Episode 129: The AI Solution the Grid Desperately Needs with Gridmatic

Molly Wood Voice-Over: Welcome to Everybody in the Pool, the podcast where we dive deep into the innovative solutions and the brilliant minds who are tackling the climate crisis head on. I'm Molly Wood.

This week, we're back to talking about energy and grid flexibility and AI. In this case, an AI solution for meeting energy demand and hopefully for helping to meet the demand for new data centers that's being driven in large part by AI.

Gridmatic is a company that was formed specifically to make renewable energy assets easier to manage and integrate into the grid. It uses AI to predict renewable energy output, optimize battery storage, and help massive electricity customers be a little more flexible. Once they're connected. Let's go.

David Miller: Hi, I am David Miller. I am, uh, the Chief Commercial Officer at Gridmatic. And Gridmatic is a power company. We serve customers, electricity, and it's backed by, uh, renewable energy, and we use AI technology to manage the risk of that power supply and do so cost effectively.

Molly Wood: Amazing. Okay. Now tell me more. So, yeah, like when, uh, you know, you're, you're sort of known, you're called a modern energy company, um, focused on decarbonization. Are you, as we might think of a power company generating electricity, distributing, all of those things? Like what's the product here?

David Miller: Yeah. Yeah. So, um, taking a step back, our, our founder, uh, helped build the AI that runs Google Ads. Uh, and he realized that similar AI models could be used, that are used to predict, you know, when to show or which ad to show, uh, can be used to predict where and when the wind will blow or where the grid will be stressed for not having enough power supply.

And so he founded Gridmatic in 2017 and we've grown and expanded by now optimizing the output of big batteries on the electricity grid. We sign renewable offtake contracts and we serve end customers, uh, electricity. And we use AI to manage the, uh, the, the optimization of the forecast and, and risk, and, and risk.

And so we, we operate in both wholesale and retail electricity markets. Though our customers primarily know us as a retail electric provider. Um, so that's the name for, uh, businesses that are competitive providers of powers to end use customers in markets where that exists.

Molly Wood: What are the markets where that exists?

David Miller: Um, in the US, think Texas, the Mid-Atlantic, the northeast. Those are the, the primary markets.

Molly Wood: Um, so you, so in those markets, you are able to sell directly to consumers, like consumers can say, click this button, get energy from Gridmatic?

David Miller: Well, so, um, yeah, so those markets are markets where, where customers have a choice to buy, uh, electricity from whoever they want.

Molly Wood: Mm-hmm.

David Miller: As opposed to like one utility company, which is what exists for the other markets. And, but the experience of the retail customer, it varies based on, uh, kind of their category. So for a, a residential consumer looking at residential retailers? Yeah, they, they can, you know, choose who they want, they can sign a contract. Um, uh, they'll see advertisements for, you know, unique rates like free nights and weekends kind of things, the way that the rest of us see cell phones.

Um, and for a commercial industrial customer, the way it works is your retailer will, will tailor the price of electricity for you that's specific to the profile that you consume electricity, rather than a generic rate that would be the same for everyone. Um, so that's what, that's what that, that rewards customers that, you know, use power when it's most readily available. And, and that's what Gridmatic does. So we, we only sell to these commercial industrial customers and then, you know, one of those segments is, is data centers. But we serve across the commercial industrial segment.

Molly Wood: So then do you only operate in, operate in the United States? At least in states that have deregulated markets like that?

David Miller: No, not entirely. Uh, because we still also operate at the wholesale level, so…

Molly Wood: Got it.

David Miller: In other states that don't have retail choice, our retail business doesn't exist, but we can trade in the market, we can operate, um, uh, battery storage and sign renewable contracts and then sell power products to other customers that, um, you know, either utilities or or other entities that, um, that fill that load serving entity role.

Molly Wood: Gotcha. Okay, great. And then how, and then talked about, talk about the optimization of renewable resources in particular and why this product was so necessary.

David Miller: Yeah, I think, um, you know, kind of the, the reason why a company like Gridmatic needs to exist is sort of as renewables have gone from a small percentage of the mix where they can just sort of, you know, kind of be in the margin of error of the supply and demand balance to a much larger portion where they are kind of critical source of energy that's relied upon. Um, we've kind of had challenges doing that and, and battery storage has, has helped to address those challenges. But fundamentally, it, it's understanding, you know, is the renewable supply gonna be available or not? And if not, how can I back it up?

Can I back it up through battery storage or can I back it up through some other mechanism? You know, trading with another resource that's available at that time period or diversifying across the grid. And that type of analysis is, uh, very challenging.

So if you look at, um, you know, what is the kind of error in load forecast for the next day? It's, you know, a couple percent maybe. Uh, the way, uh, load forecast work. If you look at that same error for like wind output in a given hour, it might be like 15%. So there's just so much more volatility associated with the renewable output that you need more advanced forecasting and kind of more advanced risk management if you're gonna actually rely on those renewables.

So a lot of companies that are in our position, you know, don't really rely on the renewables. They maybe have mostly thermal power generation and then a little bit of renewables, which is kind of like on the margin. But if you want to like actually rely on the renewables and, and not have them totally backed up one for one with thermal resources, you need a better, a kind of forecasting and risk management approach.

Molly Wood: Right. Okay, so there, it seems like there's sort of two pieces of the technology. One is the forecasting, and then the second is the optimization, right? Around the battery storage? Tell me how sort of each one works. Like how, first of all, how do you forecast the sun and the wind? Like, how tapped in are you to the clouds here?

David Miller: Yeah. Uh, as much as we can be. [Laughs] Uh, so the, there's a traditional way of doing that. And then there's the Gridmatic approach of doing it.

The traditional way of forecasting on the grid is to use power flow model. So that's a model that kind of replicates what the grid looks like. It says, you know, here's power plants and here's the power lines and here's how much they can operate. And there therefore, you know, here's what the prices on the grid are gonna look like. Um, and that's what the grid operator does. But it's a really hard model if you don't have all the information that grid operator has, it can be, you know, very skewed and go off in different directions.

And so Gridmatic’s approach is to build, you know, a true AI statistical approach where we don't try to replicate the grid entirely, we use lots and lots of data to replicate results that the grid produces. And so we are, uh, you know, bringing in lots of historical data around, uh, production of the grid. We're bringing in grid topology data. Even if it's not a powerful model, we still need to understand what the grid looks like, 'cause that changes over time.

And then one, weather is really a fundamental task here. And so we do a lot of work on weather, um, including, you know, understanding how do we, uh, there's a lot of weather data out there, so what's the weather data that's relevant? How do we, uh, use it? How do we piece it together? Uh, how do we augment it or, or build our own weather model?

There's a lot of innovation happening right now in AI-based weather models. Um, and so that's a big area of research for us, continues to be.

Molly Wood: Fascinating. Okay, so then, and then let's talk about the, the kind of the result or the output. Like for example, if you forecast a demand surge and you work heavily in battery storage optimization, what happens?

David Miller: Yeah, so just taking a battery example, um. A lot of the stuff that happens in wholesale electricity markets with generation, including batteries happens, um, in the day ahead market and then in the realtime market. And the day ahead market really exists for kinda the fuel plants to figure out what they're gonna do and turn on over multiple hours or, or different periods or, or schedule their fuel to them. So it's a really important period where we decide kinda what plants are gonna be online the next day.

And most of the reliability services across grids, like the ancillary services, the reserves, those get planned in the day ahead market as well. Um, and so for battery storage, when you're figuring out how you're gonna operate a battery, when am I gonna charge, when am I gonna discharge and when am I gonna provide these reliability services that involve me standing or, you know, being on, on standby, uh, you need to make that decision typically at the day ahead timeframe. So for us, you know, what we are trying to do is figure out is there gonna be a price spike, unexpected price spike in the realtime market tomorrow?

Molly Wood: Mm-hmm.

David Miller: And if there is, then we wanna stay out of these reliability services so that we're available for it.

If there's not going to be, then we wanna make sure that we're, you know, in the services that, uh, that don't require us to discharge energy during that period. And effectively what that means is you're predicting a price spike that's an unexpected price spike in the real time market, a day ahead of time, which is just a very hard analytical challenge for anyone to do, but is what all battery operators in the market have to do these days.

Um, and so that, that's a problem that we have spent a lot of time focused on is how do we have the best predictions on the next day basis?

Molly Wood: Are you a solution that like can only exist because of the deployment of gridscale batteries? Um, like sort of, you know, previously in renewables deployment, would this have worked?

David Miller: The batteries, uh, are kind of a, a great, you know, kind of a, a, a big, uh, user of our technology, but they're not the only way the technology can be used. So, for instance, we work with flexible load customers that have the ability to curtail their output. And in some cases they can participate very similar to the way a battery can and, uh, use kind of the same forecasts and participate in the same market products, um, to, um, to kind of save money or, or reduce their overall power bill. And so our technology applies in those cases as well.

Molly Wood: Got it.

David Miller: It, it could apply for, for other, uh, technologies in addition, but it applies best in areas where you have some kind of dispatchability, some kind of control of your output.

Molly Wood: Got it. So it's really just it no matter what the, it's, as long as there is supply that can be manipulated in some form, you can use this combination of kind of forecasting and action to manage.

David Miller: Yeah.

Molly Wood: Okay.

David Miller: That's right.

Molly Wood: How do you see this ultimately as a climate solution, not withstanding and including the fact that you only deal in renewable energy electrons.

David Miller: Yeah, I mean, so we, we are set out to enable the energy transition and the way we see that is energy transition becomes harder in some ways as you have more renewables on the grid because there's, the renewables are all producing at similar times and have these similar risk factors. And so they just have a bigger impact if things go wrong. If we, you know, have a big storm that, um, freezes the wind turbines or have, um, you know, uh, bad weather, that that means solar production is gonna be low for a period.

And so, uh, as, as we move to a higher renewable grid, our view is that we need this better kind of forecasting of output and risk management to allow us to still have, uh, you know, reliable power because, um, the grid only works if it is cost effective and reliable. And we also really want it to be clean. But managing those first pieces is critical in order for the third one to have a chance of success.

Molly Wood: Right. And so it, one, among the many things you can do is, is incentivize that usage, but also prioritize that usage, it sounds like? So we're not in a situation where we have to like fire up peaker plants to manage load growth because some of that variability is accounted for by a solution like this?

David Miller: Yeah, that, that's right. I mean, if, if you're, um, one way to say it is, you know, uh, it's, it's, it's very, I mean, firing up the peaker plant at the last resort is very expensive. And so if we tell people you're gonna be served by, you know, clean energy, but then we misforecast what's gonna be available and we end up buying very expensive alternatives. You know, they're not gonna be pleased with that as, as an outcome. So doing this effectively, um, is, is, you know, sort of critical for the success case of people to be pleased with the power supply they're getting.

Molly Wood: Right. What does the, um, what does the grid itself need to do some of this better, like to, to manage things like maybe even oversupply and overproduction. You know, like what is, you're a, you're sort of a bolt-on solution in a way to ease this transition. And I just wonder from your perspective, what, you know, what the grid at large, this crazy like machine slash several different machines could use to make even your job easier. You're making its job easier.How could it make your job easier?

David Miller: Yeah, I mean, I think there's, there's different ways of, of going about that. I think, um, you know, one piece is just at the very basic level, just having kind of wholesale electricity markets that are based on, um, you know, uh, occasional marginal pricing. It makes a big difference.

So there are parts in the country where we, uh, have dispatch through markets and there are parts of the country where dispatch is done by, uh, you know, some independent entity that is, is not a, a, a system operator. Um, and so, uh, moving toward a market, a, a, a model of more markets is, is already helpful.

Then once you have the market, it's being very transparent with the data. So in order for any party, including us to, to do this, you need to have a lot of data inputs that come from the utility itself. So what, you know, what happened on the grid yesterday and you know, a week ago to understand, uh, if 

we were making sure that our forecasts are correct to get that ground truth.

And some of the grid operators today are better at that than others, uh, in terms of producing that data reliably and in, uh, usable formats. Um, so that's the second piece.

And then I think, you know, getting those kind of basic fundamental, but most impactful building blocks outta the way. I think the, the grid operators themselves, um, you know, play a role in, in, in terms of the market operations, in terms of the forecasting. Um, they have their own forecast that a lot of the market participants use. Um, they also use those forecasts internally for some of the ways that they dispatch the power fleet. And I think there is an opportunity to make those, um, more accurate and more, therefore, efficient in terms of their operations, um, by incorporating kind of better forecasting methodologies. And the outcome from doing that would be improvements in cost and reliability

Molly Wood Voice-Over: Time for a quick break. When we come back, we'll get into the data center situation and the extremely sexy topic of controllable load resources, which takes us back to all those VPP conversations we've been having, but like a little bit on steroids.

Welcome back to Everybody in the Pool. We're talking with David Miller of Gridmatic.

Molly Wood: Um, all right, let's, now let's talk about load growth and then specifically in your customer segment data centers, like how big, you know, when you talk about commercial industrial customers, how big a part of your customer base is data centers and more, and how big a part do you see it becoming?

David Miller: Yeah, I mean, um, it's, it's, um, today data centers are an important segment of the market. They are not a majority of the market, but they're growing. And so, uh, we're seeing, uh, between now and 2030, uh, forecast that roughly half of the load growth in the US, uh, is forecast to be data centers.

Uh, so that's a huge number 'cause it's never, you know, before in the past, data centers have been 10% or less of the, the load on the grid and, uh, excuse me, they've been, they've been less than 10% of the load on the grid. And so half of the new load going to data centers is a big change.

Molly Wood: Mm-hmm. Um. What is different about, what makes them different from in, in addition to just the sheer demand, right? Like, which is staggering. What else makes them different from other customers on the grid?

David Miller: Yeah, they're, they're quite different, um, in terms of the interconnection process.

So traditionally, when someone wanted to connect to the grid as a, as a consumer. You know, there wasn't a ton of uncertainty around that process. Um, you can imagine like, um, a university wants to build a new hospital and the location of that hospital is, is pretty fixed. The price of electricity or the availability of electricity, it's not like the primary decision point for that project. Uh, they have I'm sure a lot of other factors that they're, you know, concerned about.

And so the grid operator would study that project and, you know, make assumptions around how much power it needs, um, and then determine if, if grid upgrades are needed and if they're needed, they just, you know, they go into the schedule.

Molly Wood: Yeah.

David Miller: That might take some time for the project to get on schedule.

And with, with data centers, the development factors, they're all flipped. Um, the price of electricity, especially the availability of electricity really matter. But the location could be a lot more flexible from the owner's perspective. And so what we see now is there's a lot of data centers, um, you know, working on multiple projects simultaneously. And while they're still kinda learning the cost and availability of, of power and, you know, de-risking other factors in development. And so, but, and they also wanna move really quickly. They have more demand for services than, than available power. Um, and so, you know, they're, they're just going everywhere at once.

And the impact is that, um, you know, we're just seeing these, these big, uh, queues or, or waiting lists, uh, for interconnection that are driven by the data centers. In, in Texas right now, the waiting list to plug into the grid is 230 gigawatts. And for context, the state only uses about 90 gigawatts on its hottest day.

Molly Wood: Hmm.

David Miller: Um, so it's like three times the amount that of the state that's currently being used is, is is waiting in line to plug in. I mean, look, nothing like that a year ago. It's been massive growth of that interconnection queue and of that queue, uh, data, data centers are like 70%.

Molly Wood: Hmm.

David Miller: And so not all that is gonna get connected. Um, but even so, it just puts utilities and and grid operators in a really tough spot. 'cause they're inundated with these applications. And the current methodology assumes that like they'll all run, you know, and, and pull power from the grid at full capacity all the time, which is just not feasible for the grid to support.

Molly Wood: Right. Okay. So then where does Gridmatic come in?

David Miller: Yeah, so we can help the data centers operate more flexibly. Um, and that helps minimize the ongoing costs of power and potentially can help with the interconnection process as well, although that's a more nascent area. And you know…

Molly Wood: So you're saying once they come online, like, so are you saying like, leaving aside the interconnection issue, once a data center comes online, you can help them be more of a flexible producer of power or consumer of power, rather?

David Miller: Yes. And, and potentially, uh, and this is an area that's under discussion now, but potentially we can help them, um, if, if they can be, you know, if the, the grid operator believes that there'll be a flexible, uh, consumer of power that might accelerate the interconnection process as well.

Molly Wood: I guess then my last question is just about flexibility. Like data centers historically and so far have not been legendary for their willingness to be flexible when it comes to their power usage. Where do you think that's, where do you think that's gonna go? Especially considering like public pushback, possible regulatory pushback, you know, maybe you've gotta accept a different way of being. Maybe not every single data center needs five nine's reliability all the time.

David Miller: Yeah. I mean, it's, it's gonna be, this is a key area of innovation that's being worked on right now, 'cause I think the data centers recognize that this is the situation they're in, that they have to figure this out in order to, um, get the, uh, power supply or, or the grid connection that they're, they're looking for.

Molly Wood: Mm-hmm.

David Miller: Um, and so, you know, I think there's, there's lots of different approaches. There's, you know, flexible compute, there's onsite storage or, or other onsite, uh, generation. Um, I think it's kind of, uh, right now. Lots of different, um, areas of innovation and, um, I think we're optimistic that, that those will bear fruit.

Um, but what we're focused on mostly is kind of the, the mechanisms by which the company's doing that can, can be successful. And so there's a, there's a program right now in the Texas market called the Controllable Load Resource Program that is getting off the ground. And uh, I think that is an example of one where we're already seeing customers trying to figure out how they can be flexible enough to be eligible to be into that program.

Molly Wood: What does, what does that program do?

David Miller: Yeah, so there's a specific, um, like resource designation and it refers to, uh, a load or, you know, just electricity consumer that is treated closer to what a power power plant is treated. Um, so like in the past, uh, you have demand response, which is like loads that are grid resources, but they're really kinda manual things where they get a call once or twice a year and they turn off. Um, and it's ultimately in their control of what they're doing.

The controllable load resource program flips it so that the grid operator is control. So the load has to receive a command from the grid operator every five minutes and keep its output at that power level, uh, every five minutes and make sure the grid operator knows what it's doing, um, all year round.

And that's, you know, much harder to do, um, than the traditional kind of curtail a couple times a year. And so in the past, there've been very few, you know, loads that could actually do that. Just a few very specialized facilities. And then actually the main one has been grid scale battery storage, which, uh, in the past was treated as a controllable load resource while it was charging.

So that's where Gridmatic kind of got our start in this controllable load resource space, is that we've been operating batteries as controllable load resources for the past three years.

Molly Wood: Got it. And then now you're saying as you grow, you could operate, you can control all the loads.

David Miller: Yeah. I mean there, so it's, it's a hot topic right now. Uh, it's a super hot topic because it hasn't been realized, but there is a lot of hope, uh, including from ourselves that that will change, uh, with within this year.

Um, so there's some regulatory changes that are trying to make it easier, uh, for, um, uh, consumers to become controllable load resources. They basically get rid of this frequency response requirement that batteries could do, but it's really hard for everyone else to, um, so that could make more, um, customers control load resources.

And then in addition to that, control load resources, uh, will get a cheaper price of power if they're located in the best locations. Um, so right now, you know, if you're next to a, a super cheap wind farm that's being curtailed, you don't get to take advantage of that really. But if, um, as a controllable load resource, um, that's kind of the carrot is that you, you could get to take advantage of that low price at that location compared to kind of the average price that all the other consumers are paying.

Molly Wood: Got it.

David Miller: And then controllable load resources can also participate in reserve services, make more money.

And then the last one, which is kind of the, the big if true category is that, uh, potentially it will allow data centers to interconnect faster. Because right now the grid operator kind of studies these customers as if they're gonna be always on, including at the worst times. But if you could study them where they're not always on, whether they're under control of the grid, this could allow them to connect faster. Kinda like, you know, getting, getting, getting the fast pass lane, um, and, and, and moving faster if you promise to sign up for this category.

Molly Wood: Got it. Okay. That's why, that's what you meant when you were saying that you could potentially help them with the interconnection cues, is that you can say like, look, if you work with us, you can achieve this designation and therefore maybe skip the line a little bit because you're, you're a, you're a, you're not a, just a cost center for the grid.

David Miller: Yeah. Well, well, you're, you're, because you're promising not to consume at the worst times, you're not causing the same impact.

Molly Wood: Right. Got it. It, it's interesting. That seems to be the, that seems to be the key phrase that ties together like a lot of the innovations and conversations that I and everybody else in this space has been having about like, what does it look like if you all of a sudden have grid assets in the form of people's HVAC systems or the electric cars in their driveway that are charging with bidirectional or, you know, charging systems like that, that everything becomes like a, even a house becomes a little virtual power plant.

David Miller: Yeah.

Molly Wood: Like, is that, is that fair that that is the umbrella term that makes it all go on the grid?

Because I feel like the grid has been the missing piece in so many of those conversations. It's like, cool, but what does that mean? How does, how is that digested by the grid?

David Miller: That's right. Yeah. In, in the past when you've had these resources that have had some level of flexibility, they haven't really been used for that flexibility 'cause the grid hasn't trusted it.

Molly Wood: Right.

David Miller: It's like, okay. Great, you signed up for this main response program, but because I can't guarantee that you're gonna shut off, I am, you know, not really gonna count on you. And so the, to make this real and to make it work, we have to have programs where the grid operator can truly count on this resources.

And that could be this controllable load resource program. It could be virtual power plant programs like you described. But it's really that, that, uh, market participation model that is needed. Um, and I mean the, the, the, the impact of it is, is it's just huge.

Because, you know, if, if we look at, um, increases in electricity costs, which is a huge issue. Most of the increase in costs that we're seeing comes from, you know, transmission and distribution costs. And so if we're building a bunch of new stuff and we're building a bunch of new poles and wires, uh, in order to, to, um, which, which, you know, that is what the transmission distribution costs are effectively all the, the great equipment that we can see.

Molly Wood: Mm-hmm.

David Miller: Um, you know, that that's what's gonna continue to raise prices for us over time.

Um, but if we can have this flexibility and the load resources can, can, can turn off at the right times and be, you know, located in the right locations, um, then you, you, you, you, you won't need to build all that equipment and you can actually have more consumers with the same equipment which would bring costs down.

And so, you know, flexibility is, is, is such an important factor because it is the difference between increases in costs and actually decreases in costs, not just bringing things neutral.

Molly Wood: Right. Right. Okay. I love that you're the first person who is like, you're the first person who has closed that crucial loop for me, which is sort of like, okay, I see the vision that we're all generating power and it could potentially be flexible, but it doesn't, but the grid feel, you know, then I go like, okay, well how does that work with the grid? And it's, it's a little bit crickets. You know, like, without that, this vision can't entirely come to pass.

And it sounds like what you're saying is that, that the layer that makes that happen is one buy-in. Either on the consumer side or the, or the industrial customer side, but two, like software. [Laughs[ You know, forecasting and delivery.

David Miller: Yeah, I mean, it, it, um, you, you, you need that forecasting in order to actually do it at the right times and, uh, deliver the promise of what the technology potentially allows.

Molly Wood: David Miller is the Chief Commercial Officer of Gridmatic. Thank you so much for the time today.

David Miller: Thank you for having me. I've enjoyed the conversation, Molly.

Molly Wood Voice-Over: That's it for this episode of Everybody in the Pool. Thank you so much for listening.

Hey, big news. We now have a Discord server. If you don't wanna wait for a podcast every week and you'd like to join this conversation with your fellow listeners, you can find the link to join our Discord in the description here on whatever app you're listening to. It's free. It's open to everyone. It's just so that we can build a community of climate tech, early adopters and evangelists and get to hang out together.

Also, please feel free to email me any thoughts or suggestions you have to in[at]everybodyinthepool.com. And if you'd like to support the show directly, you can become a paying subscriber to this podcast or the newsletter or both. To subscribe to the podcast and get an ad free version of the show, hit the link in the description in your podcast app of choice. Thanks to those of you who already have.

Together we can get this done. See you next week.