This transcript has been generated using machine learning technology. While this technology was trained with CIBC terminology, it may contain errors.

Tom Heintzman: Welcome to The Sustainability Agenda, a podcast series focusing on the evolving complexities of the sustainability landscape. I’m your host, Tom Heintzman. Please join me as we explore today’s most pressing issues with special guests that will give you some new perspectives and help you make sense of what really matters.

Josh Wong: I think we are all using AI in various facets in our lives. I think for the grid specifically in five years, AI would no longer or should not be a new thing. It should not be a hype. It should not be a buzz. It should be business as usual.

Tom Heintzman: Welcome to our multi-part series on the role of electrification in the transition to clean energy. Like last year, we’re producing a number of episodes in the lead up to CIBC’s second annual electrification summit, taking place on April 23rd, 2025. Throughout each episode, we’ll explore key issues related to our electrified future. On today’s episode, we’ll examine the role of AI in optimizing energy distribution. As we’ve previously discussed on this show, a lot is happening at the edge of the grid. Demand is now growing rapidly for the first time in decades. New types of demand like EVs and heat pumps are proliferating. Distributed generation such as rooftop solar is becoming commonplace. With all this change, our grid gets more complicated and harder to manage. Today, we’ll discuss AI-driven approaches to grid management and how this is helping to overcome traditional challenges to energy distribution. We’ll delve into examples of long-term and real-time applications and the impact of this automation on the energy transition. To lead us in this exploration, I’m delighted to welcome my guest, Josh Wong, founder and CEO of ThinkLabs AI. ThinkLabs AI is a specialized AI company empowering critical industries and infrastructure towards global energy sustainability. Josh has over 20 years of experience in clean tech, including in grid modernization, energy storage, and distributed energy resource management systems, or DERMs. He was formerly the founder and CEO of Opus One Solutions, which was acquired by GE, at which point he became the general manager of grid orchestration at GE. Participants at our inaugural CIBC Electrification Summit in 2024 will recognize Josh from our panel on edge of the grid distribution and distributed energy resources. It’s my pleasure to welcome you today to the show, Josh.

Josh Wong: Thank you, Tom. Great to be back with you on a very exciting topic.

Tom Heintzman: So Josh, before we dive into your technology, let’s take a step back and start with the way electricity grids have historically been managed. Could you explain to our listeners how electricity grids have historically been planned years ahead of time and then operated in real time?

Josh Wong: We definitely recognize that the grid is perhaps the most critical infrastructure sustaining society. Now, overall, I would say the historical ways of planning then operating the grid is based on long-term, worst-case conservatism. Basically, plan, build, set, and forget, and keep forgetting for the next 20, 30, 40 years. So what that means is, number one is we don’t have a lot of room to hit margins, which means we have a lot of buffer room, which leads to significant overbuilds. So this tendency for conservative overbuilds means that we have capacity to spare, and that allocates room for load growths, like increasing demand, such as now, as you said, heat pumps or electric vehicles, or when we need to reroute power upon outages. Now, why so long term, though, is because historically we have a very steady predictable demand increase or load growth. Now for the past few decades or so it has declined a bit, but it’s fairly predictable. And with predictability, we don’t need to be too smart about it because we can prepare for it and then forget about it. Now it’s of course being more challenging with operations and things being variable. So this long-term predictability drives long-term planning. Again, this set and forget type of mode. Now, I think another key characteristic though is that it’s very, very complex. So we have called the grid the largest and most complex machine man has ever made. And it is very intricately balanced, very analog.And then the operational side is basically rules driven, or very much like human intuition based on processes, standards, standard procedures, et cetera. And that has governed the grid for the past few decades and it has served our society pretty well since.

Tom Heintzman: Effectively, the load growth was modest and so we could build out for 10 or 20 years and be pretty comfortable that the infrastructure we built was going to meet demand. But the grid’s changing a lot now and demand’s growing quickly. So what are the major changes that are going on in the grid and how does it challenge this traditional way of managing the grid?

Josh Wong: Yeah, I think a number of factors and great question on that. So if there’s anything consistent over the past decade on utilities, it’s that they need to move faster. So the stakeholder pressure has definitely been challenging the utilities to make decisions faster. And so the time pressures around things like interconnections and the ability to submit plans and update the plans such as like integrated resource plans or integrated distribution plans. The second is more commonly heard is grid complexity. The grid is absolutely getting more and more complex. Why? Well, you mentioned a lot about distributed energy resources. So what used to come top down from big generators to long transmission lines to a spaghetti bowl of distribution lines. Now you have millions of distributed resources, each with their own variability and unpredictability. The third, would say the data readiness. So utilities never really had a lot of data, but suddenly now with things like smart metering, they have a lot of data which they are struggling to generate value from. And the data quality, data readiness is a challenge as well. And this does lead us to the AI conversation. But that is another big problem. I think another one is really this whole notion of generating solutions or design. So I’m moving to like Gen.AI type of design. What we have been doing to generate solutions for the grid, let’s say you have a grid congestion, and how do we solve this grid congestion? Well, for all these decades, we have been basing it on business rules, standard tables or practices, and a lot of trial and error and human intuition. The need to quickly generate solutions in a rapidly changing grid, that is a real problem right now and probably the biggest value prop for automation. The last but certainly not least is the driving workforce behind the utility. I think for the past maybe 10 years or so, I’ve been regularly hearing still that 50 % plus of the utility workforce are eligible for retirement. And so, It’s creating also a challenge to recruit new resources. Well, utilities are not like the hyperscalers. We don’t have the most sexy type of work environments. So I think being able to recruit and hire as well as to train a new work force as the previous generation is retiring is a huge challenge. And basically the risk to the grid is you have an increasingly complex grid escalating on one direction, workforce declining on the other direction and we are expecting unprecedented amounts of grid investments. So are we basing all these billion dollar capital investment plans for grid upgrades on a junior workforce that is learning how the utilities have operated without a lot of automation. So this is creating like the perfect storm right now.

Tom Heintzman: Okay, so there’s a lot, we got a lot going on here. We got rising demand for the first time in a long time. We have lots going on at the edge of the grid, including solar, which is both intermittent and two way. We’ve got new loads like EVs and heat pumps. We’ve got a plethora of data that’s hitting us. We’ve got more and more junior workforce. We’ve got big capital spend, lots and lots to manage there. So how does AI help overcome these challenges and manage the grid more effectively?

Josh Wong: Well, just ask Chat GPT, right? I think that’s the ongoing joke. But I think, whenever you look at innovation, it’s typically not done in silos. So a lot of innovation is cross-pollinating different industries into what’s our industry, which is the grid needs. So we’re learning from parallel sectors such as FinTech or from, especially an analogy I love to use is from autonomous driving. So let’s double click on that that analogy for a sec. The overarching objective is to create a safer, more reliable, affordable, and cleaner electricity system. To be able to like move and make decisions, we do need to overcome the complexity or lack of resources. So automation or in the fancier terms now, agentic AI, so AI agents, can really be leaned upon to accelerate change in decision making with the grid by AI-based decision making. And so the analogy is more like autonomous driving. So can we get like a driving assistance for the grid? What is that AI driving assistance for the grid planner or for the grid operator? Well, if you look at the analogy, it’s actually not that different between roads and the electric city grid. Both have traffic patterns like roads and highways. Both has directionality. So one way that power flow versus two way power flow. Both has a congestion like traffic jams. Definitely traffic jams or congestion on a grid is a huge problem for grid capacity and asset utilization right now. Both have lanes. For example, you don’t want to cross over even if you want to cross over to a nearby lane, then signal for us is we have voltage lanes. And now we have voltage lanes that are actually changing all the time because of something called dynamic line ratings. So even the lanes are shrinking or expanding by the minute or hour. Then both we have turn by turn instructions. So for Google Map or RouteU, for the grid, it’s capital plans for the next few years or switching orders for the next day or for the next week. And so, the ability for the AI to help assist the user or operator is really a big thing that we can follow the same, again, the design patterns. Now, where we focus first, though, first, if you look at, again, driving assistance, the first order of operations is, can we teach AI how to read the roads? Now, if it’s on, let’s say, a Tesla, that’s like camera vision. If it’s on something like other autonomous vehicles, it might be radars. If it’s like Google Maps, then it’s really loading up the maps so that we can route traffic around the maps. For us is teaching AI how to do power flow, understanding where power is coming and where power is going on a limited capacity grid. That is where we really want to focus on. So hopefully that analogy drives a bit of where AI is going to overcome or assist with the grid.

Tom Heintzman: That’s fascinating, Josh. I get the concept at a hundred thousand foot level. Are there any sort of practical examples that make this concrete? Some problems that have bedeviled the industry, but that AI can solve, you know, much, much more quickly or any examples that make this really to come to life for the listeners?

Josh Wong: So grid planning is really your planning ahead. So previously, let’s plan 40 years ahead for a big generator. And now with solar, we want to plan literally next month ahead in terms of connecting new resources. There’s a lot of plans that we need to accelerate the modeling and decisions of. And so where the AI comes in is can we take traditional engineering studies and enable AI first or AI native studies. So let me delve into the tech a bit more. Traditional engineering studies are based on mathematics. So for power flow will be something like a Newton-Raphson method or for backward sweep method. But those tend to be very slow. As an example, if you want to do like we have been saying like snapshot studies for the next decades. But now with things being changing all the time, such as with solar, we’re moving more and more towards time series studies. So can we look at every hour of the day and simulating that forward for a year and up to 10 years. So these type of time series studies for planning are really clunky and takes a very, very long time to solve typically 24 hours to run one study. That’s just one analysis and you need to do a lot of them. And so what we did is we started looking into what’s called physics informed AI. So can we teach AI with traditional physics and mathematics and engineering? So basically can I use traditional tools to train an AI model to replicate itself? We call these surrogate models. So can we have AI based power flow models? But now to do that, by going with the AI route, we can do it with super speed, hyper scale, and really allowing us to look into generative solutions, so GenAI. So back to the example, because I know there’s a lot of big words in there, but there’s an example. Previously I said a 24-hour time series analysis takes for years, takes about 24 hours. We have shrunk that down to 90 seconds. So this type of accelerated analysis gives us a lot more situational awareness and granularity of studies to do like real time studies and continuous studies rather than ad hoc studies. Scale-wise, previously we do one circuit at a time. And for the average listener, our circuit may be covering like 500 households or maybe up to like 1,500 households or so. So we do it one circuit at a time. Now with AI, we can hyper-parallelize to do entire provinces, states, and utilities within minutes to hours. So very large scale decision making. With that speed and scale, we can also cover a lot of scenarios. So what I mean by that is what we do today is at one scenario at a time ad hoc. Hey, I want to go project 50 % load growth. That’s one scenario, spend a week on that. Then I want to do 30 % penetration of solar, spend on that a couple of weeks on that study. But now what we do is because AI allows training. So what that means is I am doing bulk training for the AI to learn. So what that translates into is I’m actually training the AI to see hundreds of thousands of scenario in training. So all types of load growth, all types of solar penetration, all types of electric vehicle and switching and outages. So every single day, pretty much the AI would have seen more scenarios than a utility department would see in their entire career time. Now going back to these applications. So in planning, what that means is I can process interconnections, which currently is a big pain point for everybody from months to literally overnight. All those studies, all those analysis can be done overnight, which to us, which translates to about a couple, two to four hours, let’s say.Another one is capital plans. The game with regulations with utilities is they prepare their capital plan every single rate cycle, do a lot of ad hoc analysis on where they need to build up their grid. And so they submit the plans, thousands of pages of documents, and they just fight it out in regulatory hearings. But then if we look at it from the technical analysis perspective, they have assessed maybe three, five scenarios to justify the capital plans. And then the rest is challenging the assumptions and negotiating some middle ground, et cetera. But now the AI allows them instead of three, five scenarios on sensitivities. But we have generated and analyzed hundreds of thousands of scenarios. So because the volume and scale of high speed analysis is so big, we basically bring up the robustness of capital plans. Let me just pause here to see if that makes sense. But these are the planning use cases that we’re addressing today.

Tom Heintzman: Fascinating. So many thoughts going through my mind. I’m not native AI, you know, it’s come along very late in my lifetime. And so I have this innate, I guess, fear of AI, both the human dimension, fundamentally, cybercrime, you know, people breaking into it, or it malfunctioning. And you see that sometimes with these AI trading algorithms and the stock market. And so I am wondering if we entrust so much responsibility to AI for our electricity system and our source of energy that we’re going to run the economy on, what the risks are and how we mitigate those risks. Any thoughts in that regard?

Josh Wong: All right, so we talked about how we are dealing with a mission critical infrastructure. So it has to be trustworthy. With that, we are leveraging physics-informed AI, which is again using math and physics to teach the AI to be itself, but a lot faster, cheaper, better. So that maintains a source of truth and explainability, and we don’t have a lot of room for hallucinations. Now, cyber is a critical component because AI does open up new attack vectors. And so since we are dealing with utilities, we are bound by utility rules in terms of cybersecurity standards like SOC 2. And what we typically do is we actually don’t host any data or compute on our side because utilities spend a lot more money than we do as a company on securing their data and compute. We actually bring our AI factory or what we call the AI pipeline into the utility environment where both the data and compute are secured. Now, I think the bigger, as we always know with cybersecurity training, the bigger challenge is actually on change management in a user. And so as we bring on automation type capabilities, a big part of the task is not just creating great models, but also working with the user to look at process changes and to be able to leverage AI to manage a more complex, unpredictable, and changing grid. So we do keep both the technical elements as well as the user in mind.

Tom Heintzman: So last question. If you look out five years, what role do you see AI having in grid operations?

Josh Wong: To all of us, AI is transforming our lives faster than we can ever realize, like from search engines now to chats and really relying on AI for everyday driving. I think we are all using AI in various facets in our lives. I think for the grid specifically in five years, AI would no longer or should not be a new thing. It should not be a hype. It should not be a buzz. It should be business as usual. Utilities have been using AI for quite a number of years, but we are leveraging more advanced AI. So from forecasting, asset management, now to look at like even problem solving or the notion of agentic AI or AI agents. So we see AI really being business as usual. What that plays out in practice is the AI should be analyzing the grid all the time. So again moving from ad hoc studies, analyzing the grid all the time, looking at various like congestions, constraints and violations or destabilization vectors, and proactively generating solutions and recommendations, and finally partnering with the user to execute actions. So this type of trusted partnership between the AI and the user, we see that as the everyday norm in five years.

Tom Heintzman: That’s fascinating, Josh. I can’t wait to see it all play out. And I wish you and Think Labs AI all of the best of luck. Thanks for taking the time to join the show today. And thank you to the listeners for tuning in.

Josh Wong: Thank you, Tom, for having me here.

Tom Heintzman: If you would like to learn how electrification trends will impact your business, join us for CIBC’s second annual electrification summit on April 23rd, 2025 in Toronto. The summit will bring together leaders from across the electrification value chain, including developers, generators, utilities, heavy consumers, some of the largest investors and lenders in the space, as well as government and regulators. To register, please contact your CIBC relationship manager. Please join us next time as we tackle some of sustainability’s biggest questions, providing you different perspectives to help you move forward. I’m your host, Tom Heintzman, and this is The Sustainability Agenda.

Disclaimer: The materials disclosed on this podcast are for informational purposes only and subject to our Code of Conduct as well as CIRO rules. The information and data contained herein has been obtained or derived from sources believed to be reliable, without independent verification by CIBC Capital Markets and, to the extent that such information and data is based on sources outside CIBC Capital Markets, we do not represent or warrant that any such information or data is accurate, adequate or complete. Notwithstanding anything to the contrary herein, CIBC World Markets Inc. (and/or any affiliate thereof) shall not assume any responsibility or liability of any nature in connection with any of the contents of this communication. This communication is tailored for a particular audience and accordingly, this message is intended for such specific audience only. Any dissemination, re-distribution or other use of this message or the market commentary contained herein by any recipient is unauthorized. This communication should not be construed as a research report. The services, securities and investments discussed in this report may not be available to, nor suitable for, all investors. Nothing in this communication constitutes a recommendation, offer or solicitation to buy or sell any specific investments discussed herein. Speakers on this podcast do not have any actual, implied or apparent authority to act on behalf of any issuer mentioned in this podcast. The commentary and opinions expressed herein are solely those of the individual speaker(s), except where the author expressly states them to be the opinions of CIBC World Markets Inc. The speaker(s) may provide short-term trading views or ideas on issuers, securities, commodities, currencies or other financial instruments but investors should not expect continuing analysis, views or discussion relating to those instruments discussed herein. Any information provided herein is not intended to represent an adequate basis for investors to make an informed investment decision and is subject to change without notice. CIBC Capital Markets is a trademark brand name under which Canadian Imperial Bank of Commerce (“CIBC”), its subsidiaries and affiliates provide products and services to our customers around the world. For more information about these legal entities, as well as the products and services offered by CIBC Capital Markets, please visit www.cibccm.com.

Powering forward: AI-driven approaches to grid management

Tom Heintzman

MD & Vice-Chair, Energy Transition & Sustainability

CIBC Capital Markets

For the full catalog, visit