PJM, the largest grid operator in North America, has launched a multiyear partnership with tech giant Google and its Alphabet X moonshot Tapestry to streamline PJM’s grid interconnection process using artificial intelligence (AI) tools. The companies said the effort could significantly improve planning and decision-making as grid pressures intensify.
Under the initiative, announced on April 10, Tapestry, powered by Google Cloud and Google DeepMind, “will build on its core technology and develop a new set of AI tools and models” to “intelligently manage and optimize interconnecting power generation to the PJM electric grid,” said Ruth Porat, president and chief investment officer for Alphabet and Google.
The technology will be designed to fast-track how PJM connects new energy sources to its grid by deploying Tapestry’s AI-powered tools, she said. The tools will help automate the application intake and data verification process, unify disparate grid modeling databases into a single, collaborative platform, and support the faster integration of variable energy resources—such as wind, solar, and storage—that now dominate PJM’s interconnection queue.
She said the key aim is to ease the burden on planners and developers, speed up project approvals, and enhance affordability and reliability across the PJM region. However, the effort could also serve as a blueprint for grid operators facing similar concerns, offering a scalable model that could be adapted to accelerate clean energy integration across other U.S. regions and international systems.
“This is the first time artificial intelligence is being used to manage the entire energy interconnection queue and process, as opposed to point solutions,” said Page Crahan, general manager of Tapestry, during the April 9 press roundtable. “We are bringing more energy capacity on the grid faster. We’re driving efficiency and affordability by enabling more projects to send power to the grid and meet the region’s energy needs as efficiently as possible. And we are integrating more diverse energy resources—solar, wind and storage project capacity currently makes up over 90% of the PJM interconnection queue—and our Tapestry’s AI powered tools can support the rapid and reliable integration of these sources into the grid.” Ultimately, she said, “it’s worth reminding ourselves what’s at stake here, what the opportunity is. Investing in energy infrastructure unlocks significant growth in prosperity and economic activity.”
Considerable Potential for PJM
The partnership brings together PJM, the largest regional transmission organization (RTO) in North America, with Alphabet’s internal innovation lab and cloud computing division to address the growing backlog of power generation projects. By the end of 2023, more than 2,600 GW of capacity was waiting in interconnection queues nationwide, more than double the size of the current U.S. power fleet, according to Lawrence Berkeley National Laboratory.
As POWER reported earlier this month, grid operators have stepped up efforts to reform interconnection processes, but backlogs remain a critical concern. PJM in February told lawmakers that in 2023, only 4,800 MW were interconnected into the system with 180,000 MW of capacity. Of the approximately 50,000 MW that have completed PJM’s process, most are still not moving forward. That poses precarious implications for the grid operators supply picture, which, like other grid operators, is grappling with a confluence of system stresses that could threaten reliability.
“We project in the 2022 to 2030 timeframe that we could lose up to 40,000 MW—or 40 GW—of generation off the PJM system, and that’s due primarily to government and corporate policies that are putting pressure on our fossil fuel fleet,” said Aftab Khan, PJM’s executive vice president for Planning, Operations, and Security on Wednesday.“And there are economic factors, but you can see that that represents about 21% of our installed capacity.”
At the same time, new capacity is not arriving fast enough to fill the gap. “The concern that we also have is the pace of new entry of generation resources is also not happening at an adequate enough pace, we believe, to replace the risk of retiring resources,” Khan said. “So taken together, our concerns around retiring generation, slow pace of entry and significant electric demand growth poses significant risk for PJM to maintain reliability going into the future.”
PJM’s 2025 forecast reflects a dramatic shift in load trends. “Starting in 2023 and accelerating in 2024, and most recently, earlier this year, with our 2025 forecast, we started to project very significant peak electric demand growth—and this is driven by data centers, but also electrification of transportation and heating, and a resurgence of manufacturing,” he said.
“Generators who want to connect to PJM need to go through an interconnection cycle process, and we’ve been taking several actions to improve that,” Khan said. “We used to do a serial, first-come, first-serve interconnection process that resulted in a very significant backlog of generation interconnection requests. So we worked with our stakeholders beginning in 2021, ultimately leading to a FERC filing to transform to more of a grouping or cluster-based approach, and putting in criteria that’s more of a first-ready, first-serve.”
But even with those reforms,PJM sees room for improvement, Khan said. “Even though we’ve made significant progress with tools and automation to manage large numbers of projects in an interconnection cycle, it’s still end-to-end about a two-year cycle process,” Khan noted. “And we know that we need to continue to invest and look at all avenues to improve our interconnection process.”
Khan said the collaboration with Google and Tapestry is part of a broader strategy to accelerate project approvals and improve the interconnection cycle. “We see a huge potential to drive efficiency and quality improvements in the overall process,” he said. “It’s very difficult at this time to quantify and say, ‘Hey, this is what we’re going to achieve,’ but I can only say that I think there’s a huge potential for us.”
Tapestry’s Vision: A Unified, Intelligent Grid Model
Tapestry’s Crahan described the initiative as a “first-of-its-kind” strategy aimed at consolidating and contextualizing grid data. “We think about creating the world’s first knowledge graph for the electric grid,” she said. The idea draws directly from Google’s success organizing the internet: “We set out to do the same thing for the electric grid,” she explained, “and the information required that brings together all sorts of disparate data into a single place.”
She added: “You can think about something like the impact of Google’s knowledge graph on the way we search for the Internet—making it easy to ask a question and have somebody, a source, bring the information together to inform your trusted decisions and follow up.”
While the power sector already generates large volumes of data, she stressed that it’s often disjointed and unusable for decision-makers. “Sometimes it may even be too much data, so much that it’s not helpful for the decision makers to make sense of it all, to bring it all together in a single view,” she said. “The folks who are making decisions about our network are using multiple screens, spreadsheets, different software, and technical diagrams—not to mention the actual cacophony of alerts that go off in a control room in a challenging moment.”
Tapestry’s solution is a cloud-based, version-controlled, and collaborative model of the PJM grid—what Crahan called “Google Maps for electrons.” The goal, she said, is to replace the current patchwork of tools—each modeling different aspects of the grid—with a shared, unified environment that can “track changes so that developers, planners, and operators can access everything they need to make really critical decisions in one place.”
The project will roll out in phases over several years. “We’re developing a new set of tools expanding on that work, helping PJM to connect energy sources to the grid much, much faster,” Crahan said. First, the platform will introduce tools to streamline the interconnection application process, using natural language processing to, for example, “drag and drop a PDF and have it intelligently assess what is in that application.” These tools will automate and improve the data verification process for key factors such as land rights, equipment specs, and grid impacts.
The second major feature targets the lack of a unified model. “Grid planners assess whether they can connect new projects to the grid [by consulting] dozens of tools. They’re looking at different maps, databases, models and evaluation tools, and it is a lengthy process which can create some of the bottlenecks that we’d like to address,” she said. These tools don’t talk to each other, she noted. “Every time a change is made to that one model, it needs to be applied to all of the other models in consideration,” she explained. “Since these models are siloed, it is extremely difficult to manage all of this data and keep a consistent and updated set of information.”
Crahan described two core systems already deployed internationally: a grid planning tool that enables long-term simulations “at hourly resolutions up to 20 years in the future,” and GridAware, a suite of tools that uses computer vision and vehicle-mounted cameras (like Google Street View) to rapidly inspect physical infrastructure. “That dramatically accelerates grid inspection and repair processes,” she said, and “allows grid operators [and] field service technicians to proactively maintain the network for a more resilient grid.”
Ultimately, Crahan said, “we want to really bring the data that is engaged and available in the network and turn it into knowledge, so that the experts have what they need to make decisions about our grid of the future.”
The Planned Rollout
The companies said the tools will be rolled out in phases, beginning with development and limited testing in 2025. While PJM has not yet committed to formally adopting the AI tools as part of its standard interconnection process, the collaboration marks the start of an intensive co-development effort aimed at eventual integration.
“This is the kickoff of the collaboration. I would just emphasize this is really innovation, first-of-its-kind work,” said Crahan. “So nothing will be integrated into PJM process tomorrow, but we are beginning the work of developing these tools together,” she said. “Our plan is to be delivering solutions that PJM can start using this year—in 2025—but again, the most important thing above all is ensuring that whatever we deliver and build with that Tapestry meets the reliability and the sort of interoperability and operational needs of the PJM working team. So we will be working very closely on developing the tools, starting now imminently, and we will be working with PJM on how they integrate into their process over the next several quarters and throughout, hopefully, as we build upon this great work, perhaps even the next several years.”
Crahan also said that PJM will make decisions on what the best tools are that can conduct their responsibilities. “Tapestry is responsible for building something that is really incredibly helpful, and delivering that in a reliable, meaningful, deployable, secure way,” she said.
How the Initiative Compares to Industry-Wide Applications
Tapestry’s AI-powered grid planning tools have already demonstrated measurable success in international applications. “Because Tapestry doesn’t have our own electric grid to experiment with, we have been working closely with partners around the world for the last seven years, working in New Zealand, Australia, South Africa, the UK, working in Chile, and, of course, working in the U.S.,” said Crahan
A collaboration with Chile’s national grid operator, Coordinador Eléctrico Nacional (CEN), helped planners dramatically accelerate long-term simulations, she noted. “What we’ve seen is the planners that are using our grid planning tool are able to simulate their grid 86% faster,” she said. “They’re running 30 times the number of scenarios that they used to run a single scenario in.” She added that Tapestry has also layered in advanced weather forecasting capabilities from Google DeepMind to improve wind predictions, enhancing confidence in grid planning decisions.
Tapestry’s initiative echoes broader efforts to modernize the grid with AI, but it appears to stand apart—for now—for its focus on integrating siloed models and offering real-time, version-controlled collaboration. By contrast, the Open Power AI Consortium—launched just weeks ago in March 2025 by EPRI, NVIDIA, and startup Articul8—is developing domain-specific large language models trained on curated energy-sector datasets. The goal is to help utilities streamline operations, accelerate interconnection studies, and prepare critical filings such as licenses, permits, and rate cases. EPRI says the consortium’s first open AI model, trained on hundreds of NVIDIA H100 GPUs, could cut interconnection study timelines by up to 5x. The consortium also includes power providers like Duke Energy, PPL, PG&E, Exelon, and Portland General Electric, alongside tech partners like AWS, Oracle, and Microsoft.
Other efforts are emerging in parallel. The National Renewable Energy Laboratory (NREL) is developing eGridGPT, a research initiative to apply generative AI in the control room. Developed using open-source models and validated against digital twin simulations, the system supports real-time decision-making by grid operators through scenario analysis, action simulation, and equipment model mapping. Its multimodal capabilities include natural language interpretation of operator prompts coupled with explainable AI outputs grounded in physics-informed modeling.
Federal engagement in AI for grid planning is also ramping up. In November 2024, the U.S. Department of Energy (DOE) launched its Artificial Intelligence for Interconnection (AI4IX) initiative, offering up to $30 million to fund AI applications that streamline the nation’s generation interconnection process. The program, managed by ConnectWerx under a Partnership Intermediary Agreement with DOE’s Grid Deployment Office, targets a key chokepoint: reducing deficiencies in interconnection applications that delay analysis. Among its goals are accelerating application intake automation, improving site control verification, and enhancing data transparency—in line with the DOE’s broader i2X roadmap to modernize interconnection nationwide. DOE says more than 90% of interconnection applications submitted to some RTOs are currently deemed deficient, prolonging project timelines by years.
Numerous commercial platforms also use AI to improve grid operations, but few address interconnection queue management. IBM’s ELVIS, Siemens Energy’s Omnivise, and GE Vernova’s ThinkLabs and GridOS DERMS focus on reliability, forecasting, and DER coordination. GridUnity is a notable exception, offering automated workflows for interconnection.
“I see workflow automation happening for addressing the interconnection queue and transmission planning, and certainly applying AI in that domain is great,” said Crahan. “In our observation, that really doesn’t get necessarily to the heart of the challenge of unifying all these disparate models.” She emphasized that Tapestry’s work is complementary to existing efforts focused on point solutions or simulation tools. “What Tapestry views as a missing piece—and a real unlock for speed and data validation—is synthesizing those things, unifying them so planners and decision-makers can see them all together in one place.”
The scale of the PJM initiative is also unprecedented, she said. PJM’s footprint covers spans the District of Columbia and 13 states across much of the industrial Mid-Atlantic and Midwest. The technology could ultimately deliver power that is more reliable and affordable for the 67 million people PJM serves, noted Amanda Corio, head of Data Center Energy at Google.
Google’s incentive to participate in the solution is to bolster its broader initiatives to add firm, clean capacity to the grid. “We remain committed to our goals to decarbonize our electricity footprint, 24 hours a day, seven days a week, and matched on an hourly basis by 2030,” Corio said. Google is also backing advanced nuclear and enhanced geothermal, and has co-located data centers with new generation to ease grid burdens, she noted.
—Sonal Patel is a POWER senior editor (@sonalcpatel, @POWERmagazine).