly right” information, acting essentially as an advanced version of auto-complete that helps to jumpstart solutions. But getting an assumption incorrect for critical grid components in a regulatory filing could have catastrophic impacts. Getting the design parameters incorrect on a critical circuit could be equally disastrous. AI liability and risk areas for operators remain murky.
Finally, the industry will need to train and upskill the workforce.
In many industries, the deployment of AI has raised the question of if AI comes in and does lower-level tasks, how do we train and upskill the next generation of key workers? Given how tentatively the power sector is adopting AI, the question may still be premature, but it’s still worth asking as skilled labor in the industry continues to age. GridFWD feedback made it clear that many people remain concerned about overreliance on AI tools.
Until these and other areas are adequately addressed grid operators will tread carefully in the AI era, and they should. Electric grid operators have existed for over a century in most places — and their jobs are becoming even more critical as the grid undergoes transition. How to protect critical data, ensure safety and accuracy, and a long list of other questions need to be answered before wider AI use accelerates.
Bryce Yonker is the executive director and CEO of Grid Forward. The opinions represented in this contributed article are solely those of the author, and do not reflect the views of Latitude Media or any of its staff.
