AI-generated image credit: Anne Bailey / DALL-E
AI-generated image credit: Anne Bailey / DALL-E
In April, an AI development and deployment company quietly spun out of industrial conglomerate General Electric. Dubbed ThinkLabs, the company emerged from stealth today, bringing with it a flagship offering in the form of an autonomous orchestration copilot for grid operators.
Founded by Josh Wong — who previously founded the DERMS software startup Opus One Solutions — the company is looking to act as both Google Maps and cruise control for the electric grid.
- The top line: ThinkLabs Copilot uses the physics of the grid to build a digital twin that can be used to generate real-time models and analytics for operators faced with increasingly complex decisions as load shifts and grows. The startup is entering a grid orchestration market that is somewhat crowded, with both other startups and incumbent software vendors, but hopes its utility ties through GE Vernova will push it to the front of the pack.
- The market grounding: Utility responses to grid challenges ranging from capacity constraints to retirements run the gamut. Some are adding fossil fuels and others are experimenting with tools like dynamic line ratings. And while a few are experimenting with artificial intelligence, recent research by Latitude Intelligence reveals that most utilities are still taking a “wait and see” approach to deploying AI, and aren’t expecting to make strategic investments in the technology for another five years or so.
- The current take: ThinkLabs has a major advantage in what is an extremely challenging market, said investor Micah Kotch of Blackhorn Ventures: its relationship with GE Vernova. “A lot of investors won't touch deals that serve utilities because of challenges in go-to-market,” Kotch said. “We think that some of the key benefits of the relationship with GE are that they provide credibility, critical data from early integration with an [advanced distribution management system] and an established channel with large utilities.”
The ThinkLabs spin-out came just before General Electric’s three-way split in early April, which included the spin-off and IPO of its energy arm GE Vernova.
Building the ThinkLabs team inside of GE was key, said Wong (who sold Opus One to GE in early 2022). However, he clarified, the end goal was always to get ThinkLabs to the point where it could stand on its own feet.
“We see the need to combine the speed and agility of a startup with the global reach, strength, and installed base of GE,” he told Latitude Media. “We’ve always had this understanding that if this picks up more pace, we would be able to attract external capital and spin this off as a separate company.”
About the tech
The best analogy for what ThinkLabs has built is something like Google Maps, Wong said.
That ubiquitous app acts as a day-ahead or hour-ahead study, rather than operating on the ten-year-ahead timelines that are so common in utility analyses. “You have continuous, real-time analytics of traffic patterns,” he said. “Why don’t we have that on the grid?”
Grid planning is based on manual studies and worst-case scenarios, Wong added: “There’s no solution generation.”
Extending the analogy, Wong also pointed to driver-support tools like lane control in passenger vehicles, which knows when a car may be about to leave a lane and can help a driver steer back.
“On the grid we have nothing like that,” he said. “We have to run more studies for voltage violations. We have to run studies for reverse power flow, and it doesn’t generate solutions.”
But artificial intelligence now has the potential to enable those solutions for the grid.
“We can actually use all of the physics of the grid to train an AI surrogate model of the grid, so that the AI mirrors the physics,” Wong said. “And that means it’s far more real-time; you can actually do second-by-second traffic pattern analysis on the grid.”
That situational awareness can enable a control room to make decisions it isn’t currently equipped to make, due to the “infinite number of control variables” on today’s grid.
The next step, Wong said, is pre-training solutions for scenarios like storms, added solar, or unusual EV charging activity.
“Normally you’d need to do a study and then a contingency, but in this case, we do all the work in pre-training,” he said.
ThinkLabs can run thousands of scenarios on the grid’s digital twin, training it to replicate the grid’s behavior in the face of different circumstances. The data used for that training isn’t perfect, in large part because utility data on which they’re trained isn’t perfect, Wong added: “But those models can sufficiently train an AI to replicate [the grid’s] behavior.”
It’s that physics-trained element of ThinkLabs’ AI that sets it apart from others in the field, said investor Emily Kirsch, at Powerhouse Ventures
“ThinkLabs is differentiated through their proprietary, physics-informed AI digital twin, which includes AI algorithms that generate grid models, provide analytics, and are validated against the physics of the grid,” Kirsch told Latitude Media. “ThinkLabs provides real-time state estimation capabilities as well as more efficient scenario planning — both are critical to resolving network issues before they lead to outages.”
Use cases and proof points
The biggest technical breakthrough for ThinkLabs came when the platform proved ready to conduct “distribution state estimation,” which entails predicting the current operating conditions of the distribution network at any given time.
Solving for state estimated power flows in distribution systems has typically been extremely challenging, and most assume that automating the process is several years out. But Wong said his team was able to build out that capability in just two months. Now, he said, the technical proof points are done, and an alpha release of the platform is already in utility hands, with a beta version coming later this year.
Another key use case Wong sees for ThinkLabs is in bringing real-time data to power flow analyses — something other energy players are also experimenting with.
Power flow estimations today rely on manually evaluating operating conditions. (The actual real-time visibility of the grid is less than 0.5%, even with smart metering, Wong added: “good luck trying to do real-time power flow!”)
Creating that estimation relies on engineering-informed formulas, which tend to be “slow, inaccurate, or even no results,” he explained. And, those formulas don’t allow operators to prepare for all possible scenarios.
“The big differentiator here is that now we actually get a trustworthy, accurate understanding of the grid’s power flow,” Wong said.
And that means better insights into things like congestion, and a slew of potential solutions, combining different resources and business models: “You can have more sophisticated decision making when you have all these different resources and different models,” he added.
The utility response
Both Wong and his investors said the initial response from utilities has been both surprising and encouraging.
“Utilities are far more receptive to the grid copilot that’s AI-driven because they are very familiar with the engineering-based use cases,” Wong said. “Fundamentally, we are not reinventing the wheel; we are solving old use cases in a far more sophisticated way.”
As to the most common question surrounding AI these days — “does the model hallucinate?” — Wong said the answer is a simple one.
“We’re not training our models with the internet,” he said. “We’re just trying to replicate the physics…and [utilities] understand that.”
And, he added, the noise that utilities already get from SCADA measurements is larger than any margin of error from ThinkLabs’ models. It’s a level of data risk they’re both familiar and comfortable with.
Learn about the pathways to adopting AI-based solutions in the power sector in a first-of-its-kind study published by Latitude Intelligence and Indigo Advisory Group.
.webp)
Learn about the pathways to adopting AI-based solutions in the power sector in a first-of-its-kind study published by Latitude Intelligence and Indigo Advisory Group.
Learn about the pathways to adopting AI-based solutions in the power sector in a first-of-its-kind study published by Latitude Intelligence and Indigo Advisory Group.
Learn about the pathways to adopting AI-based solutions in the power sector in a first-of-its-kind study published by Latitude Intelligence and Indigo Advisory Group.
To date, utilities seem most interested in ThinkLabs’ congestion management and DER management capabilities, Wong said, adding that a “large utility customer” is currently beta testing the platform, and there’s a pipeline of utility customers that will hopefully take it for a test drive later this year.
Kotch said Blackhorn Ventures conducted “a lot of customer diligence” with utility representatives while considering an investment in ThinkLabs, and also spoke with Indigo Advisory Group to get a sense of the utility mindset around AI deployments. (Editor’s note: Indigo co-authored Latitude Intelligence’s recent report digging into the utility AI landscape.)
“It’s not a homogenous market,” Kotch explained. “There [are] certainly folks who are a little more forward-thinking and ready, and there are others who will attempt to build some of this solution in-house, and will probably revert back to third-party entities like ThinkLabs three years from now.”
But there’s another key element at play in the “perfect storm” facing utilities, which is the shrinking utility workforce, he added. Research from Indigo Advisory Group found that in the next decade, 50% of the utility workforce will retire.
“The thing that attracted me in the beginning was this challenge around workforce and talent,” Kotch said. “I think that in the best case, this is really a tool to help the next generation of grid operators operate the grid of tomorrow with the tools of today, as opposed to the tools of yesterday.”
Wong said labor challenges seem to be top of mind for ThinkLabs’ pipeline of utility customers as well.
“A large utility in the States I’ve just spoken to said that one in every six weeks they have an experienced operator retire,” he said. That’s a massive amount of experience that utilities are losing, at the same time as the grid is becoming “infinitely more complicated,” he added.
Despite the reportedly bright-eyed response from utilities, ThinkLabs (and its competitors) aren’t yet removing humans from the decision-making process — and Wong said that eventuality is probably pretty far off, Wong said.
When it comes to utilities, “trust has to be earned and not given,” he said. “And what we have been providing are real-time analysis and recommendations as a supervisory automation mode,” he added, which can be very beneficial for operational planning but ultimately rely on a human grid operator to make decisions.
The market for copilots
Despite general utility hesitance to bring AI into the control room, the market is hardly empty of competitors. Grid orchestration start-ups like Camus, as well as incumbent utility software vendors, are all trying to carve out their niche.
The big fish players are a greater threat to the ThinkLabs market than other startups in its sector, said Kirsch at Powerhouse, pointing to recent partnerships between Microsoft and Schneider to offer a copilot product for control rooms, and to a similar collaboration between IBM and Siemens.
But, she said, that competition comes with a major caveat: “Neither product is ready for market, and development timelines are anticipated to take at least several years.” ThinkLabs, on the other hand, is going to market now.
Given the speed at which utilities tend to move when adopting new technology, Kirsch said, market timing and sales cycles could both prove challenging. “But ThinkLabs will be bundled alongside ADMS procurement processes that already exist and has GE Vernova as an exclusive channel,” she added.
Quantifying the potential impact of a tool like ThinkLabs is challenging, said Kotch at Blackhorn Ventures, but investors are hoping it yields results like decreases in power outages, fewer maintenance truck trips, and more renewables on the grid.
“Can we drive an increase in network capacity? Can we drive an increase in hosting capacity? In reliability? Can we drive an increase in DER penetration and reduce the interconnect process time for new projects? I think those are the impact metrics that I would probably look to,” he said.
.webp)
