Great. Thanks. Well, hey everyone, thanks for joining us. Final session of the 2024 Evercore ISI Clean Energy and Transition Technology Summit. For those of you who don't know me, my name is Doug Dutton. I'm a VP on the Auto and Transportation team here at Evercore ISI. I'm happy to now welcome SES AI to the stage, a team that we've had a long-standing relationship with since their SPAC debut back in 2022. For a brief intro before I hand it over, Boston-based SES AI is a battery technology company focused on developing long-range lithium metal batteries and building a resilient EV supply chain. The company was founded by Dr. Qichao Hu, who has a PhD in Applied Physics from Harvard. He joins us today up from Boston. SES has entered into joint development agreements with several major automakers and is working toward validating and commercializing their technology.
With that, I'll hand it over to Qichao, and thanks everyone for being here.
Thanks, Doug. So SES, we develop lithium metal batteries and then mainly for transportation applications, so EV and UAM, urban air mobility. And then we do apply two types of AI extensively. That's AI for science and also AI for safety. So the three core pillars to what we do include EV, urban air mobility, UAM, and AI. So we actually start from a very fundamental level. So we start from molecules. We actually map the entire small molecule space to find the suitable molecules for electrolytes. And this is not something that other companies do. And also this is a really core part of AI for science. We map basically the entire small molecule space. And then we also formulate electrolytes for lithium metal using the molecules. And then so today we have access to the world's largest database of small molecules.
And then our electrolyte has the world record in terms of highest Coulombic efficiency on lithium metal, 99.6%. And then we take that electrolyte and we build these cells. And that's where the A sample and the B samples come in. Back in 2021, we announced we had the world's first A sample JDA for lithium metal. End of last year, we announced the world's first B sample JDA for lithium metal. And then when these cells are in the actual EV and the actual UAM, we continue to collect data to monitor health and then predict incidents. And then today we have the highest prediction accuracy for lithium metal. Here's our team. So we're based in Boston, and that's where we do all the fundamental R&D and AI work.
And then all the A-sample lines and B-sample lines and then the UAM lines are split between Korea and Shanghai. And then, so 2021, we announced three A-sample JDAs, GM, Hyundai, Honda. And then now we have two B-sample JDAs with two of them for EV applications. So for EV, just quickly, why OEMs are interested in lithium metal? Because lithium metal can do two things. One is, so you have two types of batteries. You have the economy battery and the premium battery. So when you go from lithium-ion to lithium metal, the new economy battery can have the same range as the old premium, but it'll be cheaper. So this is basically LFP paired with lithium metal. And then the other option is you have high nickel paired with lithium metal. Then you can achieve higher range for the new premium vehicles.
This is why the OEMs are interested in lithium metal for EV, not just for premium vehicles, but also economy vehicles. I'll skip this. Then recently we announced the second B-sample with Hyundai. And this is our second B-sample JDA for EV applications. But one thing that's unique about this is it's actually the first time that we build a B-sample line in someone else's facility. So with Hyundai, we will be building and operating this B-sample line in their facility in Uiwang, South Korea. And this is actually a very good practice because in A-sample, we basically build all the lines in our own facility. B-sample, now we're starting to build this line in their facility. C-sample down the road, we want to get to a situation where we have a joint venture.
So us and the OEMs and likely another large battery company will run this C-sample line together. And also for us to build this B-sample, to build and operate this line in Hyundai's facility, it also gives us a chance to license our Avatar AI for safety model to that line so we can collect quality data. I'll skip this. And then for UAM, so UAM, actually their desire for a lighter battery is even greater than EV. And then UAM, it's not a complete separate development. Basically, when you hit B-sample for EV application, from a technical perspective, it's actually basically your commercial for UAM. So it's a natural adjacent market for our batteries. And then on the EV side, because now we're focusing on the B-sample lines, what do you do with the A-sample lines?
So we're converting the old A-sample lines to make UAM cells. That's what we're doing for the UAM. And then UAM, the fundamental economics is most of these UAM companies will operate in a fleet business model. So it's about dollar per passenger per mile. So if you can make the battery pack lighter, you can put more passengers or you can fly farther. So that lowers the cost in terms of dollar per passenger per mile. And then that improves the profitability of these UAM companies. And then actually from a technical perspective, lithium metal actually performs better under the UAM mission profile than EV. And then UAM as a market is actually growing faster than we expected. Already there are some companies that have received type certification, TC.
And then also more UAM companies are expected to receive TC this year, the upcoming Paris Olympics, also second half of this year, more UAM companies are expected to receive TC. So this market is taking off. And the volume is actually quite suitable for our A-sample line that just got converted. And then we can actually, the higher energy density that lithium metal batteries can provide, it's really something that the UAM OEMs are interested in. And the last part is AI. And then so to make EV and then UAM happen, there are two really important AI that we have to do. One is AI for safety, another is AI for science. AI for safety is we really have to predict incidents before they happen. And this is a really important thing for the OEMs, both EV and UAM. Another is AI for science.
How can we stay ahead of competition? How can we accelerate material development? So AI for safety is basically we treat a battery as a person. So the health, the natural health of a person depends on genetics inherited from the family, nutrients during mom's pregnancy, and then your lifestyle. That's a person's natural health. And then same thing for a battery. A battery's natural health depends on the cell design, depends a lot on the manufacturing defects. Most of the recalls and the incidents that we see in the fields are due to manufacturing defects. And also vehicle operation. Do you have an abusive driver or do you have a nice driver? So one thing that's really interesting is we started implementing Avatar in 2022 in A-sample.
So the more number of cells we make and the greater number of quality checkpoints we have per cell, the more data we have to train this model. Because no one in the world has any experience about quality manufacturing for lithium metal. It's just not done before. Our quality team has very experienced people from the lithium-ion industry, but they don't have experience with lithium metal. No one has that experience. So for us to accumulate this data, then we can train the model and then be able to say, okay, in this manufacturing process, this step has a greater impact on quality than this other step. For example, your electrical alignment impacts your quality more than your jelly roll hot press pressure . Things like that. These are things that in the lithium-ion industry will take 20 years to accumulate.
We can do that in about one year by collecting data and then training this model. So this is something that we did in the A-sample line. We made less than 1,000 cells a year. Now we're making 1,000 cells a month a line. And then so we have, and then also the number of quality checkpoints went from less than 200 per cell to now more than 1,000 per cell. So one of the biggest reasons that we do A-sample and B-sample is to build the cells and then collect the data and then train this model, train this Avatar quality model. This is something that we're going to license to the B-sample JDAs in their lines and then down the road also to other manufacturing partners. It's really important to ensure quality and safety. And we actually did some field testing.
We put 14 of the A-samples, so about 5 kWh, into an actual drone. We actually accumulated about 75 flight hours of actual drones using lithium metal modules. Then we collected the mission profile data and then also used that to train this model. Then the other part is AI for science. So now our current generation of lithium metal is in B-sample. What about the next generation and the generation beyond that? How can we develop future generations of lithium metal much faster? Because lithium-ion took a long time to get to where it is today, 30 years. And it's actually really impressive progress, but then also it was quite slow. So to put things into perspective, all batteries, lithium-ion or lithium metal, comes down to small molecules.
Then obviously the entire universe of molecules is infinite, but the universe of small molecules is about 10 to the 16. Then the universe of small molecules that could be used for batteries is 10 to the 12. The last 30 years, the entire battery industry only explored 10 to the 3 molecules. And out of those, we've already identified molecules that can achieve 99.6% Coulombic efficiency. So next for the future generations of EV and UAM batteries, we're going to map even more molecules so that we can find even higher Coulombic efficiency electrolytes to further improve. So it took 30 years for lithium-ion to get to where it is today. I think for lithium metal, by using AI and machine learning, we can probably do that in a few years. So that's just the pace of learning will be quite different.
So yeah, so for this year, the focus is still for EV, basically complete the B-sample lines and then making sure we install Avatar on the lines so that we can collect data. One of the most important things for us, the B-sample, is to be able to collect the data so we can train the model. And another for UAM, it's a near-term earlier opportunity for revenue, which we think we can get to first half of next year, is to actually supply UAM cells and modules plus Avatar, this software to UAM OEMs. And then just in general, for both EV and UAM, our goal is to improve by using Avatar, our ability to improve the manufacturing quality and also the ability to predict incidents before they happen. So thank you.
Excellent. We can open it up for some Q&A now if anyone in the room has questions.
Could you discuss the cost structure of the cell?
Could you discuss the cost structure of the cell and economics and how you could see that progressing going forward? Because I know the roadblock for a lot of next-gen battery tech is, it's expensive. So anything on the end of the path there?
Yeah. So even starting in A-sample, we've been very transparent with the JDA OEMs about the cost structure. So if you look at the COGS, most of that is manufacturing and then the BOM. Manufacturing assembly process is the same, right? So these lithium metal cells are made using lithium-ion process, like a pouch format, tabs on the opposite side, and then stack it. And then within the BOM, cathode is the same. Cathode is actually sourced by the OEM. So the two biggest factors are basically exactly the same. And then the only things are electrolytes and the lithium metal anode. Electrolyte, everything is totally new. Like the molecules is completely new. We actually discovered it, make it in-house, not something you can buy. But the manufacturing process is a scalable chemical process. So that one we don't expect to be that different.
The anode is different, but two of our partners, Applied Materials and then Tianqi Lithium, two of our shareholders will help with the sourcing of that. Also, anode itself as a percentage of BOM is quite small. Overall, in terms of dollar per kWh, lithium metal should be similar to lithium-ion using the same cathode and the same assembly process.
Thank you. What about on the Avatar AI? It sounds like that's pretty promising in terms of discovery and in terms of refining and getting lithium metal to a point where lithium-ion took 30 years to get to. Was there any world where that is licensed separately? I know you're licensing it to some of your JDA partners, but it seems like it could be lucrative. Is that something that you've considered? Is that an avenue to sort of bridge the gap between now and maybe late 20s, early 30s?
Yeah. Yeah. So actually we didn't start this conversation. So UAM OEMs started the conversation. So we offered to provide lithium metal modules with Avatar. And then they asked, okay, can you also provide Avatar, this monitoring system to their lithium-ion modules? Same mission profile, why not? And the Avatar, that model is agnostic of the chemistry. It doesn't care if it's lithium-ion or lithium metal chemistry. It just needs to see the data and then get trained. So absolutely for the AI for safety part. And then AI for manufacturing, for now we start with the two B-sample lines for lithium metal. And then to show really without Avatar, it would take lithium metal probably another 10, 15 years to get to that similar level of quality.
But with this, I mean, for example, we had no idea like electrical misalignment by 0.1 millimeter has a greater impact or less impact than electrolyte failing off by 0.1 gram. How would we know that? We wouldn't know that. But then after, so last year we had about 4,000-5,000 cells. After all the data trained the model, the model actually told us. And then it actually gave us a ranking. Okay, so this step in the process has 0.7% impact on your quality. This has 0.3%. So it actually ranked it for us. So that's really helpful. So AI for manufacturing is actually very powerful. So we start with lithium metal and then so licensing this for lithium metal. I think once they get convinced, we can also license this to their lithium-ion lines.
Yeah. Amazing. Awesome. And that's part of the 1,500 points that you're monitoring that was in that table that you showed.
Yes. Yeah. 1,500 points for lithium metal. I think for lithium-ion, the points will be even more because lithium-ion is more mature.
Yeah. Absolutely. Awesome.
Yeah.
Any other questions from the audience? Awesome. Well, Qichao Hu, thank you for joining us today.