And so far, just in the preliminary training that we have done, we actually found some very, very interesting new molecules that can be used as electrolytes. And then once we have these new electrolytes, then the new materials can extend the cycle life and then reduce the overall cost of the sale. So I think we are committed to the B-sample spending and also probably accelerating AI spending, because it’s just such an exciting area. And then if we don’t do it, I think other companies will do it. But then also in areas like building sales for UAM, then we do plan to save some costs by just converting the old A-sample lines until finding a new line.
Shawn Severson: Great. Thank you. That’s very helpful.
Operator: Thank you. I would like to now turn it back to Kyle Pilkington for any — for some additional questions.
Kyle Pilkington: Thank you. Yes, we did have a number of questions which were submitted in advance of the call and we’ll go through a selection from those pre-submitted questions now. The first question is, has SES produced any B-sample using the existing lines? Please elaborate.
Qichao Hu: So, we’re still building cells using the A-sample lines, and then we are building the B-sample line. And the B-sample line is not ready yet and we expect the B-sample lines to be ready towards end of this year. And then we are improving the A-sample lines towards meeting the specs of the B-sample. So the goal is by the end of this B-sample JDA — the two B-sample JDAs, then we will actually have the B-sample cells. But for now, we are improving the A-sample cells to meet the specs of the B-sample cells.
Kyle Pilkington: Okay. And the next question is, can you share any progress on the new B-sample lines?
Qichao Hu: Now the details, because we do have confidentiality with the OEMs, but we are making very good process in terms of designing the lines, issuing the PoS and also the vendors have already started building the lines. So we do expect both lines to be operational by end of this year.
Kyle Pilkington: Super. The next question is, do you have any update on B-sample JDAs in addition to the first recently signed?
Qichao Hu: We’re not. So we’re focused on the two B-sample JDAs, because we’ve made commitments. And in terms of getting a third or fourth additional B-sample JDAs, that’s not really a focus for us because a lot of the work are similar. So we do want to make sure that we deliver to the two B-samples that we’ve already committed and that’s our top focus.
Kyle Pilkington: Okay. And then the last question we’ll take from the pre-submitted questions is, currently how safe is SES’s lithium metal cell? There is much energy packed into a small volume and lithium metal is very reactive. Further, you talk about AI being used to improve reliability and SES invested in explosion bunkers to evaluate issues. A skeptic might conclude that these unique and extended efforts are needed to make your cells safer. How safe are your cells today?
Qichao Hu: Yes, because lithium metal does have higher energy density than lithium-ion. It does have a greater safety concern than lithium-ion, because it has higher energy density. That’s just the fundamentals of a battery. When you have a battery with a higher energy density, you will have higher safety concerns. And historically, in the past, lithium metal did have some safety issues. This is why we are spending a lot of effort on Avatar AI. So all the cells that we build, all the good cells, bad cells, actually sometimes we intentionally make the cells fail, so that we have both positive signals and negative signals, and then we use them to train Avatar AI. So the goal of this Avatar AI is to try to ensure as close to 100% safety in the field as possible, on top of making the electrolyte safer through this AI for science effort and also making the cell design as safe as possible.
But we know from a pure hardware perspective, no high energy density cells can be 100% safe. That’s why we need this software, this Avatar AI on top of this hardware to make sure the actual operation in the field can be as close to 100% safe as possible. I think not just for lithium metal, but any high energy density batteries in any vehicle operations in the field must have Avatar AI.
Kyle Pilkington: Great. Thanks. Thanks, Qichao. I’ll pass it back to the operator to see if there’s any additional questions on the line.
Operator: Thank you. [Operator Instructions] And we now have a follow up question from Jed Dorsheimer with William Blair.
Mark Shooter: Hey, guys. It’s Mark again. Thanks for the additional question. I just wanted to dial in a bit more on the AI. I was wondering what you’re using for the training data? Is this — and you’re saying that since being installed in the lines and especially the lithium metal line with the partner Hyundai. That has me thinking, is this — are you looking at metrology of the lithium metal surface to try to predict yields or future current density hotspot issues? So, does it help with yields or is this more of a — we’re building cells, we’re doing cyclic voltammetry and stressing the cells at certain cut off voltages and monitoring in that way? So is it cell testing or is this kind of a yield thing as well?
Qichao Hu: So it’s actually three types of data. First is the cell design. So like the cathode loading, the anode thickness and then the electrolyte, basically the design information of the cell. And the second is manufacturing data. So that includes, like you said, lithium surface morphology, roughness, electrolyte stacking, overhand misalignment, electrolyte filling and pressure during heat sealing. So all the parameters, all the data during the manufacturing process. And the third is testing. So that includes like the cyclic voltammetry at high temperature, low temperature, high pressure, low pressure during cycling, the first cycle, the second cycle, all the way till 500, 600 cycles, until the cells die. And then all three types of data are used to train this model.
