Utilidata and Nvidia think their AI-enabled smart meters can help solve tricky grid challenges, from growing data center demand to more rooftop solar and EVs.
Conversations about AI and the power grid tend to focus on the demands that the developing technology will place on the country’s aging energy infrastructure. But Josh Brumberger, CEO of Utilidata, has a vision for how AI can actually help the grid.
The Rhode Island-based grid technology company is working on what it calls “edge AI intelligence” — smart meters or grid control devices embedded with chipsets designed by leading AI chipmaker Nvidia. Those devices have the computational capacity to process massive amounts of data and make split-second decisions, enabling utilities to better manage increasingly complicated power grids.
On Tuesday, Utilidata announced $60.3 million in funding to expand production and deployment of this technology with a growing list of utility partners. The new round brings Utilidata’s total venture funding to $126.5 million and was led by Renown Capital Partners and joined by existing investor Keyframe Capital, as well as Nvidia and Quanta Services, a major utility grid, energy infrastructure, and data center engineering and services company.
“We want to make it as easy as possible for hardware manufacturers to embed AI and distributed intelligence into their devices,” Brumberger said. “There’s this concept that AI is going to be crucial as we go about developing our next-generation infrastructure — in this case, the power grid. We were kind of on the edges of those discussions a few years ago. Now we’re at the center.”
Utilidata and Nvidia began to jointly develop their technology in 2021. The next year, they launched a consortium of U.S. utilities, along with leading U.S. residential solar and battery installer Sunrun, to support its deployment.
Since then, the technology has been selected to play a role in several cutting-edge utility grid projects.
The projects share some common features, Brumberger said. They involve collecting massive amounts of data, such as subsecond readings of the voltage and frequency of power flowing through utilities’ distribution grids. Those data must then be processed by integrated circuits running hefty mathematical calculations before grid operators can make use of it.
The in-the-field computers must also be reprogrammable to perform a shifting variety of tasks, rather than “hard-wired” for only a preset range of duties, he said. And those tasks may require autonomous decision-making, like coordinating utility and customer-controlled devices to respond to changing grid circumstances, which is possible only with technology capable of acting faster than traditional low-bandwidth wireless communications from central utility control rooms.
Utilidata, which got its start providing grid voltage control equipment to the utility industry, restructured its business in 2020 to focus on these kinds of “grid-edge” challenges. The goal was to develop new versions of long-standing utility technologies that simply don’t have the speed necessary for the modern grid.
Take the more than 100 million smart meters deployed across the U.S. since the mid-2000s. Those meters — essentially stripped-down, weatherproofed computers linked via wireless networks — are collecting energy billing data, alerting utilities to power outages, and enabling some basic grid control capabilities today.
But the chipsets in those earlier generations of advanced metering infrastructure — AMI 1.0, in utility parlance — were designed to do a preset list of tasks and to be cheap enough to be deployed in the millions. The underlying computing technology has gotten both cheaper and better since then.
Major smart-meter vendors such as Itron and Landis+Gyr have been boosting the capabilities of their latest “AMI 2.0” systems to carry out increasingly complex activities. Utilidata and Nvidia claim that their technology platform, dubbed “Karman,” exceeds the capabilities of its peers in the field, though their price point is likely higher too. (AMI vendors tend not to disclose per-unit prices, and cost and pricing vary widely depending on order volumes and vagaries of industry demand.)
Utilities take a long time to move from testing a technology to deploying it at large scale. Utilidata’s earliest pilot projects, launched in 2022, embedded Nvidia chips in devices that attach to existing meters. Last year, the company signed a deal with Aclara, a longtime smart meter manufacturer and subsidiary of electronics giant Hubbell, to develop an integrated smart meter using the Karman platform.
Utilidata and Nvidia’s projects with Portland General Electric, Duquesne Light, and other utility partners aren’t going to completely replace those utilities’ existing smart meters — at least, not right away. Instead, these initial projects are tied to strategic deployments at parts of the grid where the utilities are seeking more granular information, Brumberger said.
One major area of interest is in assessing the grid impacts of rooftop solar systems, backup batteries, EV chargers, and other so-called distributed energy resources, he said. A growing number of utilities are looking for ways to enlist these kinds of devices to reduce strains on their power grids — say, by instructing batteries to store solar power at midday to release it later when it’s more valuable to the grid, or by coordinating when EV charging happens to avoid overloading local grid infrastructure when everyone charges at once.
These virtual power plants (VPPs) or distributed energy resource management systems (DERMS) can sometimes be handled in a command-and-control fashion by utility grid operations centers communicating to in-field devices via hard-wired, cellular, or broadband internet. But more advanced tasks require complex computations of local grid conditions and real-time communications between multiple local devices — exactly what Karman was designed to handle, Brumberger said.
“How can you have a VPP that’s, from a capacity perspective, as big and as reliable as a gas-fired plant, without accelerating computing and AI? You’ve got so many disparate resources that have so much untapped value that we ultimately have to unlock,” he said.
Portland General Electric, which is planning to rely on distributed energy resources for a significant chunk of its future grid needs, sees technologies like Karman as a way to better understand the reliability of VPPs and DERMS, Ananth Sundaram, senior manager of integrated grid at Portland General Electric, told Canary Media in a 2023 interview.
“We can look at grid services, we can look at disaggregation of the power, and what customer behavior and customer signatures we can track,” he said. “That will not only provide us a solid platform for serving our clients, it also helps us harvest massive amounts of data we need to understand what exactly is happening on the grid edge.”
Utilidata is hoping that utilities and regulators will keep these future needs in mind when planning the next cycle of large-scale smart meter deployments. Brumberger noted that Quanta, a new investor in the latest funding round, is a key partner in many large-scale utility infrastructure and smart meter deployments.
Utilidata’s near-term deployments are also dependent on the Trump administration preserving the DOE grants supporting its first-of-a-kind utility projects. The administration has frozen and threatened to end federal climate and energy funding approved during the Biden administration, as well as to eliminate large swaths of the federal workforce, including DOE offices that manage these grant programs.
“We have not received any word that those projects are not happening,” Brumberger said about the grants. “If you sort of peel back the layers of our project, at its core, it’s next-generation AI infrastructure. That theme does seem central to this administration, when they talk about winning the AI race, about hardening our critical infrastructure.”
The latest round of funding will allow Utilidata to expand to new markets, both outside the U.S. and outside the utility grid, Brumberger said. In particular, it’s exploring the prospects for embedding its Nvidia-enabled distributed energy control devices within data centers themselves, enabling them to better understand and control power usage down to the server level, he said.
Utilidata and Nvidia’s computing platforms will also be collecting, analyzing, and “training” from the data they’re collecting, much like large language models (LLMs) “train” on massive amounts of human-generated text and images, Brumberger noted. The data might include things like differences between the grid voltage signals that accompany power disruptions caused by people turning things on and off in their homes and those caused by external impacts like tree branches hitting power lines.
“It’s no longer just a sensor, but a little hub of activity where you can train locally, so not every piece of information needs to leave the site,” he said. “The question is going to be, is this the kind of tech you need on 10% of your territory, on 50%, on 80%, or 100%?”
Jeff St. John is chief reporter and policy specialist at Canary Media. He covers innovative grid technologies, rooftop solar and batteries, clean hydrogen, EV charging, and more.