AI in Power Trading with David Miller (CCO, Gridmatic)

Hello, everybody. Welcome back to Transmission. It's me, Quentin. And this week, we're talking about one of my favorite topics, which is battery optimization.

Now every single time we do an episode with an optimizer, we get tons of feedback that people really like the episode and loads of downloads, and this one will not disappoint.

We're talking to David Miller from Gridmatic in the US. Now Gridmatic's approach to optimization, they're they're they're big in the US. They got a gigawatt under management, mainly in ERCOT in California.

But their approach to the optimization business problem is quite different. So I think you're gonna really enjoy this. We go deep on AI and their fifty million dollar fund and loads of other stuff. Stuff.

And David is a really deep thinker on optimization. So as ever, if you like this episode, please hit like, subscribe, and hit all those good buttons. It does increase our reach and means we can do more awesome stuff for more people. Alright.

Let's jump in.

Hello, David. Welcome to the podcast. Thank you for having me.

So firstly, if you're listening to this and you haven't come across Squidmatic and seen some of their their content and some of the awesome stuff they put out, then you need to go to the website right now and do that. I just wanna firstly as a as a person who who believes greatly and respects good content, I just wanna take my hat off to you and the team for what you guys put out there.

That's very kind. We feel the same value.

That's a good start.

And so let let's for our audience, if you haven't heard of Gridmatic, David, can you just take us through what Gridmatic does? And then we're gonna talk about the the title of this episode is, is rather grand. It's called AI in power, which could mean lots of different things to different people. So we're gonna talk all about AI and what how AI can change market access and market participation.

But first, what is Gridmatic, and what do you guys do?

Sure. We were, founded by an ex Google engineer, and he saw the need, for AI for the grid of the future and understood that wholesale power markets were really the foundation of that grid of the future. So he began by developing algorithms to trade in the wholesale market's ISOs, and he modeled the grid, using AI models at the nodal level and had success trading. And so now those same algorithms from trading are used for, optimizing battery storage on the grid as well as load resources on the grid. And we continue to do that financial trading, meaning trading not backed by, physical assets on the grid across all the US ISOs.

And as a power marketer, we sell power to end users, retail electricity. And then we hedge that power by contracting for a supply with batteries, with renewables, lock the powers, and and then we we optimize energy storage for ourselves and others.

And so lots of people know about Gridmatic from a energy storage perspective. How does energy storage fit into the business, and how much of what you do is is is battery related?

There's really, three parts of our business, at least in terms of how businesses are traditionally organized in our area. So we're doing trading, we're doing battery storage, and retail. So in that sense, you could say it's a third, although our view is that these pieces are all intertwined and will function as one business over time.

I feel that we should talk about two news articles from GridMassett recently. The the first one is is is fund related. You guys had an announcement around that. We talked about that. And Heather what is you surpassed a gigawatt hour of assets under management. Are those operational assets, and and where are they?

Yeah. So we, have, storage assets in both ERCOT and Caiso, operational as well as contracted in that metric.

And those, are, projects that our battery storage fund that you mentioned has worked with as an offtaker, or service provider. And so we're signing tolls as well as revenue floors for these these the batteries that we're then managing on the behalf of the owners.

And so what's the big vision for for Gridmatic? Where are you guys headed? What does Gridmatic look like in twenty thirty?

Right. I our our view is that as electricity prices really become lower on average but more volatile with the increase in renewables, that there's a need for a new type of of load serving entity, and that's really kind of our marketing business that we're interested in. And so for us, the key thing is using battery storage and just batchable resources to balance a load portfolio and serve customers with low cost power.

And would you would you class Gridmatic as an optimizer? Is that a word that you use for your business?

It is. Yeah. That's a part of our business because we really focus on battery storage optimization, both for ourselves and for our customers. And so when we're doing that, we do it across different models. And one of those models is a very traditional service model where we work as an optimizer on behalf of our customers.

So let's talk about AI for a few minutes then because AI is a big part of what Gridmatic does.

I've been struck by how much time Gridmatic spends talking to the market about AI forecasting.

And so it must be really important to you, and I'd love to go down that rabbit hole. So could we start from the start from the start? How does AI fit in an optimizer? And then why specifically the forecasting problem is that a good one to solve? And then what how do you turn that into more dollars or more ROI for customers?

Yeah. Absolutely. I would start with really the the foundational activity in power marketing, which is forecasting electricity prices and understanding how to participate in power markets. Traditionally, that's been done through physics based models of the grid.

These are powerful models that take into account things like, all the generators and the lines and the fuel prices and efficiencies, and we try to estimate what prices will be on the grid based on all those things, how the grid operates. And that approach works well if you're the grid operator and you have all that perfect information. But for everyone else, it's quite tricky because these models need to be finely tuned and they require a lot of human oversight and ongoing tuning. And if you don't have access to all the data, it's hard to get them to produce accurate results.

AI takes a totally different approach. Instead of trying to actually replicate what the grid is doing, it learns from data and just infers how the system behaves.

And what we've seen is that just as AI has outperformed traditional methods in areas like voice recognition or playing board games or or generating images, it turns out that your AI can really do a better job of forecasting electricity prices as well. And because AI keeps learning and updating, it can really automate a lot of that decision making, which has a big benefit for folks in the market. And so tying that back to the battery storage, you're running a battery. It's it's not just about, you know, buy low, sell high each day. It's really these ongoing complex probabilistic decisions, that you have to do all the time to maximize revenue.

And a person running, you know, a traditional forecasting model, it just can't keep track of all these variables in in in one head. And so this is really kind of the exact type of problem that AI is built for, which is why I think that, you know, as storage markets evolve, AI driven forecasting opposition will become the market standard for for operating old batteries.

And what what's the input data? What's the alert? What's the training data?

Yeah. There's there's a lot of data. Our, our CEO, who I mentioned comes from Google, he he sometimes says, obviously, Google has tons of information. They index the whole web.

But for the problem that we're looking at, you know, we have just much more data than than would ever be accessible to someone like Google because it's very specific but voluminous data that we need. And it really revolves around weather and then also the grid and and what the grid looks like. And so, you know, on on the weather side, we're very interested in AI based weather models and understanding, you know, how to get earlier and eat earlier in terms of just forecasting weather. And then on on the grid side, it's really like, how do you take in the information of the grid and the representation of the grid in a way that is accurate but also interpretable by an AI model?

And we're we're constantly kind of looking at how we're gonna do that. The the methods matter as well as the data too. So, like, just to give an example, you know, one question we might ask ourselves is how do we understand the impact on the grid of a big weather event? You know, if a storm is gonna come, maybe it's gonna produce a bunch of wind, and the wind power on the grid will be producing at all time highs.

Or maybe it's gonna ice the turbines and take them offline entirely.

And those, you know, divergent outcomes could be just a matter of a few degrees of what the storm's gonna look like, and and maybe they're happening at the same time in different parts of the grid. And so how you represent that appropriately in an AI model is really, an evolving field of research.

And how many people are there at Gridmatic solving this problem?

We just passed fifty.

Fifty. And how many of those are engineers?

Well, it's it's roughly, I would estimate something like thirty five. The the engineering team breaks down into a few different groups, folks focusing more on AI models and optimization and kind of software infrastructure, data science. But most of the company is is based in in the engineering groups.

I have to ask you a broader question, which is, in my job, I've had the privilege of seeing lots of different power markets and got to see the UK go from pretty much zero to where it is now, which is, you know, six, seven gigs online.

And one of the debates over the last five to ten years in the UK and that is now being ported to lots of other markets. Of course, you know, ERCOT and California surpassed UK in size now. But one of the debates that's that's just continues to go on is about whether humans are best at making decisions or whether it's humans in the loop guided by technology or whether we're moving to a world where it's fully automated. Gridmatics, one of those countries that has really planted a a a flag on automation. Right? And are you so so, operationally, do you have humans in the loop, or is it is this is this really truly automated?

There are humans, but we try to keep them in the loop as infrequently as possible.

The big risk factor in an AI based model is what if something is gonna happen that the model has never seen before and has no way to predict?

And in the history of GRIMATIC, you know, we've seen that multiple times.

In twenty twenty, we had hurricane Laura, which island did part of MISO. There was twenty twenty one winter storm Uri, which was in multiple ISOs, but especially ERCOT. There's been wildfires in CAISO. There's been others as well. And all these events impacted the grid in ways that didn't have exact historical parallels.

And we were active in the market for these events, and they're they're very hard. And and through these events, which we had, you know, both successes and failures, we've developed a a risk management framework where effectively we're asking ourselves a question, is there anything coming up that we have never seen before so the model can't, you know, we can't accurately trust what the model is seeing or what it thinks it's seeing? And if there is, that's when a manual intervention happens. And that manual intervention is pretty blunt, and it's it's really dialing back the risk level significantly.

But then once the kind of maximum uncertainty fa passes, we go back to that fully automated approach right away, and the models are able to learn and recalibrate really quickly, faster than any human. And these interventions, I mean, they're they're quite rare. They happen, you know, less than once a year in a market. And our history is that more often than not, the model would have done better without the intervention.

No way.

So even though we're trying to do it as infrequently as possible, we still still should be doing it less frequently. But we have to keep that lever in place to really manage these tail risks and errors.

So to build that, are you putting in guardrails for known unknowns? Well, this is a very complex problem. Right? Because you have the known unknowns or the unknown unknowns.

And, of course, you can't. It's, ultimately impossible to size the unknown unknowns. But the tricky thing is that, especially in the markets where you have a big presence in Akron, California, historically, the periods of the unknowns, known or unknown, are where there's been a lot of money on the table. So it's interesting that you say that the the the system or the the AI model would historically have made more in those periods as well.

But, yeah, interested to how how you build the system and define those guardrails and when the human steps it.

Yeah. I mean, at at some level, it it it comes down to risk management, both human intervention as well as quantitative risk management. So the models are constantly running, and they are flagging alerts if there's something that that looks unusual. Sometimes it's mundane.

Right? A data feed goes down. That's that's kind of a a very common thing. And so sometimes you have to intervene for something that just, you know, some some things is broken.

But in terms of these, you know, sort of extreme events, they're infrequent, but they do kind of require a human to to look and say, okay, do we think that the model can accurately capture, you know, the risk of what's going on here. Our CEO sometimes says, you know, quantity markets have these structural dynamics where it's not hard to be right and make money a lot of the time. What's really hard is having risk management in place to not be, like, really wrong at at any particular time. And so that's just a big part of what we focus on on the quantitative risk management is we do look at metrics.

We look a lot at different tail risks and risk metrics that focus on the tails to understand, you know, is there anything that this model, you know, is not seeing properly? So an example could be, you know, we're seeing let's say we're in California, and it's, you know, the hottest day we've ever seen. And the model doesn't really see any expectation of the the the price cap being hit. Well, it said something's wrong there.

Right? Like, why is there maybe it's not the base case, but, like, why is there not a scenario where that's happening? That would be a trigger where we'd say, okay. Let's go in and look and see.

We need to investigate further. And while we investigate, we need to pause what's happening to understand, you know, why the potential range of outcomes that we would expect to happen here is not being shown.

Let's let's talk commercials a second because you're a commercial guy. Gridmatics offering is a little bit different to a traditional optimizer, if you like. Right? So you you do the the the revenue share, market access, control the battery.

Of course, your your angle on that is is AI related amongst I'm sure it's amongst other things in customer service and and all of that. But from from speaking to the the team and seeing what you guys are doing in the market, it it's very technology focused.

But I was really struck to see that you launched a fund, fifty million fifty million dollars, right, to to underwrite flaws, and you also offer a a full toll. So you do everything from the bid optimization, the the the the data instructions to revenue share to full top. And I think that's quite quite interesting. How how do you manage those three different contract types and different types of customers? And, if I may ask a a tricky question, which is, yeah, how how do you prioritize those three types of customers in the system? Sure.

I mean, for us, it it really depends on on what works best for the owner and their capital stack. And sometimes these different contract structures will work best for the same owner because they're looking to mix and match their level of merchant appetite across their portfolio of projects. Right now in ERCOT, we know we're seeing a lot of interest in fixed price offtake and tolling or or floors. But, you know, that that changes over time, and we wanna be available in the market depending on the needs of the owners.

Now the way way we think about serving all these these customers is that we think in any of these relationships, you really need strong incentive alignment. We don't think the model works really well where it's a typical SaaS model and the battery owner kind of play pays a fixed monthly amount to an optimizer or a software provider. The range in outcomes and optimization are are just so wide that the owner needs a strong optimizer. And whether that's us or someone else, the optimizers will wanna be paid based on their performance.

And so, you know, I think the owner should want that too. It incentive alignment really helps for building relationships, but that's key for us. So so in our service contracts, we try to structure it such that we're really only getting paid if we're performing well.

And the same, you know, thing happens in the offtake contracts where we only make money if we're outperforming the the floor or or the toll. That structure keeps us highly incentivized to to work on all these projects. And then on the technical side, you can segregate the activities such that no no no one model is getting more prioritization or more eyeballs than the others. They don't know what the others are doing, so they're each on equal footing.

Yeah. That was my next question. You must get this all the time. If you're selling a service which provides the instructions, you know, the the price the scheduling and the pricing for bids, as well as the the full control market access optimization to the point where you actually have skin in the game, right, because you're offering full tone and and revenue flows.

How do you, as a business, prevent prioritizing the second mark there?

The real answer is on the technical side, there's just a wall between the models. And so, if we operate two batteries that are owned by two different counterparties, each one doesn't know what the other battery is doing any more than it knows what all the other market participants are doing from disclosure data after the fact. So there's no sharing of information across those. And that's really, I think, the fundamental key to prevent any kind of either unfair activities or or market power abuses. We also do things that are maybe more for optics. So, like, on our trading side, we won't trade on any locations where a battery storage exists. Even though that model doesn't know what the battery is doing, we just don't wanna even, you know, raise any eyebrows because it's it's there's plenty of nodes straight on, and it's it's not worth any kind of controversy.

Yeah. It's it's it's a it's a tricky one, isn't it? Because even if you're you have to play it so careful, the impartiality, because even if you're doing everything right, the the perception becomes a reality, doesn't it? So it's awesome to hear that you have those kind of rules in place. Okay. So you've got a gigawatt portfolio, something like that, right, of grid scale batteries.

What's the proportion of them that are taking feed optimization as a service at a scheduler versus revenue sharing versus a full toll. I I know you can't you can't share commercial data, but I'm just wondering what's the percentage split between in your in your business between those states?

There there is a mix. We're seeing it, vary a bit by market. In general, in in ERCOT, there's much more interest in offtake because there's no forward capacity market. Owners that have a lot of ERCOT exposure are are really have a strong desire to have some level of contracted revenue.

In KISO, there is interest in offtake for folks that are looking for to kind of combine an energy ancillary service offtake with an RA contract. But because the RA contracts already exist and, provide, you know, a large portion of the contracted revenue for the project, There's a little bit less of a pain point. As a result, we're seeing in that market, a lot of the owners are looking more towards service contracts, and that's that's our portfolio reflects that.

Coming back to your approach, your AI first approach, my my intuition is that you have traders who can trade the market and control assets, then you have a an an AI solution that can do it. And they're operating factories in a similar way, but one of them makes more money. Right? This is the big fight.

But does AI take or your solution, does it take different types of decisions to what a human generally would? Or does it mimic you, but does it behave like a really great I guess my question is, does an AI system behave like a really awesome human trader, or does it behave on something completely different? A bit like I don't know if you've you you must have seen the, the DeepMind documentary about the the Go. Right?

And the fact that that that move, you know, that move in that documentary that totally blew the socks off the entire Go community.

Does an AI optimizer do anything really cool like that?

Sometimes. Yeah. It it, you know, it really depends on what's happening in the market. There are certain days where it's it's not so surprising what's gonna happen.

And so the AI is, whatever, participating in the ancillary service or selling the energy at the at the peak hour, so of similar to what a manual operator will do. But there are other days where there are weird things that happen, particularly when there's big divergent spreads between the day ahead and the real time market. And so sometimes we see strange things like, oh, the power prices were high in this hour, but we sold the reg down and had the charge during them. Like, why do we do that?

Well, actually, it was just selling reg down because it was trying to manage its state of charge for, you know, optimizing some other ancillary service that were coming later. That makes sense after you think about it for a little while. So we have these cases where you have to as a someone who has, some trading acumen could look at it and have to think, okay. Why is it doing that?

And you really have to investigate, okay. What was the forecasting? What were the all the different components? And it is hard for that to all stay in in one person's head.

And so I think in those cases, we do see things that are a little different than what a typical manual trader would do.

You're operational in ERCOT and Caiso right now. Interesting that you said that ERCOT, the market's more interested in off taking tolls.

I thought that was really interesting. Could you talk a little bit more about that and how that's changing? And is it is this cycle? Does risk appetite shift back and forth in in your experience?

I'd imagine it does.

Yeah. I think a big portion of it is is reactive to recent price levels in the market. We're, an ERCOT Moto subscriber, and we pay close attention to, the Moto best index in ERCOT, which shows those revenue performance. And, you know, we we love the service you guys provide, and and we use it for additional analysis internally. And I think, you know, the headline for twenty twenty four that mostly gets picked up is that storage revenue in ERCOT was down relative to previous years. And I think that's to the point about offtake and and and and service models. I think that's impacted a lot of desire for for revenue certainty and this balance between fear and greed.

But for us, there was another factor that really jumped out, and that was that there was a really wide range of revenue performance across the fleet in the market. We looked at the batteries that were just operating for the full calendar year of twenty twenty four And we saw that among those, I mean, there were batteries that made on average eight dollars per kilowatt month and batteries that made less than eighty cents a kilowatt month I mean, it just got us really interested to dive deeper into what's driving the differences. So, you know, we saw a lot of contributing factors, but we saw the biggest factors were battery location, meaning sighting, and battery optimization, so the the optimizer quality. So I think I think as operators are struggling between those two things, some of them, as a result, are are looking for offtake to kind of derisk a portion of their activities.

Let's talk about location for a minute. Location matters so much in the markets you operate in. And no doubt you spent a lot of time thinking about this. So what can you share with our audience about location of of batteries?

And how should we think about location?

Clearly, you know, developers know that it's really those volatile nodes that are interesting for storage. But our data looking at at twenty twenty four shows some complicated results. We wanted to look at the Texas market and and isolate, locational value ignoring other factors. And so what we did is we created a top bottom or TB, baseline for each of the existing battery nodes in Texas.

So for example, for a two hour battery, we create a TB two baseline which just looks at and what if you sell energy at the top two priced hours and buy it at the two bottom priced hours? To simplify baseline is kind of what a typical battery should be able to achieve. It's not a perfect metric. You have to normalize it for battery duration.

It has other limitations like it requires perc foresight, ignores efficiency and other revenue streams. But we think that focusing on kind of this TB metric or as a baseline using energy spreads is a good baseline because now most of the revenue for batteries in ERCOT is coming from energy. And so we looked at these t b two and t b one values for all of the batteries in ERCOT and found interesting results. The the top t b two opportunity for an existing battery in ERCOT was this volatile node in South Zone in twenty twenty four that had, a t b two baseline of ten dollars a kilowatt month.

And the lowest, t b two value was a low volatile node in the North Zone that had it was revenue of of four dollars a kilowatt month. And so the opportunity of just picking the best node versus picking the worst node in Texas was, like, roughly a two point five x increase in revenue, at least from the opportunity of the baseline, which is, you know, a pretty pretty big number in terms of the range of of possible revenue outcomes.

Another interesting note was that the relative value changed over time quite a bit. We look back over over history and in twenty twenty two, if we're looking at the different zones, it was really Houston and West that were tied for the top kind of volatility values measured by this t b two baseline. But in twenty twenty three and twenty twenty four, it really changed where West stayed as the highest volatile node with the most opportunity for for battery revenue. But Houston actually had the lowest TB two value, lower than the south and north zones, and had a big negative impact on batteries located in Houston in twenty twenty four. So the takeaway for us was that the this backward looking volatility analysis isn't that useful. You really need our unique insights about what future congestion is gonna look like to be good at exciting batteries, which is it's a very challenging problem.

Hypothetically, a developer comes to you, David, and says, alright. Which nodes are hot? Apart from doing a, you know, a fundamentals model, we could talk about that at another time, which, of course, in Moda, we spend a lot of time thinking about that. What what's your approach? What's the framework of thinking about where nodes would be hot nodes?

What are the what are the things that you think about or or look at?

Well, we do have to think about kind of what's what's the time frame for which the question's being asked. So if it's, you know, what's the next ten years of asset life or or for a total contract, that's one thing. If it's, you know, really kind of where should we focus our efforts on existing portfolio the next year, it's quite a different thing. And for us, you know, we we look at a variety of different model types.

I think for the short term, we're we're strong believers in these AI focused models. And even in the medium term, I think there's big applications. As you go further and further out, you really do have to rely on fundamentals models to model the big changes that are gonna occur on the grid. And so, it's a big research area for us of just kind of understanding how to build best in breed fundamentals models that kind of include some of the the the large datasets that we use for the short term model.

So we do that. We model that for ourselves.

Well, I guess you especially. Right? Because if you've got this fifty million dollar fund that you're underwriting long term toll agreements or floor floor price agreements or whatever you wanna call it, then you've really gotta have a view on the future if you're gonna underwrite these. Not only do you have to optimize them, but you really need to have a pretty strong so, you know, he has a strong conviction in the value of those notes. Interesting.

That's right. I mean, that's that's that's, that's what it takes to be, you know, be a market participant trader because we're we're doing that in our storage activities. We're doing that when we write multiyear retail contract. We're doing that when we trade speculatively on ice. It's it's really a core function of the business across the different business lines.

It's different to say an optimizer that signs a twelve month contract and says, I'm gonna make you as much money as you can make in the market. Ultimately, if it's a revenue share, then, of course, their revenue goes down if the market goes down. And that's actually a tricky thing for to optimize this forecasting cash flow is quite a tricky problem in itself. But if you're gonna sign long term total agreements, then you have you you have so much reliance on your your medium and long term view of those notes. Fascinating.

Sorry.

Okay. So following on on your point there about analyzing TB twos and nodes, when you think about the value of optimization and when you're selling optimization or when you're analyzing the market to to to figure out what what you know, what's the true alpha, how do you do that? What did you find?

We use the the same TB baselines to evaluate kind of optimization performance relative to the opportunity, for revenue at a specific location. We call this TB two uplift in case of a TB a two hour battery. And so it's just the achieved revenue divided by this TB2 baseline that we create, looking at the node.

And so we can then rank batteries by their TB2 uplift to account for the comparison between optimization performance of different batteries normalized by the location. Again, it it's not perfect. You really can only compare batteries of similar durations. So we could do a t b one comparison or t b two comparison, but not between the two. But regardless, you know, if we just compare all the two hour batteries, we saw that in twenty twenty four, some of them made a less than half of the value of their TB two baseline, what was the opportunity at their location, and some of them made, you know, thirty percent or more above that TB two baseline.

And so the spread between the bottom to the top in terms of the TB2 uplift is a little over two point five x in terms of revenue, which, coincidentally, was was very similar to what we saw in terms of just the TB2 spread of locational differences.

And that tells us that, you know, there's really big spreads between locations on the grid for batteries and also for optimization conditional on location.

And it means, you know, if if you wanna operate a battery, you need to do multiple things really well. You you can't just focus on sighting or just on optimization because they're both so important that to be good, you need to be really good at both of them.

And then where does gridmatic start and start and finish? So if, you know, if I'm an asset owner and I wanna sign a contract with gridmatic to to optimize my battery, do you assume that you can control my battery via API? And do you assume that my battery is gonna give you a certain quality of data?

Or are you willing to take a bet on less capable systems? Or do you put hardware onto sites? You You know, what what's your approach on integrating with battery assets? How do you do that?

It can be a nightmare. Right? This is why I'm saying it. It can be an absolute nightmare.

And it did this can be for an optimizer, the integration part on getting quality data from assets could end up being kind of a third to half of the the OPEX of the business, just with all these different types of assets. So as a as a as a business, how is GridMagic thinking about that problem?

So we we're a a a queasy intercut scheduling entity. As a result, we're we need to connect to the battery and have a SCADA connection to the physical plant that follows the standards and requirements, laid out by the market operator. And so that part of it, you know, it can be a challenge, but everyone sort of somewhat knows the rules. And so, you know, the battery operators need to be, providing telemetry data that meets the requirements of the market.

And that's actually sufficient for us to operate a battery. We don't need additional connections to site beyond that necessarily because we're not necessarily controlling the battery directly. We're submitting bids to the market operator and the market operator is then submitting controls the battery. Now we do often try to get more telemetry and control to the site, and that's because there are additional data points that provide value beyond what the grid operator requires.

These are things related to battery health, cell balancing, these types of issues. And if we get additional telemetry for those points, we can also, you know, potentially conduct activities that maintain battery health in a way that's that's better for the system. That can be a win win for us and and and and for the battery owner.

You guys should do an integration or integrations with some of the battery analytics companies, some incredible companies out there doing cell health work. And if that's a real value to to your system, then perhaps you guys could get integration with them.

Absolutely. Yeah. We're we're following that space closely. It's it's, I think we're we're a prime candidate to be a user of those services.

David, I've got to ask you about this fund because it is very unique, and I've got lots of technical questions about it. I'll see where we get with it. So how how does it work? So you've got a fifty not as as well as being an optimizer, you've also got a fifty million dollar storage fund or an optimization underwriting fund or I think you call it a storage fund.

Right? So how does that whose money is that? And what what does a fund do? Can you talk us through that?

Sure. Effectively, the fund serves as credit facility. So we have outside limited partners. We're also one of the limited partners in the fund.

So we put up some of our money. And that money goes to collateralizing the offtake contracts with counterparties. So we're not investment grade counterparty. And so when an investor, a debt provider, for instance, to a project looks at the offtake contract with someone like us, they need some assurance that that contract is gonna be worth something.

And, and there's typical ways to handle this are are posting collateral. The the owners is typically also posting collateral to us, to make sure that they're, delivering the project.

And so that requires capital. So we use that capital to to to to post the collateral, for to sign a contract. And then, you know, we try to make money from the operations of the battery, and the money that we make from there serves as a return to the investors based on the collateral that they posted up from.

So when did you guys get the idea to do this? Because this is this is one of the the big problems with offtakers, especially if you if you wanna be a start up in this space. Even if you have the best products, sometimes the best product doesn't always win because it's about balance sheet strength.

But Gridmatic has gone sort of round the back door to provide balance sheet strength through this fund, which is really interesting. So when did you get the idea, and then how long did it take you to raise this raise this fund?

Yeah. Well, so we this is our second fund. Our first fund focuses on the, the financial trading, not backed by assets.

That fund launched in twenty twenty one, so we had raised it before that. Then the second fund we launched at the beginning of twenty twenty three.

So we had the idea, you know, really from near the beginning of our battery businesses, this was a potential option for how we would, you know, go to market as a a provider and optimizer of batteries that wanted to, you know, use our trading capability to have more skin in the game, but not looking at, you know, it being a a an asset owner ourselves.

So what does fifty million get you? How many megawatt hours can you can you back with fifty million dollars?

It's sort of a complicated formula because it depends on lengths of contracts and kinda size of systems. But the rough order of magnitude is something like five hundred megawatts of batteries that that could be backed.

Say gridmatic ten x is hopefully more, but let's say ten x is from here. So to back that, to do a ten gig portfolio, you really need, like, a billion dollars.

We need to underwrite that. Right?

Yeah. Absolutely. That's where we wanna get to next.

So you so as well as being a an optimization business, you end up being a capital custodian, you know, a fund management business. How do you how you set up to do that? What does the organization look like for that? Is it totally separate arms length, or is it within there must be all sorts of regulations and rules about this.

How how does that work?

Yeah. So, we operate two commodities trading funds that are regulated by the CFTC. So we have our finance organization that has folks who know much more about that regulatory regime, than I do and and have to maintain the books accordingly. But I think that is, you know, that's the reality that that energy is a capital intensive business. And we think that, you know, providing the most value that's needed, does require us putting up capital and and absorbing the risk from owners that that that that, you know, that's their model is to is to share that risk with counter parties.

It's absolutely fascinating. I'm sure that, some of our European listeners will be thinking, when's the when's the when's the next optimizer in Europe gonna come up with the fund? Watch this space. So now here's your opportunity to plug something. Is there anything you wanna talk about to our listeners about Gridmatic or some news or an announcement? And then we're gonna get to my favorite question, which is about your contrarian view. But first, any news to share?

I think just quite simply, you know, we're we're in the market, as a service provider, an offtaker, for tolls, for physical revenue floors, for the US ISOs for batteries. And so if you're a battery owner interested in these services, please reach out to us. We're happy to talk.

Okay. And now to your contrarian view, David. What what do you believe that not a lot of other people believe?

I think the the landscape of storage IPPs or independent power producers, today, it looks a lot like the renewable IPP landscape, but in the future, it will look much more like the gas IPP business. On its own, that might not be a remarkable statement, but I think some of the impacts of that view, I think, would be considered constrained relative to how we've seen the storage IPP businesses, developed to date. The reason for that, so wind and solar, they're really all about front loaded value. The economic decisions that matter the most happen before the project is built. It's things like siting and procurement and offtake contracting. But then once the project's up and running, there's not that much active ongoing management.

Whereas battery storage and gas are really operationally intensive. A lot of the value is created after the project starts operating. It means that these projects require constant active management to succeed. And this difference, this fundamental difference has, has a big impact on who owns and operates these assets.

In the US, wind and solar ownership is, is fairly fragmented that wind the top five wind IPPs own about thirty percent of all the wind market and solar it's, it's a bit lower. And beyond those, a lot of the owners are financial firms and they, they outsource the operations.

Whereas with gas, it's, it's a totally different story. If, if you exclude the regulated utilities and you just look at gas IPPs, the top five gas IPPs own seventy percent of the installed capacity of gas plants. And part of that is gas has been around longer, but a big part is that in gas, the operational performance really matters and the best operators thrive and the non specialists over time have straggle.

And their customer relationships look really different too. We have large corporations sign a lot of long term renewable PPAs, but they don't sign PPAs with gas IPPs. You know, gas is working with loads or entities, not corporates. And so I think, you know, large gas IPPs, they're not just generating companies.

They also operate retail businesses in competitive markets, and they use a gas as a risk management tool for their own load books and for a little bit of spec trading. And so right right now, for the most part, we don't we don't see that with battery storage IPs. Battery storage looks like wind and solar. A lot of the IPs have come from the wind and solar world, and they're focused on project development, siting, and procurement.

But because the storage requires this active management, I think over time, we'll see storage IPs evolve to look more like gas IPs. And so there'll be consolidation where the best storage operators will benefit and the benefit the the business models that they have will become more intertwined with with trading and retail operations. And so I think, you know, the current storage I could be landscape won't sort of grow and expand consistently the way it has, but instead will look quite different in ten years from the way it looks today.

What about the ownership structure? So in the gas world, you have a a lot more vertical integration.

And so do you expect that to happen in in our world, in storage?

Yeah. Absolutely. I mean, it's it's really the load serving entities that are the ultimate users for these balancing services, whether it's from gas and storage. And so in the gas world, what that means is you have utilities that own a bunch of gas, and you have gas IPPs that then hedge their fleets with a retail book. I I think the same thing is gonna happen with storage where there'll be a lot of storage that's owned by utilities, load serving entities, and then the large storage IPPs will also be thinking about storage primarily as a hedge for a portfolio of operations rather than as a dedicated kind of merchant operation.

One of the things I've noticed in the last few years is that the asset management part of operating, there's been a a a lot of wind and solar asset managers who have turned their hand to storage. And some of the things are the same, you know, going to, you know, be on-site, managing safety, site access, permit to work, repairing inverters.

That kind of stuff is the same. But there's this massive chunk of it, which is so commercial and market driven, which they have to upscale so much. There's been very few of the wind and solar asset management companies who have managed to make that really work.

And I think it's because, as you say, it's the complexity element.

So how does gridmatic fit within that new paradigm then? How do you think about this is really interesting. How how do you think about an optimizer fitting within that world that looks more gassy?

Yeah. I I think there's basically two directions for optimizers to go in. One option is to kinda sell software, it is a SaaS model, to regulated utilities or other storage owners that aren't so highly motivated to care about revenue performance. And I think that's a real business for storage optimizers.

But the other option is to really align very closely with customers because you have to understand that the good ones are gonna get big and focus on commercial constructs that align the incentives really strongly, things like, you know, pay for performance only service contracts or off take structures that share in the rewards. And we've really started chosen that second path. I think I think some optimizers are still kinda trying to serve all the markets. I think that will be a challenge in the long term as consolidation occurs.

Has grid massive got any plans beyond the US?

Well, not today, but certainly all power markets worldwide are of interest to us. It's, certainly an area that we're we're we're interested in, but our active growth today, we still see a lot of opportunity for expansion in the US.

Alright. Well, David, that was fascinating. Thank you very much for joining us on the podcast.

Alastair's no doubt will really like this episode. It's always great when we get optimizers on because being that close to the market, you have to think in a very entrepreneurial way, which is always fascinating.

So Tam's taking the time to speak to us. Good luck. Really try to see what comes next from Gridmatic, and, speak soon.

Thanks so much. Good to chat.