who we're gonna be talking about a very interesting topic on the AI strategy at the company. As a reminder, at Water Tower Research, we're an open access, non-permissioned research platform. You can access, information and research on SES, including all past, fireside chats and research at www.watertowerresearch.com, and you can also sign up for our distribution there. And it can also be accessed, of course, on Bloomberg, FactSet, Refinitiv and all the research aggregators. So excited to have Qichao with us. Welcome today, and I'm ready to jump in if you are, Qichao. Yes, I am. Thank you, Shawn. Great. Well, as I mentioned, we're gonna be talking about AI, and obviously it's been getting a lot of press, you know, across, you know, across the tech sector. But, you know, looking at some of these use cases and how it's getting used in specific applications brings us to what, what you are doing at SES, and a very interesting topic. You know, we talked about in the last quarter, the all-in AI strategy, and, how this positions, SES for success. So let's begin with that, and, and what does that mean, all in on AI? Yeah. Yes, so in the past, we focused on lithium metal batteries, next-gen battery technology. And then going forward, we continue to focus on lithium metal, but also we want to go beyond that, and that is in three directions: speed up the development of new materials, and also speed up the qualification of manufacturing quality, and also improve safety, especially for next-gen battery technologies like lithium metal. So, what we mean by going all-in on AI is, for next-gen battery technologies like lithium metal, we want to really speed up the adoption in three areas: science, manufacturing, and safety. So when you look at taking, I wouldn't call it a strategic shift, but an advancement here, and leveraging the technology that's out there in big data, what do you think the, you know, OEMs are thinking about this? So if I'm going back to kind of what the catalyst was behind you taking the company in this direction, help me understand what brought you there. In terms of the OEMs and what the OEMs— Well, yeah, the OEMs and maybe some of the other aspects of this. You know, I'm curious, particularly how your OEM partners... 'Cause as a reminder for investors that don't know, SES is involved with some key OEM partners and, you know, you're all the way through, you're making some good progress there on the battery side, but I wanted to address how are they looking at AI? Yeah, absolutely. So, for these three, science, manufacturing, and safety, the biggest thing that the OEMs care about is actually safety. And then recently, we also hear more of these fire incidents with EVs on the road. And then when you have more EVs, and then safety is a concern. So the ability to monitor the safety of the car, during charge, during operation, and also during parking, is really important to the OEMs. And the second is, manufacturing. A lot of the safeties are due to manufacturing defects, and then the OEMs are just never satisfied with battery manufacturers' ability to detect defects. And then if there are more advanced ways of detecting defects, that is very important. Then, in terms of materials, the innovation in batteries has always been slow, and if there are ways that we can use AI for Science to speed up the adoption of new materials, then that is also quite urgent to the OEMs. So primarily safety, and then actually, we have been working on using AI to monitor battery health way back since 2017. So this is not a new approach. It just, it's more important with the fire incidents recently. I think, you know, as an investor and analyst, we always look for products that result in an ROI for the customer, right? For what the company is providing. We certainly hear a lot about safety, so I'm assuming that if you look at the adoption curves and the willingness of OEMs or desire of OEMs in terms of safety, it's got to be at the top of their list, right? Because the publicity from stuff like that is just devastating. I'm assuming when they look at, they look closely and would look quickly, and I'm talking about adoption curves and technology, for anything that helps with the safety of a vehicle. Is that fair? Absolutely. Yes. And then obviously there are different ways of improving safety. One is just inherent battery safety, and then also OEMs are looking at BMS, battery management systems, and then ways of improving the accuracy of the BMS. And then AI for Safety doesn't really replace those, it's just another layer of safety on top of that. And so, you know, I think the use case is clear in the passenger side, talking about safety, you know, on the safety side in particular. But what is the motivation with UAM? How does AI fit into that? And maybe just give 30 seconds, if you can, on that effort at the company, because it is a key one, and maybe not all investors know what you're doing there. Yes. Yeah, urban air mobility is quite interesting because it's for flying applications. You want the batteries that are very light. So very high energy density in terms of weight. And the lithium metal compared to lithium-ion, one advantage is basically it's lightweight. And then safety is important for EV. It's also really important for UAM. And then also for UAM, certification will actually require you to have a minimum of reserve. And then also if you can have a lighter battery, that means you can carry more weight, so that's more passengers, more payloads. You can also fly farther. That does significantly improve the economics of the business model. So safety and the lightweight for UAM are very important. I want to talk a moment about the commercialization of AI and how this gets monetized. We always get down to that, and what is this gonna do to revenue and margins, and how do you make money off of it? But even before we get to that, what gives you a unique position? What gives SES a unique position to be able to do this? 'Cause I'm trying to understand better the competitive advantages you have to develop AI for some of these applications. I assume some of it's gonna go back to the data sets and things like that. But what really gives you an advantage to make you the company to do this? Yeah. Yes, actually, if you look at AI for Science, for example, in battery applications, very few companies, whether they are battery companies or material companies or AI companies, or that have complete ability from in terms of ability to simulate all the molecules and have a big database of all the properties of molecules, and then deep understanding of the electrochemistry, the batteries, and then ability to design molecules and then synthesize molecules, and then test them batteries, and then verify. But we have the complete ability on the AI side, on the synthesis side, on the battery side, and the validation side. And also for the manufacturing and the safety, again, we are both a battery company and also an AI company. I think a lot of the companies out there that do this AI are not really battery companies, so they don't really have the data, and then also lack the deep understanding of the technology, but we have both. Because it does come down to the data critically, right? I mean, you know we always hear about garbage in, garbage out, right? So when you look at AI and you look at use cases, you know, for large language models, so much comes down to what is the input, what is it learning from, right? And I guess in your case, that's, that really is... You have the manufacturing data, you have the chemistry data. You've been using forms of AI for battery health monitoring for quite some time, right? This is just sort of come into the limelight, let's say, in terms of the press. But in application, you've already been doing this for safety for some time. So, I guess when you look at your business model, you have a data set that provides you a substantial advantage. How does that data set grow? Because obviously, it needs to get bigger and more incoming, you know, commercial data coming back to you. What's the next step there in the data gathering for you? Yeah. So for lithium metal, we have pretty high-quality data, materials, the number of cells we build, and the number of quality checkpoints per cell, and also the amount of test data per cell. So the quality of data that we have for lithium metal is quite good. And then in terms of how we want to grow that, we really want to collect data from different chemistries, both in terms of different chemistries within lithium metal, different cathodes, high nickel, LFP. Also chemistries beyond lithium metal, because AI fundamentally really does not care about chemistry. So if we can prove that this model is indeed agnostic to the chemistry, to the design, first different designs within lithium metal, then even beyond lithium metal, then we can hopefully grow the amount of data that we have, because lithium- ion still has way more data than lithium metal. And then if getting lithium metal data helps with lithium- ion understanding and vice versa, getting lithium-ion data help with lithium metal understanding, then it'll be helpful. So expanding on that a little bit, I wanted to make sure investors understand there are three pillars or three applications for the AI, and I think you just touched on some of the, you know, science, I guess, that would fall under the AI for Science. But can you talk about the AI for Manufacturing, just a brief description of what that is, so investors understand that, and the other two being safety and science. So when we look at your AI strategy, we understand there are three clear applications, and they each have a different monetization opportunity and commercialization opportunity. So just spend a minute or two on each of those so that we break that out in our minds. Yeah. Yeah, so if you look at AI for Science, basically AI for Science, it focuses on electrolyte, and the electrolyte as a material has always been our core. And then that relies on the small molecules, because we use small molecules for the salt, for the additives, for solvents. And then it's basically the same database shared between lithium-ion and lithium metal. And so what we are doing is we are going to map basically the entire small molecule universe. And then to give you a reference, the last 30 years, the entire battery industry only looked at about 500 unique small molecules for all electrolytes. 500 unique small molecules. But there is a possible ten to the eleventh, so 100 billion possible small molecules that could be used. We're going to map that entire small molecules, and that is going to provide us a very powerful database that you can use to solve any battery electrolyte problems, not just lithium metal. But obviously, lithium metal is the core. But it could also be silicon anode, low-temperature fast charge, because really, again, AI for Science does not distinguish between lithium metal or silicon or lithium-ion chemistries. So that's that. Just a question on that. Yes. So if I simplify it, really what the AI is able to do is to look at combinations and multiple chemistries and I guess you know build up. To simplify it, it could determine, does this work? What is the performance characteristics of this or likely performance characteristics of this chemistry? So it's able to take that large data set and look at all kinds of different combinations and options that it would take you know a human doing this in an engineering or lab environment you know decades right? Yeah. Yeah. And also, first of all, humans have never looked at a database that big. And AI has the ability to look at a database that big. So, it is basically using a database that's much, much bigger than anything that the humans have used. And then, so for example, you can study the molecular properties of a subset of small molecules, and then based on that, it can learn, and it can extrapolate, like, to a bigger universe of small molecules. So that one is very helpful. And then in terms of how we're going to commercialize that, for now, the plan is we basically develop these electrolytes and then license the electrolytes. And the lithium metal, we are focused on lithium metal, and then use this tool to improve the electrolytes, Coulombic efficiency, cycle life, safety for lithium metal. At the same time, we can use this to develop new electrolytes, for customers and then license that. Now, you, of course— have big data sets in lithium metal, but what about the other chemistries? So for AI for Science, again, our goal is to get to ten to the eleventh. Right now we are close to about ten to the sixth, in terms of the amount of small molecules that we have mapped the physical properties of. And then we are in the process of collaborating with a few major national labs and companies on the compute power, because this will require significant compute power. And then to get from ten to the sixth to about ten to the eighth, our goal is ten to the eleventh, but once we get to ten to the eighth, then the AI tool can extrapolate from ten to the eighth to ten to the eleventh. So then we will have the most complete mapping of physical properties of all small molecules. It's super interesting. And to carry it through to the monetization, this is something where you may come up with a new chemistry, right? Now, you would own— Yes And patent that chemistry, and then you would license it— Yes To other battery makers, OEMs, whoever, right? Is that, is that my understanding— Yes Of the economic model of AI for Science? Yes. Yes, and then this part is actually very exciting. It's a model that the pharmaceutical industries have done a lot, like, drug discovery. You have companies that come out with new molecules, and then they will license to, for example, Pfizer, and then they would go through all the manufacturing and licensing. So this basically we will come out with a new formulation that we can license to other battery or OEMs. And then because we also have very good synthesis capabilities, so not only we can predict these molecules, we can actually synthesize, and then not only just give a recommendation, but actually synthesize, actually testing batteries. And then when we give something, it's actually something that's tested in batteries. It's really intriguing because you look at, you know, you hate to say it's a better business than making batteries but in some ways, if you have the ability to build and manage and create these data sets, and you have the background in manufacturing already, so as you said, you can synthesize, you can put it in, does this work in a practical manufacturing application, right? That seems to be the key, because— Yes The battery industry has been full of great ideas that never commercialized because the manufacturing is impossible. Yeah. I assume not impossible, but just it's lab experiments, right? And in this case, I'm assuming, correct me if I'm wrong, but that the AI would be able to also optimize for those and understand which chemistries are best for commercialization for larger scale manufacturing. I assume that cost and understanding of the— Yes Manufacturing process is part of the intelligence, right? Yes, yes. Yeah, when we do AI for Science, for sure, we consider cost, and then also how practical it is to actually synthesize the model. Yes, we definitely do that. I think we could talk about AI for Science all day, an extremely interesting topic. I did wanna go back to the other two applications in Safety and Manufacturing. If we can, let's spend a few moments on those— Yes So investors understand the process and commercialization there. Yeah. Okay, great. And then the other one is AI for Manufacturing, and then AI for Manufacturing is actually really interesting and very practical. This is not some science projects, this is, like, happening today. And then I give you one example, right? For example, most of the fires that we have in EV are due to battery manufacturing defects, not because the cells were designed incorrectly or the materials are not safe. It's actually the cells are safe, just they had manufacturing defects. One example is, for example, CT. Because of cost, companies don't do CT on every cell they manufacture, and they do CT scan, for example, one out of 50 or one out of 100. And then the cells that they do CT scan, they can see defects, and they can catch that. But then the cells that they don't do CT scan because of cost, they don't catch the defects. So there is a balance, and you try to lower the cost, but then you have these safety risks, and then you miss a lot of the defects that you don't catch. Now, with AI, what's powerful is you can actually learn from a few CT scans, and then the model, and apply this to every single cell. So you don't need to do CT scan on every single cell, because that would be cost-prohibitive. Based on the selected CT scans that you do, you can apply the learning to every single cell, and the model can detect defects as if you actually run CT scan on every single cell. So this we have found to be very powerful, and then it can help our manufacturing line to catch defects that otherwise we would have missed. And then now on every line that we have, every lithium-metal line that we have, including the ones, the B-sample lines that we have for Hyundai and Honda, and also the lines that we have for UAM, we basically don't have a line that makes lithium metal that does not have AI for Manufacturing, because it is so effective at catching defects. And this, to us, is very practical. And then we also want to apply this to even lithium-ion lines. And then some of the OEMs have asked us, "Okay, can we apply this to lithium-ion lines?" And then now our process works for pouch stack cells, lithium metal or lithium- ion. And yeah, it's very powerful at detecting defects. And to clarify, when you're talking about thermal runaway and things that happen in the application in the field, that's not generally coming from some type of physical damage to the battery in a passenger vehicle, right? It's coming from something that happened in the manufacturing process that eventually breaks out, right? That comes to bear. Is that correct? And so this is stopping them at the manufacturing side before they ever get into the field. Exactly, exactly. And for example, you could have a very mild soft short as a result of a misalignment or electrical overhang, but then it's so small that your traditional manufacturing process would have missed it as a defect, but this will help you catch that. And then the next part is in the field, right? Or I guess, so you're talking. It's interesting when you say, 'cause even AI for Manufacturing versus AI for Safety, even on the manufacturing side, in my understanding, we're still talking about safety in many cases. Is that— Yes Is that correct? Yes. You're just catching it— Yes On the manufacturing line. Exactly. Exactly, yeah. And then AI for— Yes Safety is monitoring in the field, right? So in the use cases. So if you could talk about how that works as the next level of safety management. Yeah, yeah. And then AI for Safety is once the cells are already in the vehicles in the field, and then it's, it does not replace traditional BMS. And the traditional BMS, you rely primarily on physics-based models, so you have a deep understanding of the cell behavior on the different environments. So what happens is we actually take the charge and discharge voltage profiles during cycle, during charging, discharging, parking. Actually, parking is actually quite important recently, because a lot of the fires occur when the batteries are being parked. Because when they're parked, the car try to save, you know, electricity, try to save battery, and then don't send the signals as frequently. So we basically treat this charge and discharge profile similar to a sentence. So you can think of a battery with 300 cycles as like a book with 300 pages or 500 pages, and then you can train large language model in a very similar way. And then it's going to do two things. One is, it's going to predict incidents. It's really accurate at detecting anomalies. And then, in our own case, we have found, for example, of all the incidents that we had, the traditional physics-based models were able to detect about 80%, which is pretty powerful. But then there are these very few incidents that the traditional physics-based models were not able to detect, but then the AI models were able to detect. So it's very powerful. And then also, a very accurate estimate of the cell health, cell charge. And then that, this SOC, SOH, we'll call it SOX, sometimes that drifts as you cycle. So, for example, this battery after 500 cycles, 600 cycles, your estimate of the SOX becomes less accurate. And then when that's less accurate, then you're applying a current that's less accurate, so you could actually hurt your cycle life. If you have a more accurate estimate, then you can extend the battery life, which is meaningful, because you could extend the warranty. And the way to monetize this is, you can charge a fee, per kilowatt- hour, per 1,000 miles during the warranty period. Most cars will have, say, 100,000 miles as warranty. You can charge that fee, and then you can monitor, provide health monitoring, incident prediction. And then, if you can extend the life, because now you have more accurate estimation of SOX, then you actually extend the warranty period. So let's run with that because that was my next question. I think we've really established the use cases for this, right? And that there's a return for the safety, safety factors for the, for the OEMs, you know, the science side for you to... But let's talk about the monetization of those. You know, everybody loves recurring revenue models. That's a great business model. Tends to get a higher valuation per dollar than certainly other business models in the market. If I'm understanding the use case correctly, if we want to continue with the idea of the AI for Safety, maybe to start, this would be something that could be a subscription model that the OEMs would pay for, and in that process, it would give them better data on warranties, I assume, right? So they know what their warranty costs are. Importantly, less incidences in the field of thermal runaway. But they, I think you touched on it a little bit, but explain that. They pay a service fee to you to use the software, and you would monitor it for them, or is this something that they would have, kind of, you're licensing it to them? So, we're still working with the OEMs on these business models because it is quite new, and it's actually a very conservative field to the OEMs, and the OEMs don't wanna change any relationship with their customers, so we don't change anything about the way they collect data. The OEMs are already collecting a lot of data, so we don't change anything, and they already have their BMS, we don't change anything there, and then so all we do is behind their BMS, behind all the data that they are already collecting, we train these models, and then all the analysis results provided by these models, we will send that to the OEMs, and then the OEMs will plug that into their BMS. And then I think we talked a little bit about already on the science side. This would be you creating new chemistries, looking at new molecules and things like that. So that would become a monetization opportunity where you have a... You literally have a better chemistry that you could then license to battery OEMs, you know, whoever's making the battery. It doesn't matter whether it's an EV. Yes OEM or whether it's a pure battery maker. Is that, that correct? And that would be, again, a licensing-type model? Yeah. So for that one, well, for lithium metal, then we would just build the batteries, and then- Build the batteries. There'll be a in-house R&D. But then beyond lithium metal, we can definitely license to them. I guess from the manufacturing side it would be the same thing, a licensing as part of either a new chemistry into. Now, if they're doing the manufacturing as well of their battery, this would be something they would have on their manufacturing line, and again, it'd be a licensing fee to you? Yeah. So for lithium metal, because we actually manufacture lithium metal. This will be a part of our offer to this joint venture that we talk about with the OEMs, and then with a larger cell manufacturer. So that we will provide lithium metal activities, know-hows, and also this AI for Manufacturing, and then beyond the lithium metal, then this could also be used in beyond lithium metal, lithium- ion lines that are currently operational. Great. We're a little past 30 minutes or coming up to it. So I do like to keep these to half an hour, but wanted to ask one more question, and then we can wrap it up today, and this is, you know, this is from the investor and analyst perspective. I think we've established very well the use cases. It makes sense, right? There are economics to all of this for the customer, and of course, that means margin to you and you know, profit to SES, but when we look at the next steps and, you know, from an outsider viewpoint, how do we... how should we expect to see this roll out into commercialization? Will there be a small licensing agreement with one of your existing OEM partners? I'm just trying to understand how this is gonna show itself when we look at the business in SES over the next 12 to 24 months. Yeah, I think you, we can expect, for example, our current OEM partners will continue to be the most supportive and the first to adopt these new models with lithium metal and also beyond lithium metal. And then in EV, and then in UAM and in drones, we can expect for any customer that use our lithium metal cells, then AI for Manufacturing and AI for Safety will be a part of the lithium metal offer. Okay, so it becomes, you already have a built-in base, I guess, let's say, that you've already been working with. So the logical step would be with your existing partners— Yes To expand the current relationships. You already have a proven relationship with them, so that's gonna be the point to look to. And is this gonna be something that can be, you think is able to be discussed and talked about, or is it going to be something that you kind of have to do and then say, "Hey, we've been doing this for six months with this particular OEM partner?" How would it manifest itself that way? Yeah, I think it's gonna be a mix of both. I think we will have a few announcements coming up with OEMs that we have been working with for a long time. And then this AI for Manufacturing and AI for Safety would accelerate the adoption of lithium metal because it's a new chemistry. And then anytime a new chemistry gets this far into, like, a C-sample and then commercial, there are always hurdles that people are concerned about. So AI for Manufacturing and then safety will help accelerate the commercialization of lithium metal. And at the same time, beyond lithium metal, this will also be used there. Great. Thanks for this today, Qichao. This is very informative. Tip of the iceberg. We could probably talk about this all day. It's such a, such an interesting topic, but I'm sure we'll have you back to discuss it further as things progress. As a reminder, investors, you can check out www.watertowerresearch for more information. If you are sending this around and available on demand, you can use the same link that you did to register for the original event. Feel free to email questions to me or to IR at SES, if you wanted to follow up on any details. And again, you know, thanks for coming in today, Qichao, and appreciate the time and the very exciting stuff. Thank you, Shawn. Thank you. Thank you. And that'll conclude today's fireside chat. Thank you, everybody, for joining us. We'll see you next time. Thank you.