Hi, everyone. Thanks for joining the session. I am Kingsley Crane, a Technology Analyst here at Canaccord. Really pleased to have the GSI Technology team with us here today. Didier, thanks for joining.. Thanks for having me. Let's kick it off. You just reported your June quarter. You have described this as a pivotal point. What does GSI look like today versus 18 months ago, and just at a high level? Yeah. We certainly progressed the company quite a long ways. A 18 months ago, we were just coming out with Gemini-II, which is really our first commercialized part for the AI market. Fast-forward to now, since then, the Gemini-II is in production now as far as the hardware is concerned. We have gotten some third-party validation from folks like Cornell. Cornell University actually had a board of ours and did a RAG comparison with an NVIDIA GPU, and it found at comparable performance, we were 98% less power, so it was certainly a huge movement in the market space. Since then, we have also engaged in a few POCs. We have one drone surveillance POC that is being funded by the DoD. We also have recently won a Phase I POC out of a municipality in Taiwan for Smart City, which is nice because it leveraged a lot of what we're doing for the POC for the drone, so it's not a complete new lift. Since then, we've also gone quite a ways with developing an AI SDK, or AI-assisted SDK, I should say. This is important because it's going to help us enable the ecosystem. If you look at our model now, we are writing our own applications for all these POCs, which is fine to really showcase the technology, but to really scale, we're going to have to have tools that the customers can use. We are quite a ways with that SDK. We'll have our alpha version out in this fall. Let me see what else have we done since then. We've won a couple SBIRs, one with the U.S. Army, for ruggedized edge node. This could be a nice product for us and for the Army. Essentially, it's going to be a server that can do object detection, or it can do SAR imagery, and it can be done, again, at the edge. So it's a ruggedized server they can put at the back of a Humvee or something. Let me see, 18 months. We've also started our next-generation device, the Plato. Plato is going to be. It leverages some of the technology, obviously, from Gemini-II, but it's going to adjust a different market. The way that Gemini-II is that it's. We have a small bandwidth coming from memory into our chip because really, the intent of the chip is to download a model or a database one time, and then once it's in our chip, we run it really, really fast. So our internal bandwidth is extreme, so it's perfect for search applications or HPC kind of applications. Plato, on the other hand, is going to be developed more for LLMs at the edge. We're going to open up the pipe so that we can get data into the part faster, and then we'll scale down the internal bandwidth to match that. That'll actually be used for vision language models, large language models at the edge. It's going to be extreme because it's going to have almost data center performance at a 2-10-watt power budget. So it's going to be a powerhouse for edge applications. Really helpful overview. Investors are trying to understand the AI supply chain and target the bottleneck. Can you just help us understand why SRAM is so important for AI, how durable that business is, and what is a cyclical, AI super cycle this time? Sure. I'm going to answer that in a couple of different ways. First of all, it's important to our company because right now it's the cash cow. We're just starting the AI story, and so we need something to offset the bills and we've been doing SRAMs now for 30 years, shipped over 140 million devices. So we're certainly a leader in that market. It's really helped offset the bills or the cost there. As far as the market itself, the SRAM isn't directly in the AI, it's not in a data center, but it really helps the infrastructure. What I mean by that is one of our largest customers we've talked about is Cadence. They make emulation systems. These are systems that if you're an IC manufacturer or designer, I should say, you do a design and in the old days, you just do a simulation. Then, is the part working? Do I have any major bugs? It wasn't until you went to first silicon, got the actual chip, you looked at it and said: Wow, it's dead on arrival. Part doesn't work. That's a real problem nowadays because you have to create a mask set in order to get that chip. The mask sets cost upwards of $30 million a mask set, so you don't want to be wasting a lot of $30 million mask sets. Now, guys like Cadence make these emulation systems where they actually emulate the design in software, and these are large, expensive systems, and our highest-end parts go into those systems. We're really helping the front-end design. Another one of our larger customers is KYEC. They actually are part of the manufacturing process. In this particular case, they do burn-in. Burn-in is basically a way to electrically really challenge a part to make sure there's what's called no infant mortality rate. In other words, you don't want a part to be put in a system, go out to the field, and then fail because it's a weak chip. So they do burn-in to try and get rid of those. KYEC is doing the back end for all the latest GPUs that are out there, and they use one of our high-end parts. So we're indirectly supporting the AI with our SRAM division. Can you help us understand why compute-in-memory is so critical from a performance and cost perspective? Sure. There is this thing called von Neumann model. I do not want to get too much into detail on it, but essentially, if you look at the way a GPU and a CPU work, they have their processing elements. When they are being asked to do something, do a calculation or some kind of process, they have to go outside the chip to fetch data from memory, bring it back, use it, and once they have used it, they have to write it back to memory. There is this constant data transfer, data flow back and forth. It takes a tremendous amount of power, which I am sure you have all heard, data centers, what is their biggest problem? Power. With the APU, we have done something much different, and this is all patent protected because we know we have something unique is with the Gemini-II family, as I mentioned, we have a small pipeline going into the chip. We bring in data one time, and once it is there, we run it very fast. The way we do that is we have, first of all, a large memory inside, but the process or the calculation is actually done in the memory bit line, in the memory itself. Our bit processors are actually coupled with our memory. We are not going out fetching data, we are not bringing it back. We do not have this constant transfer, and so the performance is high, but the power is extremely low. As I mentioned, this Cornell paper, 98% less power, and that is because of that CIM architecture. For Gemini-II, can you help us get a sense of that path towards commercialization, just moving from proof of concept into product, and then what do you think success looks like in a year from now? Sure. As I mentioned, Gemini-II, the hardware is production ready. The software is catching up right now. I mentioned we're going to have that AI-assisted SDK in the fall. In the meantime, we have two engagements on POC, as I mentioned. One is the drone surveillance, the other one is a Smart City out of Taiwan. We're writing all of the applications for that ourselves right now, and that's kind of our model. We're a smaller company, and so we have limited resources. We're going to write an application for a certain market, show that we have a good solution and an advantageous solution, and then at that point, that coupled with the AI-assisted SDK will allow us to then use system integrators for each of these markets to address the broader market. POC for the initial and then the system integrators with our tools for the launch. We're looking at the POCs, the two we have identified now will last through this year. We have a couple more that we haven't announced yet, but we're still working on. That'll take us into next year. What we're looking at is completing the POCs between the end of this year and next year and having early production by the end of 2027. To what extent do you feel like those engagements that you're having right now are reusable? Great question. They are very reusable. It took us a while to do the first POC with the surveillance drone. They had a requirement that was called time to first token, which is essentially, you have a video feed coming in and there's events that happen and you need to identify, is this an issue? Is this a bad event? You need to do that very quickly. In this case for the drone, it had to be less than three seconds. At that point, once it identifies, okay, I have a problem, what is the problem? You have to identify it's a truck that went through a gate unannounced. At that point, you need a response. What's the recommendation? What do you do? This particular Sentinel program, it's usually military bases or government buildings. The response might be, take out the truck. We don't know what the response is, but we have to at least do a recommendation. The work we did there, a lot of it should be usable for the Smart City. Smart City now is somebody jogging in a park and clearly grabs his chest and goes down. The system has to identify this is not somebody who's just tired and wants to take a nap. This is somebody who's having a medical emergency. That's the identification. The response is alert EMT. Car has an accident. Well, that happens all the time, but car starts to smoke, catches on fire. It gives you an alert to the fire department. You have a car on fire, and here is where it's located. That going also to other POCs will have similar. Other POCs we're looking at are industrial inspection, time to first token, very important. A lot of what we use from the last POC we can use for future ones. When you think about a software developer kit or SDK, building mindshare is really important. If you move towards an AI-assisted SDK, this could further increase or improve time to value for customers. Can you just double-click on why that's so important for GSI? Yeah. It builds the ecosystem, right? As I mentioned, we're a small resource company. We don't have thousands of software engineers that are writing applications for us, and that's just not going to happen for us. We need to enable the market for them to write their own, whether it's the customer directly or whether it's some of these system enablers I mentioned. What this SDK does is allows them to write at a higher level language and have it be translated into a language that our partner understands. To give you an idea, the drone POC we did, the application we wrote there, it was before the AI-assisted SDK. It took us about a man year to write that. With this AI SDK, which excuse me, my internal team is starting to use already now, it's going to take that down to weeks. It is going to really accelerate the time to be able to get out these applications and also allow customers and integrators to use them without understanding really the machine code of our part. Maybe it would be helpful, we have talked about Gemini-II, but just a reminder of why Gemini-II is so important, why it is winning, and then the next step that Plato takes on top of that. Sure. More importantly. Sure. Let me give you a, excuse me, real-life example of why it is important. I keep talking about this drone Sentinel POC. Originally, the drone, excuse me, the drone manufacturer was going to use NVIDIA. Everybody who knows NVIDIA, use NVIDIA, right? There were two critical components to this POC. Number one is time to first token, I mentioned of three seconds. The other one was they needed a power budget of less than 50 watts. The drone needs to go up, and it needs to stay in flight for some amount of time. They looked at NVIDIA. NVIDIA gave them the three seconds time to first token, but their power was 160 watts. It was over three times beyond their budget. Then they looked at Qualcomm Snapdragon. Snapdragon gave them the sub 50-watt power, but their time to first token was 12 seconds. It's 4x the allowable amount. Assuming the reaction 3x and it takes you four times longer, a lot of times, the danger has escaped or what have you. They had to get the critical. That's when they looked at GSI, and we gave them sub 50. In fact, we're at 30 watts of power. Right off the bat, we were three seconds time to first token. Just past June, we actually did a lab demo, and I say we, us and G2 Tech who's our drone partner, did a lab demo for the DoD, and the time to first token actually came in at two and half seconds. We've exceeded the 50 watts and the three seconds. Maybe talk to us more about the balance sheet, over $70 million in cash, no debt. How much do you view that as an asset for you? Then just thinking about that enabling the roadmap. Sure. Yeah. No, it's critical to have that. As you mentioned, $77 million, no debt. We have the SRAM division that's offsetting a lot of the bills, obviously. But we're still burning some cash right now until the APU takes off. We're burning roughly $4 million a quarter, so we'll say $16 million a year. That's flat except for we will have a slightly higher expense in the spring quarter when we have the tape out for the Plato, and that'll be an extra couple million dollars. We're not using the latest advanced technology there, so it's a little cheaper mass set. Besides that, the $77 million clearly is enough to satisfy all of our near-term goals that we need and milestones that we need to cover. Yeah, we feel that we're in great shape there. Just to complete that funding picture, how about SBIRs or other sources of non-dilutive funding from the government? Yeah. SBIRs, they help on the funding, absolutely. We use SBIRs. If you are not familiar with them, it stands for Small Business Innovation Research, I think is what it stands for. Anyway, it is basically a way for the U.S. government, through one of the DoD elements, to help fund some technology for smaller companies. As we use that money as an offset to R&D costs. For us, it is also a way to engage with these entities. We have won two SBIRs with the Air Force Research Laboratory. We have won two with the Space Development Agency, and we won one with the U.S. Army. We have two that are still active. One is with the Space Development Agency, and that was to fund radiation testing on our Gemini-II part. What the SBIR is doing is taking a commercial off-the-shelf Gemini-II part, and we are doing radiation testing, looking for any kind of single-event latch-ups, which is very bad. We are also doing a total ionizing dose testing, which is basically how many ions can the part absorb before it stops working. We actually did the radiation testing in June. We are still waiting for the report, and because the testing is done by a third party, but we did have a GSI employee present at the testing, and there was zero single-event latch-ups, which is critical, and which was fantastic. Again, this is on a non-optimized commercial part. At the end of this month or beginning of September, we will do the TID testing, and we are confident we will do well there because our SRAMs have always done well, and we have similar technology. Once we have those two elements in place, it is going to help enable the DoD entities to be able to have a part that they can use in space. Any of you familiar with some of the other technologies like GPUs, they inherently are not good for space applications. Right now, if you look at some of the applications, there might be a LEO satellite right now that has sensors that are basically taking the data and sending it to Earth for image creation or what have you. They want to be able to do this in space. In other words, the satellites take those sensors, and then we can create the images on the satellite, and then now we can send that to Earth saying: Hey, these guys are deploying equipment or trucks or something, or deforestation, or what have you, whatever they are looking for. Those SBIRs are critical, not only for offsetting R&D costs but also for creating new markets. One last thing I want to touch upon was on the U.S. Army. I mentioned it earlier. It is for a ruggedized edge server, but this is something that could be productized by us for the DoD. This is not only a way to offset R&D but also a potential revenue stream for us in the future. Right. I think Elon would love to put a bunch of GPUs in space. Yeah. We'll see how quickly he can do that. There you go. In terms of needing power-efficient AI, of course, we need it in the data centers, but we need it at the edge. It's so critical there, and so thinking about use cases in industrial automation or Smart City is really compelling. You talked a lot about defense. Is defense the beachhead market where you break in? Then maybe just talk more about opportunities outside of that. Yeah, I think it is, only because those guys have shown the early interest and they've made commitments. These SBIR dollars, they're millions of dollars, so they're actually making a financial commitment that they're saying our technology is something that is going to be useful to them. I do see that the early wins and interest we have come from that market. Yeah, I do see that would be where some of our early revenue would come from. What KPIs should investors be watching in the business over the next couple of quarters to really judge this inflection? Sure. Close out the two POCs we have now. As I mentioned, the drone one. We've actually, candidly on our part, have finished all of our deliverables. The drone manufacturer now has to finish theirs for the final field demo, but then we have to see how the final demo goes. We have to close out the pPhase I POC for the smart city, at which point it's going to go to Phase II. Phase I is basically. Let me take a step back. This municipality already has cameras installed all over the place, and what they do now is they just record. That's all. That's all they do. If there's an event that somebody is interested in, they have to go back to the footage and go back through old footage to see what happened. What this municipality wants is they want to have intelligence up front, so they want to understand, is there an event happening right now? What's the event, and how do we respond? Phase I is to take video from this system for 20 cameras and go through. In fact, we have a small demo on our website you can look at. It's one of the demos that we used to win the POC, but it gives you a feel for what it does. Phase II will bring it up to 80 cameras and will also now include audio. The initial Phase I is only video, now it's audio. Because some of the issues, they're going to be having some of these at schools, and they need to understand if there's any abuse or anything happening. Abuse isn't always physical. Sometimes it's verbal, and so they need to be able to see that. Then Phase III. By the way, Phase I will have our deliverables in November. Phase II will start about then. Assuming we win Phase II and we win Phase III, that is the production deployment that would happen sometime at the end of 2027, and that could be anywhere between 2,000 cameras to 6,000 cameras. To give you a feel for what that means to us, there's one Gemini-II chip for every four cameras. It will be a significant hardware sell for us. On top of that, we have an annual recurring license to the application itself to keep it running. It will be recurring revenue there. We have the POCs we need to close out. Another KPI is obviously the AI-assisted SDK. We'll have the alpha release in fall for certain customers, and then they're going to try and break it and get any bugs out of it. Our intent is to release it sometime next year to the mass public. Other KPI, obviously the tape-out or the end of the design of Plato. Plato is going to be real important, as I mentioned, because it's going to be an LLM powerhouse at a very, very low wattage. Certainly in the springtime, we need to get that out. Just even thinking about centralized AI, we have so much more room to use more of that. Decentralized, we have even probably a longer runway for that to permeate through all these different form factors. How do you, as a company of your size, compete against some of the larger players? Or how do we make sure that the best technology does win? Sure. Yeah. Let me take. I'm going to take one step back before I answer that question. You have the Qualcomms, the AMDs, and the NVIDIAs, huge companies, and then most of the other AI companies are very small, and they have an idea. We're more of an AI startup as far as the technology goes, but we're not a semiconductor startup. That's one thing that people need to understand. As I mentioned earlier, we sold 140 million SRAMs over our 30-year career, and we did it using TSMC as our fab, using ASE as our assembly house. Fast-forward, our APU, our Gemini, and our Plato, we'll use TSMC and ASE as our two assembly houses. We have a long 30-year relationship with them. We already have a full operational team. As far as ramping, that's not going to be a problem. As far as how do we win, we bring both performance and low power, and that's critical. As I mentioned, on a compute basis, NVIDIA is strong. They have good performance. Their power is high. Guys like Qualcomm, their power is good, their performance is low. There's going to be a lot of applications where you're going to need the performance and the low power. On a performance per watt basis, that's how we beat those guys. We're close on time. Just want to make sure the audience has a chance to ask if they'd like. Now you've talked about the technology, and it's been validated, and we're starting to see an inflection there, but what else do you need to prove either to customers or to the market over the next 24 months? I'm not sure proof is the right word as much as execution. We have to get those software tools out. The hardware, as I mentioned, Gemini-II production ready right now, and Plato will be out next year. It's really getting the software tools in the hands of the customers and the market so that we can really create that ecosystem. It's really more of an execution than it is. I think we've proven that the technology is real. The Cornell paper helps and other benchmarking we've done, winning the bake-off for the Sentinel program. I think we've done the proof part. Now we just have to do the execution on the software side. Didier, thanks so much for joining us. Thank you, Kingsley. [crosstalk] seeing that execution.