GSI Technology, Inc. Aktie

Welcome to GSI Technologies Fourth Quarter and Fiscal Year 2026 Results Conference Call. [Operator Instructions]

Before we begin today's call, the company has requested that I read the following safe harbor statement. The matters discussed in this conference call may include forward-looking statements regarding future events and future performance of GSI Technology that involve risks and uncertainties that could cause actual results to differ materially from those anticipated. These risks and uncertainties are described in the company's Form 10-K filed with the Securities and Exchange Commission.

Additionally, I have also been asked to advise you that this conference call is being recorded today, May 7, 2026, at the request of GSI Technology. Lee-Lean Shu, the company's Chairman, President and Chief Executive Officer, will be hosting the call today. With him are Douglas Schirle, Chief Financial Officer; and Didier Lasserre, Vice President of Sales.

I would now like to turn the conference over to Mr. Shu. Please go ahead, sir.

Good afternoon, and thank you for joining us to review our fourth quarter and fiscal year 2026 financial results. Fiscal 2026 was a year of meaningful progress for GSI, marked by strong performance in our SRAM business, continued advancement of Gemini-II toward commercialization and the initiation of the Plato design. I am pleased with the progress we have made on several fronts, significant work remains. Our team is executing our key milestones and advancing business development for the APU and I have had several encouraging conversations on numerous fronts in recent months. We ended fiscal [ 2027 ] with continuous momentum, promoting the APU and building on our customer traction. With that, I will now hand the call over to Didier Lasserre.

Thank you, Lean. Let me start by stepping back and framing where we are today because I think the context is important. Our SRAM business performed well in fiscal 2026 and remains the revenue foundation of the company, providing cash for APU development.

For the full year, the SRAM business grew 22% year-over-year and gross margins rose to 55% from 49%. The SRAM business has benefited from increased demand from our customers that support high-performance AI chip development and manufacturing. We recently announced that we concluded our strategic review and determined that continuing to execute our stand-alone strategy is the best path forward for delivering long-term shareholder value. The stronger SRAM business and a strengthened balance sheet, along with nondilutive R&D funding are providing the resources to support our go-forward plan.

With this financial foundation in place, we are now seeing real progress with Gemini-II and Plato. Over the past several months, we've reached a point where we are seeing both technical validation and early program level engagement of Gemini-II, including the Sentinel drone surveillance POC, the U.S. Army SBIR award and a new Phase 1 smart city project, which I'll discuss in a minute.

On the technical side, in a bake-off for the Sentinel POC, Gemini-II's performance contributed to winning the contract award by achieving a time to first token of roughly 3 seconds at 30 watts of system power on Gemini-312B multimodal workloads at the edge. In this use case, time to first token is a critical metric for a drone surveillance system because it reflects how quickly a system can respond in real-world applications where response time directly affects critical decision-making.

We are working closely with the GSI tech team on the Sentinel program. We have completed the software deliverables and continue to target a June demonstration of the Gemini-II powered drone. This demonstration is planned for the Department of Defense and an International Defense Agency. In mid-April, we were notified that we had been awarded Phase 1 of a Smart City project. The project leverages our work done for the drone-based surveillance POC and marks an important step forward towards commercial deployment. In this application, Gemini-II will process inputs from distributed camera systems to provide near real-time detection of events such as fires and other public safety risks.

This project demonstrates how our platform can scale across real-world infrastructure. We expect to share additional details on the Smart City program around the time of a planned media event in late May hosted by the municipality.

Currently, we are working on several projects in tandem. What matters most for GSI at this time is not just the number of early-stage trials and demonstrations we have, but also how these early-stage engagements are helping us identify where our APU architecture provides a clear advantage, particularly in delivering low latency performance with a constrained power envelope. We are also leveraging our deployment work in 2 ways. First, we are applying what we have developed for a drone security application to a smart city application. While the end markets are different, the underlying development carries over, giving us meaningful head start in each new case rather than starting from scratch.

Secondly, as we complete the Sentinel POC and Phase 1 of the Smart City program, we can build on those results to pursue additional opportunities with new customers in those markets. We view this as a repeatable model where each engagement helps accelerate the next. What's exciting for us is that we see the end markets for low-latency, low-power AI edge is expanding as AI workloads continue to move closer to where the data is generated. These applications favor the APU architecture that can deliver higher compute per watt.

Gemini-II is ideal for these power latency constrained edge deployments where real-time response and energy efficiency are critical. Where we are winning is where Gemini-II is tested against conventional architectures requiring significantly higher power -- excuse me, higher system power for similar or slower responsiveness. We believe Gemini-II best addresses this gap and positions us well to win as more AI loads shift towards distributed power-constrained environments.

Consistent with this, we are encouraged by our progress within defense agency programs as evidenced by our recent U.S. Army SBIR, progressing from Phase 1 into Phase 2. This project is about enabling real-time infield AI deployment on small, low-power systems typically operating in challenging conditions. As part of this program, we will build and test a ruggedized node containing the Gemini-II for real-world mission-critical environments.

This SBIR positions us within a broader shift in defense spending with approximately $13 billion proposed in fiscal 2026 budgeted for AI and autonomous systems and creates a potential pathway to follow-on programs and future opportunities to supply Gemini-II-based systems.

So how do we move from where we are today to design wins and ultimately revenue? From a commercial standpoint, we are still in the early stages. Our focus is on advancing our current engagements and working closely with partners to integrate Gemini-II into their systems with the goal of moving into design level discussions. Given the complexity of these deployments, we are focusing our resources on a small number of high-value opportunities where we believe we have a clear advantage.

Although the number of engagements remains limited, we are seeing a meaningful increase in the depth of these engagements and our ability to leverage our prior Gemini-II deployment work for new related applications.

Looking ahead, our priorities are to advance current POCs and awarded programs and to leverage what we've learned from each of these engagements to drive additional design opportunities. At the edge, performance matters most when it can be delivered within real-world power and latency constraints. That is where we believe Gemini's II advantages lies.

With that, I'd like to hand the call over to Doug. Go ahead, Doug.

In the earnings release issued today after the close of the market, you'll find a detailed summary of our financial results for the fourth quarter and full fiscal year 2026. Rather than walking through the numbers again, I'll focus my comments on the key drivers behind the results and provide more context and explanation to help you better understand the business.

Let me start with the results for the fiscal year 2026 ended March 31, 2026. As Didier mentioned, fiscal 2026 revenue increased 22.4% to $25.1 million, reflecting continued strength in our SRAM business, particularly with customers supporting chip design and simulation for AI applications. We experienced solid growth in this customer segment throughout fiscal year 2026. We do see variability in customer orders and sales can fluctuate from quarter-to-quarter. However, barring any significant change in underlying AI chip demand that would affect SRAM orders from these customers, we expect this business to remain relatively stable in fiscal year 2027.

The higher level of revenue and product mix helped to lift fiscal year 2026 gross margin to 54.5%, a notable gain from the prior year gross margin of 49.4% Operating expenses in fiscal 2026 rose to $31.2 million compared to $21 million in fiscal 2025. Operating expenses increased year-over-year, primarily driven by higher R&D spending on the Plato chip design. It's also important to note that the prior year included a $5.8 million gain from the sale of assets, which makes year-to-year -- year-over-year comparison appear more pronounced.

We also continue to offset a portion of our R&D expenses through nondilutive funding, SBIR contract funds and POC-related funding. The majority of our R&D is dedicated to the APU. The R&D offset in fiscal 2026 and fiscal 2025 was $1 million and $1.2 million, respectively. Higher operating expenses increased the total operating loss for fiscal 2026 to $17.5 million compared to an operating loss of $10.8 million in the prior year. The fiscal 2026 net loss included interest and other income of $4.1 million, primarily from interest payments on the increased cash balance from the capital raise completed in October 2025 and $3.4 million of other income, consisting of a $6.2 million noncash gain from the change in the fair value of prefunded warrants, partially offset by $2.8 million in issuance costs associated with the registered direct offering in October 2025.

Looking now to the fourth quarter. Revenue was $6.3 million with a gross margin of 52.4%. As we've seen in prior periods, quarterly gross margin can fluctuate with the product mix and revenue levels. The fourth quarter gross margin reflects slightly lower SigmaQuad sales sequentially compared with the prior year quarter. From a customer perspective, we did see some variability across accounts during the quarter, including lower shipments to certain customers and higher shipments to others. At the same time, defense-related sales increased to approximately 46% of total shipments, reflecting continued demand in that segment.

Again, you'll find a full breakdown of sales in today's earnings release. Operating expenses increased from the prior year, primarily due to continued investment in our Gemini-II and Plato development programs. These investments align with our strategy to advance our APU road map while maintaining disciplined cost management. Last quarter, we expanded quarterly earnings disclosures to help investors better understand the company's cash consumption and cash generation. This information will complement the condensed consolidated statement of cash flows included in our Forms 10-K and 10-Q.

Cash flows for the quarter ended March 31, 2026, were cash and cash equivalents as of December 31 were $70.7 million. Net cash used in operating activities in the quarter was $5.5 million. Net cash used in investing activities was approximately $100,000 and net cash provided by financing activities was $2.1 million. Cash and cash equivalents as of March 31, 2026, were $67.2 million.

From a cash flow standpoint, spending in the quarter continued to reflect our investment in Gemini-II and Plato development. We expect cash usage to remain elevated as we progress through this development phase. As a general reference point, we expect the cash usage to be approximately $4 million per quarter or about $16 million annually, although this may vary depending on development timing and program activity. We ended the quarter with $67.2 million in cash and no debt. This is a notable improvement from the prior year-end cash balance of $13.4 million. and is associated with the $46.9 million net of fees registered direct offering proceeds that closed in October 2025.

The absence of debt and the improved cash balance provide us with the flexibility to continue investing in APU while maintaining a disciplined approach to capital allocation. We believe our current cash position provides sufficient runway to support the initial commercialization of Gemini-II and the completion of the Plato tape-out, both expected in late fiscal 2027.

Before I hand the call over to the operator for Q&A, I'd like to provide the first quarter fiscal 2027 outlook. For the upcoming quarter, we expect net revenues in the range of $5.9 million to $6.7 million with gross margin of approximately 54% to 56%. Overall, our strong cash position and continued support from nondilutive funding give us a runway to advance Gemini-II into early commercialization and the completion of the Plato chip design.

Operator, at this point, we will open the call to Q&A.

[Operator Instructions] The first question is from Tony Brainord, retail investor.

Yes. Can you share some color on the size, like if you do get the design wins, the size of the market we're looking at?

On which market?

On the Gemini-II?

Okay. That's a pretty broad question. So the markets we're going after initially, some of them are government military based, specifically these drone programs. And as we talked about, we're limited detail now. We'll give you more detail on the smart city at the end of May. But both of those markets are multibillion-dollar markets.

[Operator Instructions] The next question comes from Robert Christian, private investor.

Yes. I'd like to know why the Plato project has moved up from the first half of 2027 to late fiscal 2027?

So it actually hasn't been pushed out. So it might have been a mixture of calendars and fiscal quarters. So when we had first talked about it, we were targeting the beginning of calendar 2027 to have the part taped out, and we're still on schedule for that. So the tape-out means that the design will be done in the first quarter, and that would give us silicon because we have to make the assets that are used from the wafer fabs at TSMC. So we'll see our first wafers in hand in the summertime of calendar 2027. And I believe that has been always our schedule.

Yes. I think we mentioned the fiscal year 2027, that's the beginning of the 2027 calendar year.

Right. That's a good point. So the end of fiscal 2027 is March of calendar '27.

Okay. That would be great. And the second question I have is Gemini-II taped out over 2.5 years ago. Is it going to take that long to see expected sales, say, of Plato?

So that's a great question. So you have 2 components to sales. You have the hardware component, which is the chip and any kind of a board and you have the other software side. The software side actually lagged the hardware on Gemini-II. With Plato, we're trying to align the 2 of those more closely. The good news is some of the software work that's being done for Gemini-II can be used for Plato, while with Gemini-I, it was a completely new effort. And so in that respect, we can leverage some of the work from Gemini-II for Plato, and then we're also lining up the resources to be able to bring in the software with Plato.

Well, the chip is genius, and I wish you guys Godspeed.

At this time, we show no further questions, and this concludes our question-and-answer session. I would like to turn the conference back over to Mr. Lee-Lean Shu for closing statements.

Thank you again for joining today's call. As a reminder, Didier will be at the LD Micro Conference on May 19. Contact LD Micro if you would like to attend this presentation or take a one-on-one meeting. We are encouraged by the progress we are making the Gemini-II, and we remain focused on successful executing against the opportunities in front of us. We look forward to speaking with you again on our fiscal 2027 first quarter earnings call. Thank you.

This concludes today's conference. Thank you for attending. You may now disconnect.

Good day, and welcome to the iAccess Alpha Virtual Best Ideas Spring Investment Conference 2026. Our next presenting company is GSI Technologies, Incorporate. [Operator Instructions]

I'd now like to turn the floor over to today's host, Mr. Didier Lasserre, Vice President of Sales and Investor Relations for GSI Technologies, Incorporated. Sir, please go ahead.

Thank you, and thank you for joining us. As the moderator mentioned, my name is Didier Lasserre. I'm Vice President of Sales and Investor Relations here at GSI. GSI has been a semiconductor company for over 30 years, known for our high-performance SRAM products that are used in networking, defense and other demanding applications. That business remains an important financial foundation for GSI being it generates revenue and cash that support the development of our next-generation technology, the Associative Processing Unit. We are excited about our APU due to our proprietary compute and memory architecture.

Our current product, the Gemini-II is designed for power and low latency constrained edge environments such as drones, satellites and any other autonomous systems where minimizing data movements enables significantly higher performance per watt than traditional architectures. Gemini-II is already being evaluated in defense and other edge applications as we work towards initial design wins.

Today, I'll briefly walk through one of the -- I'm sorry, the technology behind the compute and memory architecture, the edge AI market opportunities we're targeting and how we plan to bring the APU platform into commercial deployments in the upcoming years. The key takeaway I want to leave you with is that edge AI favors architectures that deliver the most compute per watt, and that's exactly what the APU compute and memory is designed for.

I will be making some forward-looking statements, so we have included the safe harbor statement here. A quick overview. As we mentioned, we are a leader in the high-density, high-performance memory market. We've been partnered with TSMC for our wafer fabs for over 30 years, and this will be the same partnership that we'll be using for the APU as well. We've developed and invented the APU chip, which is, as I mentioned, a compute in memory CIM technology. We are targeting the edge. We are not looking at the data centers at the moment. And so the APU is really manufactured and designed for edge applications. To date, we spent over $175 million on the APU R&D, which has been funded using our SRAM product line. This past October, we raised net $47 million through an equity raise.

If you look at our trailing 12-month revenues, we're just under $25 million. In fact, this month, at the end of this month, we'll finish our fiscal 2026. If you compare what we're running at for this year versus last quarter -- or I'm sorry, last year, fiscal 2025, it will be about a 25% increase in revenue. We outsource the labor-intensive portions of our business, the fab, assembly, day-to-day sales. And so we're able to keep our headcount down to a very efficient 122 employees. So the majority of the employees are either hardware or software engineers.

We have a very unique architecture, which I'll be talking about extensively today, and we want to protect that. So we've been aggressive with filing patents. So we now have 87 patents, specifically for the APU. The balance sheet is strong. We have just over $70 million in cash and cash equivalents, market cap over $300 million. I want to say this morning, we're about $320 million in market cap and a high insider ownership of 20%.

So just looking at a high level, what's the -- what are really the challenges in the AI market space? What's the bottleneck? And really, the bottleneck is the fact that data needs to be moved and transferred constantly within the system. And when you're moving data from memory to where the compute elements are, it takes time. It takes latency and it also takes a lot of power.

For edge environments, which is what we're focused on, really, the compute is constrained because you have a very limited power budget. And this is where the APU fits in perfect because we do the compute or the processing where the data resides, and I'll explain how that works.

If you look on the right, that's essentially what a GPU or CPU looks like. You can see that you have the DRAM, which is the memory, which is where all the data is stored. And so if for some reason, a GPU needs to do a compute, it needs to grab the data from DRAM, has to transition and be transferred through L2 cache to L1 cache before it gets to the compute elements. And once that data is used in the compute elements, it has to be written back all the way through the same cycle back through L1 through L2 back to memory. So this constant transfer of data, besides taking time takes a tremendous amount of power.

If you look on the left, that's our architecture, very simple. We actually do the compute or the processing in the memory array itself. And so the compute bits or the compute elements, processor bits are physically in the memory array. And so we are -- the memory or the processing is actually where the data resides. So we're not having to go fetch it and then we don't have to write it back. And once we're through using it, it remains in place. So this really significantly increases the performance and lowers the power. Also in our architecture, we have over 1 million bit processors that can work simultaneously. So we have massive parallel processing with our technology.

And then lastly, our resolution or our bit width is not predetermined like GPUs. It's not predetermined to be an 8-bit or 16-bit. It's -- we're a bit engine and you can configure it any way you want, and it can change from cycle to cycle. So if this cycle, you want to do some processing at 8-bit, that's fine. Next cycle, let's say, you have a model that's most efficient at 3-bit, you can go ahead 3-bit, no problem.

So what's interesting is a lot of the AI workloads are moving to the edge for several reasons. Mainly, what you're finding is that it's real-time responses are required. And so you need to start doing the processing where the data is collected. And so that transition is happening. Now the other reasons for that transition is the fact that cloud compute besides becoming expensive, it also takes time and it's also not private. Some of the military and defense applications are working on the data that is not allowed to leave the device for security reasons. And so therefore, there's really a huge demand now at the edge for this real-time inference.

So why does AI at the edge require a new architecture? If you look at the traditional methods, again, as we talked about, the data is separated from the compute. And so you have to go get the data. Again, this constant transferring of data requires time and power. So if you look at the GSI APU, with the data in the memory array where the processing is happening, we're able to, again, lower the latency and lower the power at the same time. So GPUs are good for data centers. I mean they're great for training, and they're very good for large data centers. But when you get to the edge, power, the performance per watt is critical. And that's where the APU really shines.

So this is a true case POC. In fact, this is the POC that we announced last quarter. And this is for a drone perimeter security program. And so at the time, the drone manufacturer needed specific parameters. They needed the time-to-first-token to be no more than 3 seconds, and they needed at the time a system power to be no more than 50 watts, preferably 30 watts. And so this drone manufacturer went to NVIDIA first and we looked at the Jetson and the Jetson gave them the time-to-first-token that they're required, but it was significantly over 100 watts for that performance, and that did not work for this drone power budget.

So then they looked at a Snapdragon from Qualcomm, and they were able to achieve the power requirements, but it took 12 seconds to get the first token out. That's 4x slower than they could afford. And at that point, they looked at the Gemini-II from GSI, GSI was able to give them the performance of 3 seconds on the time-to-first-token, along with the 30 watts on the power budget. And so this was a critical win for us, and we were chosen as the hardware solution for this program.

If you look at the market sizes, a lot of folks are really concentrated on the data center. Everything is about the data center. But if you look at the edge AI market, it's going to be exploding in the future. Right now, it's estimated to be about $20 billion, growing to $120 billion by 2030. If you look at the markets that we want to address, we're about $7 billion right now, and it will be more than doubling in the next 5 years. And these markets include drones, SAR satellites, in autonomous systems, smart cities, automated warehouses, anything in that category.

So we've talked about Plato in the past. So Plato is our next-generation part. It's going to be designed specifically for LLMs at the edge. And I want to emphasize at the edge. So certainly, there are GPUs today that are working on LLMs for the data center. Those require over a kilowatt of power consumption. If you look at the Plato, Plato is going to be designed to be around 10 watts and in a lot of cases, less than 10 watts to deliver an LLM. So it's really designed again for the edge. We started the design this past quarter, and we're anticipating having the design done about a year from now. So we'll be taping out in the first half of 2027.

If you look at our strategy on how we want to monetize the APU family, we're going to be starting with the Gemini-II. And we're starting with applications like drones and smart cities and anything basically physical AI at the edge. We're doing this through POCs and through some of our government contacts, which I'll talk about how we're doing that. But really, it's these advantages with the CIM architecture for things like time-to-first-token and other surrounding awareness that are really going to be leveraged for all these different applications. And again, the APU gives you that unique architecture for that low power. We anticipate Plato to kick in sometime in 2028. And then we're already having discussions with partners on what the generation after Plato looks like.

Some of our successes have come from the mill defense areas. We've been very successful with SBIRs, which are essentially grants from the government. To date, we've been awarded $4.4 million worth of SBIRs. I kind of -- I'll start at the bottom about these SBIRs. We had our first Phase 1 win with the U.S. Army second half of last year for $250,000. We also had a win with both the U.S. Air Force Research Labs and the Space Development Agency. Both were worth over $1 million. The Space Defense Agency SBIR was extended for another $751,000. And the purpose for that extension is SDA wanted to see what our commercial chip look like on a robust level. So we're taking this grant to put our commercial Gemini-II under radiation beam and other ionization type testing to see what it can do.

And then another grant, which will be coming in is the POC that we discussed with our partner, G2 Tech in Israel for a program called Sentinel that is for DoD or DoW and another foreign defense agency. And this is for a drone camera perimeter security environment. And this is something that we'll be together with G2 Tech demoing to these agencies in the summertime.

As far as future opportunities for SBIRs, we have a pipeline somewhere between $6 million and $10 million that we have submitted. We feel especially good about one of the submissions that we've done for the U.S. Army on the Phase 2. It's for a ruggedized edge node that can be used for a lot of different applications from SAAR to object detection to drones. We are hoping to hear back on that one shortly.

And then we're also going after other funding sources are a little larger than SBIRs. We're looking at ones like STRATFI and TACFI and other ones like BAA, which will give tens of millions of dollars of grant money. And of course, we're also looking at partnerships with potential customers to have some other strategic funding as well.

As far as the financial overview, the revenues have been growing nicely over the last 1.5 years or so. We dropped off a little bit last quarter, but we're running over $6 million. Operating expenses have been running about $7 million a quarter. You could see a bump up there in our December quarter. That was with the purchase of the IP required for the Plato design to start. That was just over $3 million worth of IP that we purchased that quarter. And as far as the cash and cash equivalents between the $47 million raise that we did in October and some of the ATM purchases during that quarter, our cash went up significantly to just over $70 million.

Just on a quick overview of the legacy product line of the SRAMs. We do have the highest density, highest performance memories in the market space. We are at least 1 to 2 generations ahead of our nearest competitor. And good news is all of our competitors have frozen their road maps. So we'll continue to enjoy that leadership position for the future. The SigmaRAM, SigmaQuad family have really been driving the gross margins and the revenue because they attribute over 50% of our revenues from those families. We're taking that legacy SRAM and we're hardening it. We have hardened it. We've made Rad-Hard and Rad-Tolerant products. These are for satellites anywhere from the LEO satellites to the GEO satellites.

What's unique about this market is it's very high ASPs and gross margins. If you look at the kind of the range of ASPs, a lower density Rad-Tolerant might be a few thousand dollars and a high-density Rad-Tolerant could be as high as $30,000. And again, with gross margins north of 90%. So this is a market that we've talked about in the past. We've sent out lots of different samples and prototypes for several programs, and we're just waiting for those to go into production. It's a long design-in cycle for these markets.

So in summary, again, this computer memory device we have, this architecture really is unique in the fact that it allows us to really decrease latency, decrease power consumption makes us a perfect use case for edge applications. And again, anything from drones to satellites, any autonomous systems, smart cities, anything like that is a perfect market to address. We have proven advantages. The SBIR wins, the Cornell paper that went out last quarter that showed -- compared us to a GPU for RAG application, and it showed that we were more than 95% less power for the same performance. And then also the POC, the drone surveillance POC program I mentioned, that was a bake-off between us, Qualcomm and NVIDIA that we won. So it's certainly proven advantages. So we like to refer to ourselves as kind of an AI start-up.

But in a lot of cases, I mean, in a lot of cases, we're really not a semiconductor start-up. So remember, we've been in the business for 30 years selling and shipping SRAMs. And during that time, we shipped over 100 million SRAMs. The manufacturing process we're using for SRAMs will be the same for our APU. I mean we'll be using the same wafer foundry. We'll be using the same assembly house. We'll be using the same testing. And so we have a proven model, 30 years' experience for when we ramp the APU. And again, those SRAMs have been funding nondilutive the APU R&D. And lastly, we have a strong balance sheet with, again, over $70 million in cash with no debt.

At this point, I'll open it up to questions and -- or Q&A.

Okay. So the first question, given Gemini-II's ultra low power, low latency performance, where do you see the strongest initial deployment verticals across drones and surveillance?

So yes, certainly, drones, we've done a significant amount of software work on SAAR, object detection and now time-to-first-token. And then again, this POC that we're doing, it's with the demo going in the summer, it's with the intent to sell these to the DoD and other government agencies. So certainly, that will be the first area.

What portion of future revenues do you expect to come from defense versus commercial edge applications over the next 2 to 3 years?

That's a good question. It's going to start -- it's going to be leaning more towards the defense versus commercial. That's because that's where we've seen our early successes between the interest from the DoD, SBIRs that we've won along with this POC, we certainly see that, that's going to be the first entry into revenue. But again, we're trying to follow that up quickly with other applications like smart cities. But for the next 2 to 3 years, it will be leaning much more towards the defense side.

You see here what is the size of the Gemini-II cache?

Right now, it's 96 megabyte of memory on the Gemini-II. That's 8x more than Gemini-I. So candidly, on the Plato, it will be less on the internal cache. But as you recall, with Plato, it's going to be a different application. It's going to be LLMs. So LLMs, by definition, Large Language Models, won't fit on inside of a chip. And so what we've done is we've actually lowered the cache on that one in order to be able to really increase the pipeline, the bandwidth to be able to get extra data in the chip. But 96 megabyte for Gemini-II.

Did you publish the details for where the 3-second figure for the Jetson Thor came from?

That came from the drone manufacturer who did the bake-off. The numbers that we presented were the numbers that they did in a benchmark program that they did.

So -- I'm sorry, are you going to move from DDR4 to HBM2 to avoid external memory bottlenecks?

No. So we're actually going to be going to GDDR5 for Plato. So we -- again, these are edge applications. And so we need to keep them low power and lower cost as well. So it's prohibitive of moving to HBM for that. So we'll be going with the GDDR5.

Comment -- please comment on the ATM. So right now, the ATM is not active. I think there is a couple of million dollars that were left on it. But at this point, it's not active.

Let me see. Okay. So are potential partners already driving decisions making the Plato road map? Yes. And again, the current Plato program as well as, we'll call it, Plato 2, a future Plato. So yes, those are definitely with partners.

Okay. So we hear, okay. So is the 98% power savings reported by Cornell based on simulated HBM external memory interface? Candidly, I don't believe it was simulated. I think that they actually did a program, but the Cornell Paper is out there for you to be able to read it and go through that. But candidly, I don't remember what the memory interface was.

Is manufacturing in Taiwan? Yes. So we use, as I mentioned, TSMC as our fab. We use a company called ASEK to do our assembly, and we do our own testing. We have a facility in Sunnyvale, California, which we do all the R&D testing. And then when it goes into mass production, it goes to our Taiwan facility for the testing.

Okay. Let me see here. Okay. You received a reported 3-second time-to-first-token for Gemma-3 12B on Gemini-II. Can you confirm that the benchmark was achieved on the final product silicon? And are there plans to release a live demo for third-party audit? So yes, the 3-second time-to-first-token was done on the current production -- I'm sorry, current silicon, which is going to be our production silicon. And as I mentioned, the demo right now is going to be done in -- sometime in the summer is when it's scheduled to be.

Let me see here. Do you expect to raise more money anytime soon? And are you looking -- are you talking to any strategic investors? Also, is the SRAM business potentially up for sale?

So certainly, to scale, we're not actively looking to raise more money, but more money will be needed, I'm sure, in the future. Is the SRAM business potentially up for sale? Well, that's a good question. Certainly, if the right opportunity came, we would certainly entertain that.

At this point, I am actually out of time. Operator?

Thank you, sir. Ladies and gentlemen, that concludes GSI Technologies, Incorporated's presentation. You may now disconnect, and please consult the conference agenda for the next presenting company.

Welcome to GSI Technology's Third Quarter Fiscal 2026 Results Conference Call. [Operator Instructions]

Before we begin today's call, the company has requested that I read the following safe harbor statement. The matters discussed in this conference call may include forward-looking statements regarding future events and the future performance of GSI Technology that involve risks and uncertainties that could cause actual results to differ materially from those anticipated. These risks and uncertainties are described in the company's Form 10-K filed with the Securities and Exchange Commission.

Additionally, I have also been asked to advise you that this conference call is being recorded today, January 29, 2026, at the request of GSI Technology. Lee-Lean Shu, the company's Chairman, President and Chief Executive Officer, will be hosting the call today; with him are Douglas Schirle, Chief Financial Officer; and Didier Lasserre, Vice President of Sales.

I would now like to turn the conference over to Mr. Shu. Please go ahead, sir.

Good afternoon, and thank you for joining us to review our third quarter fiscal 2026 results. I am encouraged by our overall progress this quarter. Revenue in the third quarter increased by 12% year-over-year and 28.5% on a fiscal year-to-date basis. Demand for SRAM products remains solid. And we expect strong sales from our largest customers in the first half of calendar 2026.

After completing our financing in 2025, will [indiscernible] our APU road map. We began play to hardware development after purchasing the required IP. We have also added counter-engineers to support our hardware design team, keeping us on track to tape-out Plato in early 2027. We finalized the agreement with G2 Tech that is a base AI company for our recently announced proof of concept. We are partnering with G2 Tech on [ Sentinel ] program for autonomous [indiscernible] security using drones and cameras.

The project is [indiscernible] by the U.S. Department of War and foreign government agencies. This government funding will offset our cost to build a software stack and the liabilities needed for this project. Didier will share more details shortly. Another [indiscernible] milestones is Gemini-II time-to-first-token benchmark around in the press release earlier today. For those who have not reviewed it please see today's press release for the full detail on the benchmark result and the methodology.

Accordingly, we reported [indiscernible] time-to-first-token TTFT performance for [indiscernible] with tax and video input, consuming approximately 30 watts of system power. Compared to third-party testing of competitive graph, Gemini-II TTFT delivered up to 3x faster first token at lower power than the competitive chip on the same workload. We believe these test results validate Gemini-II [indiscernible] each use cases that need low power and low latency.

In his comments, Didier will expand on the benchmark results. We are making steady [indiscernible] continue to improve Gemini-II performance and completing the[indiscernible]. We are also pursuing early proof of concept and prototype opportunities for Gemini-II in system in the different programs, including [indiscernible] systems, a in selected commercial age deployment. In parallel, we continue to pursue non-diluted R&D funding through government defense forwards and the strategic partners.

With that, I hand the call over to Didier.

Thank you, Lee-Lean. I'll start by expanding on some of the Lee-Lean comments. On the [ Sentinel ] POC, we expect to receive more than $1 million in government funding. We will record this as an offset to R&D expenses. We plan to use it to complete key software milestones for the projects, including software development for [ GEMA312B ] on our Gemini-II ahead of the planned demonstration to the government agencies later this year. Our POC partner, G2 Tech, is receiving additional funds to develop the drone platform for this demo.

I'm pleased to share that G2 Tech conducted a competitive evaluation, and GSI was selected based on Gemini-II's performance, delivering the lowest TTFT at 30 watts. If the government evaluation later this year is successful, it could lead to a potential Gemini-II design win with G2 Tech, and we would move to pursue additional opportunities to other drone and unmanned system customers beyond the POC sponsors.

Turning to today's press release. Our Gemini-II TTFT benchmarks, we discussed preliminary results showing a 3-second time-to-first-token for a multimodal model at the edge using video and text inputs at approximately 30 watts of power -- of system power. TTFT is how long it takes the system to produce the first response, which is critical for drones and unmanned systems. The threshold for a useful TTFT in video surveillance to ensure nothing is missed is 3 seconds. That means we are sampling the video image every 3 seconds. If the TTFT is 10 seconds, it takes too long, so the surveillance video could miss something.

In preparation for the Sentinel demo, we will continue improving TTFT over the next 5 months to further reduce Gemini-II's first time to response. What's exciting for GSI about these Gemini-II preliminary benchmark results is that they demonstrate what compute and memory can provide for physical AI, faster time-to-first-token and materially lower chip power. This will help enable a broader set of viable cost-effective deployments.

For added color, at CES 2026 there was a clear shift towards edge AI and physical AI systems that must make real-time decisions under tight power constraints. In that context, Intel noted that tops the number of operations per second doesn't tell the whole story. What matters more than edge AI and physical AI is real-world load -- workload performance and efficiency. That is the takeaway for us as well.

For edge inference performance per watt and responsiveness matter more than peak training metrics. We are confident that our computer memory APU architecture designed to reduce data movement is well suited for power-constrained edge inference. Our near focus is to continue validating this with additional benchmarks and customer proof of concepts and convert that progress into design wins for Gemini-II. And to be clear, we are not trying to compete with folks at training and data centers. Our goal is to be a strong option for fast low-power edge AI applications.

Switching to the customer and product sales breakdown in the third quarter of fiscal 2026. Sales to KYEC were $1.1 million or 17.9% of net revenues compared to $1.2 million or 22.7% of net revenues in the same period a year ago and $802,000 or 12.5% of net revenues in the prior quarter. Sales to Nokia were $675,000 or 11.1% of net revenues compared to $239,000 or 4.4% of net revenues in the same period a year ago, and $200,000 or 3.1% of net revenues in the prior quarter.

Sales to Cadence Design Systems were $233,000 or 3.8% of net revenues compared to $971,000 or 17.9% of net revenues in the same period a year ago and $1.4 million or 21.6% of net revenues in the prior quarter. Military Defense sales were 28.5% of third quarter shipments compared to 30% of shipments in the comparable quarter a year ago and 28.9% of shipments in the prior quarter. SigmaQuad sales were 41.7% of third quarter shipments in fiscal 2026 compared to 39.1% in the third quarter of fiscal 2025 and 50.1% in the prior quarter.

I'd now like to hand the call over to Doug. Go ahead, please.

We reported net revenues of $6.1 million for the third quarter of fiscal 2026, compared to $5.4 million for the third quarter of fiscal 2025 at $6.4 million for the second quarter of fiscal 2026. Gross margin was 52.7% in the third quarter of fiscal 2026, compared to 54% in the third quarter of fiscal 2025 and 54.8% in the preceding second quarter of fiscal 2026. The decrease in gross margin in the third quarter of 2026 was primarily due to product mix.

Total operating expenses in the third quarter of fiscal 2026 were $10.1 million, compared to $7 million in the third quarter fiscal 2025 and $6.7 million in the prior quarter. Research and development expenses were $7.5 million, compared to $4 million in the prior year period and $3.8 million in the prior quarter. The increase in research and development spending compared to the prior quarter is primarily due to the purchase of IP for the development of Plato and associated consulting expenses. Selling, general and administrative expenses were $2.6 million in the quarter ended December 31, 2025, compared to $3 million in the prior year quarter and $3 million in the previous quarter.

Third quarter fiscal 2026 operating loss was $6.9 million compared to an operating loss of $4.1 million in the prior year period and an operating loss of $3.2 million in the prior quarter. Third quarter fiscal 2026 net loss included interest and other income of $3.6 million, reflecting a noncash accounting adjustment of $6.2 million for the change in fair value of the prefunded warrants and issuance costs of $2.8 million for the recent registered direct offering and a tax benefit of $251,000 compared to $70,000 in interest and other income and a tax provision of $44,000 for the same period a year ago.

In the preceding second quarter, net loss included interest and other income of $43,000 and a tax revision of $41,000. Net loss in the third quarter of fiscal 2026 was $3 million or $0.09 per diluted share compared to a net loss of $3.2 million or $0.11 per diluted share for the second quarter of fiscal 2026. For the prior year fiscal quarter of 2025, net loss was $4 million or $0.16 per diluted share. Total third quarter pretax stock-based compensation expense was $783,000 compared to $429,000 in the comparable quarter a year ago and $856,000 in the prior quarter.

Beginning this quarter, GSI is expanding the cash disclosures in its quarterly earnings release process to help investors understand the company's cash consumption and cash generation. Going forward, we will disclose the beginning cash balance net cash used by operating activities, net cash used by investing activities and net cash provided by financing activities. This will complement the condensed consolidated statement of cash flows included in our Forms 10-K and 10-Q.

Cash flows for the quarter ended December 31, 2025, and thousands of dollars. Cash and cash equivalents as of September 30, 2025 were $25.3 million, Net cash used in operating activities was $7.9 million. Net cash used in investing activities was $296,000, and net cash provided by financing activities were $53.5 million. Cash and cash equivalents as of December 31, 2025, for $70.7 million. The increase in cash and cash equivalents as of December 31, 2025, primarily reflects $46.9 million in net proceeds from the company's October 22, 2025 registered direct offering.

Cash used in operation activities include spending for the development and commercialization of Gemini-II and Plato. December 31, 2025, we had $70.7 million in cash, cash equivalents compared to $13.4 million at March 31, 2025. Working capital was $71.7 million as of December 31, 2025, versus $16.4 million at March 31, 2025. Stockholders' equity as of December 31, 2025, was $83.6 million compared to $28.2 million as of the fiscal year ended March 31, 2025.

Before I hand the call over to the operator for Q&A, I'd like to provide the fourth quarter fiscal 2026 outlook. Current expectations for the upcoming fiscal fourth quarter of net revenues in the range of $5.7 million to $6.5 million with gross margin of approximately 54% to 56%.

Operator, at this point, we will open the call to Q&A.

[Operator Instructions] And our first question comes from the line of Quinn Bolton with Needham & Company.

This is Robert [indiscernible] on for Quinn here. Congrats on all of the progress on Gemini-II. I just wanted to ask, you also announced during the quarter a partnership with G2 Tech as well. And that application for defense. Maybe how important is kind of the defense applications for Gemini-II? How does that establish the capabilities of Gemini in sort of real-world applications? And can you speak to potential commercial uses beyond kind of drones and defense and how this may impact the business going forward?

Sure. Sure. So certainly, the mil and defense area have been the sectors that we've had our early successes in. We've talked about it in the past. We've had some of the SBIR wins with entities under the DoD or DoW as it's called today, specifically with the Air Force Space Development Agency and U.S. Army. So we certainly have had some successes there. And in getting the message out.

We've also talked about a SAR application, a Board that we sent out to an offshore defense contractor for [ LEO satellite ] for SAR in which they're doing their evaluation now. So certainly, this is the area that has adopted our technology most quickly. And as you mentioned, with G2 Tech, it's really -- it's a nice partnership because they're able to actually bring a product using our subsystem APU to create a true product. In this case, like you said, a drone and camera surveillance system.

Can you repeat the last part of the sentence -- of the question though.

[indiscernible]

So a good point. So the effort we're doing right now with this POC, this time-to-first-token and the whole [indiscernible] lends itself to other applications outside of drones and unmanned vehicles, things like smart cities, things like robotics. And so certainly, we will be able to leverage all the work we're doing now for this current POC with G2 Tech for these other markets as well.

For sure. And just one more on -- you mentioned the government funding as a catalyst for 2026. Can you talk through maybe potential time lines? When do you expect this funding to come in? And any other details that you have on that front would be great.

Yes. So for SBIR, we have a continuous pipeline of submittal. So we have a handful right now that have already been submitted, and we're waiting for word on whether we've been awarded or not, and we have others that we are putting together. This is an ongoing process. And there are different levels. They fall under the classic SBIRs. There's also other areas like BAA, which stands for a broad agency announcement, I believe, [indiscernible]. There are other programs where other fundings are involved. And we're active in all those areas. Again, the benefit of this funding is first. And secondly, it allows us to get more exposure within the DoD elements for future business.

[Operator Instructions]

[Audio Gap]

Since there are no more questions from the investor. So thank you all for joining us. We look forward to speaking with you again before our fourth quarter and the full year fiscal 2026 results. Thank you.

Thank you. This concludes today's conference. You may disconnect your lines at this time, and thank you for your participation.

Ladies and gentlemen, thank you for standing by. Welcome to GSI Technologies Second Quarter Fiscal 2026 Results Conference Call. [Operator Instructions]

Before we begin today's call, the company has requested that I read the following safe harbor statement. The matters discussed in this conference may include forward-looking statements regarding future events and the future performance of GSI Technology that involve risks and uncertainties that could cause actual results to differ materially from those anticipated.

These risks and uncertainties are described in the company's Form 10-K filed with the Securities and Exchange Commission. Additionally, I have also been asked to advise you that this conference is being recorded today, October 30, 2025, as the request of GSI Technology. Hosting the call today is Lee-Lean Shu, the company's Chairman, President and Chief Executive Officer. With him are Douglas Schirle, Chief Financial Officer; and Didier Lasserre, Vice President of Sales.

I would now like to turn the call over to Mr. Shu. Please go ahead, sir.

Good afternoon, everyone, and thank you for joining us today. Let me start by highlighting two recent and important events for GSI. First, we announced a research paper published by Cornell University in mid-October. The paper verified that our Gemini-I chip performed on par with NVIDIA's A6000 on certain AI tasks, while consuming roughly 98% less energy.

This paper validates the disruptive potential of our compute in memory design, particularly for the near-term commercialization of Gemini-II with 8x the memory and 10x the performance of Gemini-I. Gemini-II is positioned to deliver superior processing at a fraction of the power when compared to existing solutions.

This brings me to my second point. The market quickly recognized the significance of our compute-in memory validation with the Cornell paper. Building on the momentum from the papers funding, we closed a $50 million equity financing. We are now deploying the capital to accelerate execution across our hardware and the software build-out, making this a pivotal period for GSS growth.

Post funding, we are working on the initiative in parallel. First, we have begun the work to acquire the necessary IP for Plato, which will allow us to start hardware development. This IP provides crucial connection to support broader system interface and prototyping for future customer applications.

To accelerate Plato's time to market and capture market opportunity sooner, we are building additional software team to support Plato. Second, to expedite the build-out of our Gemini-II software solutions and applications, we are investing in all the layers that make the platform more accessible and flexible for developers.

These software tools are essential for customers integrating Gemini-II hardware into AI and signal processing workflows, particularly in edge and the defense applications, where efficiency and the low power provide a competitive advantage.

Looking ahead, our initiatives for calendar year 2026 are centered on converting proof-of-concept projects into commercial customers and expanding those relationships into large production programs. Didier will provide an update on where those efforts stand today.

To sum up, our post-funding initiative was target and discipline -- will be targeted and disciplined. We are rapidly moving forward with the Plato hardware design and the software development, ramping up our Gemini software ecosystem and strengthening ties with key defense and government partners in our POC and Small Business Innovation Research or SBIR programs.

This action position GSI to turn technical progress into commercial success in the high-value edge and defense applications such as drone military vehicles, satellite and other use cases and ride the wave of AI compute innovation.

Now I hand the call over to Didi, who will provide more details on this topic and discuss our business development and the sales activities. Please go ahead, Didi.

Thank you, Lee-Lean. Let me expand on the topics that Lee-Lean just highlighted. We continue to advance our ongoing projects, including our SBIR and POC engagements with potential customers. Recently, Gemini-II has been approved for prototyping by the offshore defense contractor to whom we shipped a board and software to a few months back.

This POC focuses on synthetic aperture radar or SAR applications for drones and other edge systems. What's exciting here is that our solution delivers the required performance while maintaining an extremely low power profile around 15 watts, making it ideal for compact energy-constrained environments.

For added context on just how competitive the solution is, an incandescent light bulb uses about 4x more wattage than our solution. We are also involved in a joint POC involving two defense organizations and a drone integration partner. This Gemini-II project combines YOLO model, we developed with multimodal large language model processing at the edge, specifically targeting time to first token, a key performance metric for drones. Along with our partner, we successfully demonstrated the end-to-end application to one of the potential end customers.

Gemini-II outperformed the competing solution, particularly in how quickly the model produces its first response. We are now optimizing the algorithm and expect to publish initial benchmark results before year-end with a fully optimized version available in the first half of calendar 2026. This algorithm would be for defense applications such as drones, satellites and other military vehicles.

Gemini-II is a central part of the near-term commercialization road map, and we are encouraged by the customer engagement and technical validation that is being received. Turning to our Plato program.

We are embarking on the journey towards a major milestone, the tape-out of Plato chip in early calendar 2027. Over the next year or so, we plan to actively engage several strategic partners for Plato who could provide funding and collaborate on testing and prototyping early versions of the chip.

Their involvement would also support the development of software libraries and APIs, ensuring that Plato becomes a versatile, scalable solution across multiple markets, starting with defense.

In military and defense applications, the APU's high-performance and low-power capabilities provide unique advantages. And Plato will further enhance critical functions such as SAR imaging, object recognition, GPS-denied navigation and data fusion for drones and military vehicles, delivering real-time tactical capabilities in compact mobile systems.

Plato's design builds directly on the foundation of Gemini-II. To accelerate time to market, we are acquiring building block IP that allows us to focus on differentiation rather than reinventing core components.

Strategic partners would play a critical role, not just in meeting our ambitious time line, but in shaping the chip's capabilities, validating its performance in real-world applications and guiding future enhancements.

Their technical collaboration and early adoption would position us to deliver a highly optimized field-tested solution, strengthening our long-term leadership in specialized AI compute architectures well beyond the immediate financial support.

And lastly, a comment on our SBIR work. We recently received a $751,000 extension of one of our Space Development Agency contracts, which includes additional funding for radiation hardened beam testing of Gemini-II. The goal of this testing is to evaluate the robustness of the current Gemini-II commercial chip for possible use in satellite and other aerospace applications.

While it's too early to confirm specific designations, we see this as a significant opportunity. Let me now switch to the second quarter's commercial -- I'm sorry, customer and product breakdown. By revenue, I am referring to net revenue in the following comments.

In the second quarter of fiscal 2026, sales to KYEC were $802,000 or 12.5% of revenues compared to $650,000 or 14.3% of revenues in the same period a year ago and $267,000 or 4.3% of revenues in the prior quarter. Sales to Nokia were $200,000 or 3.1% of revenues compared to $812,000 or 17.8% in the same period a year ago and $536,000 or 8.5% of revenues in the prior quarter. Sales to Cadence Design Systems were $1.4 million or 21.6% of net revenues compared to 0 in the same period last year and $1.5 million or 23.9% of revenues in the prior quarter.

Military defense sales were 28.9% of second quarter shipments compared to 40.2% of shipments in the comparable period a year ago and 19.1% of shipments in the prior quarter. SigmaQuad sales were 50.1% of second quarter shipments in fiscal 2026 compared to 38.6% in second quarter fiscal 2025 and 62.5% in the prior quarter.

I'd now like to hand the call over to Doug. Please go ahead, Doug.

Thank you, Didier. The company reported net revenues of $6.4 million for the second quarter of fiscal 2026 compared to $4.6 million for the second quarter of fiscal 2025 and $6.3 million for the first quarter of fiscal 2026.

Revenue growth in the quarter was driven by strong market momentum for leading SRAM solutions. Gross margin was 54.8% in the second quarter of fiscal 2026 compared to 38.6% in the year ago quarter and 58.1% in the preceding first quarter of fiscal 2026.

The decrease in gross margin for the second quarter of 2026 was primarily due to a change in the product mix. Total operating expenses in the second quarter of fiscal 2026 were $6.7 million compared to $7.3 million in the second quarter of fiscal 2025 and $5.8 million in the prior quarter.

Research and development expenses were $3.8 million compared to $4.8 million in the prior year period and $3.1 million in the prior quarter. The increase in research and development spending compared to the prior quarter is primarily due to changes in the level of stock-based compensation expense and amounts of government funding received under SBIRs in each quarter recorded as an offset to research and development expense.

Selling, general and administrative expenses were $3 million in the quarter ended September 30, 2025, compared to $2.6 million in the prior year quarter and $2.7 million in the previous quarter. The second quarter fiscal 2026 operating loss was $3.2 million compared to an operating loss of $5.6 million in the prior year period and an operating loss of $2.2 million in the prior quarter.

Second quarter fiscal 2026 net loss included interest and other income of $43,000 and a tax provision of $41,000 compared to $149,000 in interest and other income and a tax provision of $23,000 for the same period a year ago. In the preceding first quarter, net loss included interest and other income of $13,000 and a tax provision of $54,000.

Net loss in the second quarter of fiscal 2026 was $3.2 million or $0.11 per diluted share compared to a net loss of $2.2 million or $0.08 per diluted share for the first quarter of fiscal 2026. For the prior year second fiscal quarter of 2025, net loss was $5.5 million or $0.21 per diluted share.

Total second quarter pretax stock-based compensation expense was $856,000 compared to $663,000 in the comparable quarter a year ago and $341,000 in the prior quarter. At September 30, 2025, the company had $25.3 million in cash and cash equivalents compared to $13.4 million at March 31, 2025. Working capital was $26.8 million as of September 30, 2025, versus $16.4 million at March 31, 2025.

Stockholders' equity as of September 30, 2025, was $38.6 million compared to $28.2 million as of the fiscal year ended March 31, 2025.

Lastly, for the third quarter of fiscal 2026, we expect net revenues in the range of $6.0 million to $6.8 million with gross margin of approximately 54% to 56%.

We remain focused on disciplined execution to bring Gemini-II to market, advance our road map for Plato and drive long-term shareholder value. Operator, at this point, we'll open the call to Q&A.

[Operator Instructions] Our first question comes from Robert Christian, private investor.

I'd like to congratulate you on the Cornell verification. But I'd also like to know, have you done any work with the auto industry on autonomous vehicles?

We have not yet. So as we've talked about in past calls, we've certainly have limited resources, and that takes a tremendous effort for that market space. So we're currently starting in the military defense arena, but we certainly believe our technology will adapt well in those areas. And so that's certainly a focus for us in the future, but not yet.

Our next question comes from Mark [indiscernible] private investor.

I had a question on the $50 million placement you recently did, was that with a strategic investor? What sort of investor? And was there a holding period to that stock?

No, it was just someone that was interested in the company, wasn't strategic in any way. And there is no required holding period for the shares.

Got it. Okay. And then just a follow-up to that. Were there -- have there been any strategic circling at all now post the Cornell report?

Can you repeat the question, please?

Have there been any inquiries from more strategic investors since the report came out from Cornell?

There are things that we're looking at and parties that we're talking to. But I wouldn't say that there's anything that anyone that we haven't already considered working with at this time. As Didier said, we have limited resources, and I think we have some very significant opportunities that he's already mentioned.

Our next question comes from David Zalkowitz with ISQ.

Yes. Is there any plan to have Cornell or another third party validate the Gemini-II information, a different technology? I know the Cornell report was Gemini-I. So is there a plan to do that similar type of analysis for Gemini-II?

Yes. So you're absolutely correct. So Cornell actually received this Gemini-I board many years ago, and they've actually written a few other papers. And so they -- this was a continuation of that original board. And we are talking to them about getting a Gemini-II board to them and also other researchers as well.

Okay. And then I guess you're talking -- you're working with the military. I guess, I didn't see anybody on the Board or senior management team that has real military defense experience. Is there any plans to beef up that area of the management or the Board of Directors in order to target those applications?

Yes. No, that hasn't come up as a discussion or topic on the Board. At this point, there are no plans to revise the Board. It doesn't mean that we won't in the future, if it makes sense, though.

Okay. And then I saw you're developing your own large language model, which you're going to release some information on at the end of the year. Just curious why you wouldn't just use the plethora of large language models that are already out in the market and why spend resources developing your own?

No, we are not developing our own large language model. We are working on the open source large language model like Gemma III. So...

I'm just reading a press release, the press release says currently developing a multi-modal LLM that targets edge applications.

Correct. Yes. For 12b -- so Gemma-312B that's the model, and we're developing our algorithms to work with that model.

Okay. And why would you do that as opposed to utilizing other LLMs are already developed?

In this case, it was the definition from the POC that we're working on. So as we talked about, there are 2 government entities that have approached us and a partner to do a POC, and that is the model that they requested.

Yes. Also, there are certain aspects of the model, which support multi-modal well, okay? So they can support the image very well and in addition to the text. That's why they pick on this one.

Our next question comes from [ Christian Rug from CER Holdings. ]

I was wondering, how are you differentiating your APU versus GPU competitors in terms of power, latency and cost efficiency?

That's a pretty broad question. And so if you look at the Cornell paper, that certainly hits on the power. The comparison was to an NVIDIA GPU and the use case they use, the performances were on par, but we were 98% less power. So that certainly shows that. With the SAR algorithm that we've been talking about, certainly, our image creation time is faster at a lower power footprint as well.

So what we've done is we've done benchmarking on certain use cases based off of input from customers on what they'd like to see. So there are times where we beat them strictly on power. There's times we beat them strictly on performance. Well, I shouldn't say that, we've never lost to them on power. But there certainly are times that we have the advantage on both performance and lower power.

Okay. And then my second question is, given the performance claims and potential of Gemini-II APU, have you had any engagement or partnership discussions with larger semiconductor or AI-focused companies?

So right now, we're focused on the customers at this point. We haven't had any discussions at least recently with other semiconductor companies.

Okay. And then my last question is, how does the power factors play into building AI data centers at a large scale?

So we're focused on the edge right now. And so everything we've talked about right now is the edge. And so certainly, the data centers have a real power issue as well. There's no secret there. But what we've been focused on right now with Gemini-II and certainly with the next-generation chip Plato will be at the edge.

And so if you look at -- as we discussed with Gemini-II, this project we did with this offshore defense contractor, we limited -- we limited our chip to 1 of the 4 cores that are there to get it down to 15 watts. If you look at Plato, depending on how it's used, can be as little as 4 watts and maximum 12 to 15 watts. So we're really focused at the edge, not in the data center.

Our next question comes from Michael Roberts from Roberts Capital. It seems Michael has gone silent. We will move on to the next question. Michael Cooper, private investor. You may proceed with your question.

Can you talk about the total addressable market that you're looking at over the next 5 years? And then how you expect that to ramp? I'm guessing you have a number of different scenarios, maybe a range of scenarios. You could give us a sense for how large this market is for these markets? I'm sure you're looking at various markets. And then what kind of price points your boards or chips go into products?

Sure. It's a good question. So Michael, I don't have the numbers in front of me. But certainly, there was a report very recently that was issued by one of the researcher analysts at Needham & Company that discussed the drone market specifically.

And I don't have it in front of me, but I want to say it was either -- it was tens of billions at least market size. I want to say it might be larger than that. And so certainly, it's a very, very large market. And as we've discussed, we certainly feel with the power profile of our chips, along with some of the algorithm work that we're doing for, like Lee-Lean mentioned, the multimodal inputs, whether we take an image or text or voice in the future, along with the time to first token advantage that we have. We certainly think that we're well positioned to address that market.

That was question one. The second question was -- I think it was a two-part question you had. Pricing. Yes. So pricing, I mean, we'll give you generalities. But certainly, it's going to be priced differently by market, but it could be a few thousand dollars a board to $10,000 a board that contains the chip. And then the chip will sell -- again, based on the market, but the chip could be $1,000 or more depending on the market and the volume.

And you're working in gross margins in the 80-ish percent range?

Yes, it will be above where we are corporately today. And again, it really depends on the market and how it's sold. It could be 60% to 80%. It really depends on how it's sold, whether it's in a board, in a server, whether it comes with software or not. I mean there's a lot of different aspects that would move that margin needle.

Our next question comes from Michael Roberts with Roberts Capital, who is rejoining us.

On capital deployment on the $50 million raise, can you give an idea of how that plans to be allocated, whether it's percentage or dollar amount, amongst the Gemini-II completion, software development and the new Plato chip that you referenced?

Yes. On the Plato because there's some fixed costs that we have to spend like IP costs and the mass tape-out costs. So those are fixed $15 million, $60 million, $70 million kind of range, okay? And the rest of them, I think they're probably pretty even between the Gemini-II and the Plato, that's mostly engineering costs, the internal cost, and it will be distributed even inside the company.

Evenly across. Okay. And in terms of then based on your cash runway now, what revenue or gross margin level do you expect to reach operating breakeven then?

If you can assume I don't know, 65% to 70% gross margin once we get into this. It's something that I need to take a look at. We're still putting our plans together in terms of hiring levels and so on. SoC teams or whoever we need for the chip development, additional software teams that we need for the software development. I don't have all those numbers yet to do a calculation.

Understood. But are there concrete milestones and dates then for the Gemini-II in terms of expectation of pilot shipments or expected initial production orders?

Yes. So we will be doing some pilot shipments. We've done a couple already or we plan on doing more in the first half of 2026 calendar. This POC that hopefully, we'll be able to discuss a lot more in the upcoming months, depending on the schedules on that could give more substantial revenues in the back half of calendar 2026.

Noted. And from the current evaluation customers now, has any purchase orders or letters of intent been provided yet?

I'm sorry, could you repeat the question?

Yes. Have any of the evaluation customers provided any purchase orders or letters of intent yet against that production?

They're still in their evaluation at this point. So as we talked about, the Board that we sent along with the software to this offshore defense contractor, they have done a review and they've put us as what's called good acceptance in their system, which means it's passed and been accepted. And so now we're going through the possible use cases.

They have a couple of different divisions. Two of them, we think will be a good fit. One obviously is the SAR division. The other one is what they call their AI division. And so we're looking for practical applications that can then, like you say, turn into design wins and revenue. So that's -- we're doing that with the customer today.

All right. Very helpful. And one last question, then I'll let others proceed. Can you elaborate on that software stack maturity then, the compiler SDK model porting tools and when developers outside of GSI will have access? Going towards the ecosystem adoption of what we have?

No, for the Gemini-II, we -- right now, we are developing the library and algorithm. Now after that, we will move on to the tool and the compiler work, okay? And we are developing this with the customer -- the partner and the customer we have.

Our next question comes from Robert Christian, private investor.

Yes. Can you help me understand why the company is not going after data centers in view of the environment impact with energy consumption and cooling? It seems like we're leaving a lot of money on the table, even if it was just licensed so others could use the technology.

So I'm not sure how long you've been following the company, but we had talked about another potential road map product at the time we were calling it Gemini-III, and that was going to be geared towards the data center.

And that one needed a different kind of partner and it needed a lot more funding. It would have required a very aggressive process node and would have been much more expensive. And so we are going down that road.

And again, that was -- it was targeted for the data center. In the meantime, we were getting way too much positive feedback and interest on the edge, and we were getting SBIR dollars, and there are other dollars, research dollars that we've submitted for to try and get, and it's all for the edge.

And so the decision was made. We couldn't do both. It was one or the other at this point. And so we remain focused on the edge. Not to say with more influx of cash, we can't beef up the team and go after the data center, but it strategically made sense for us to remain at the edge for now.

Okay. But there's not a possibility, say, of NVIDIA or a Micron to come in and develop the chip and we get a percentage of it?

Yes. I mean that's certainly very possible. I can't say those discussions are happening, but we had some discussions in the past where that was kind of the model we were looking at. So the answer is yes, we could do that. It's just -- there's nothing in the hopper right now.

Our next question comes from [ Marco Petroni with MG Capital. ]

Yes. You guys just recently raised $47 million net, and you had $13 million last quarter, and the balance sheet shows only $25 million now. So I was wondering where that money went, number one. And number two, going forward, what type of capital allocations do you need for -- to build out the software team to do all this other stuff that we've been talking about?

Well, the first answer is that, that transaction closed after the balance sheet date. It was an October transaction and the $27 million that you see is as of September 30. And then in terms of capital allocation, I think we answered a previous question where we're looking at some IP and other stuff that we need to purchase for Plato, and then we expect to split funding between software development and the Plato development.

So how much cash do you have on hand currently?

Well, take $27 million or $23 million -- I'm sorry, $25 million balance sheet date plus we got another $47 million. So that should give you a reasonable estimate.

Our next question comes from [ Mohammed Alsousi from Scale. ]

I just want to know if you have attracted any interest from any potential new customers after the Cornell study on APU performance.

I'm sorry, just to be clear, you're asking if we've gotten any more customer traction because of the Cornell paper? Is that the question?

No, I want to know if you attract more interest or potential new customers after the new Cornell study on APU performance.

Okay. I think that's what I just said. Okay. So the answer is the customers we've been talking to, we've been talking about this low-power advantage for some time, and we've done benchmarks on several applications with some of our customers. And so they're aware of that. And so in that respect, it's not a surprise to our customers we've been talking to that we have this low-power advantage. This just illustrated it for the rest of the public as a third-party validation of what we've been saying.

This now concludes our question-and-answer session. I would like to turn the floor back over to Mr. Lee-Lean Shu for closing comments.

We look forward to seeing you at this event and -- and your participation in the third quarter fiscal 2026 earnings call. Thank you.

Ladies and gentlemen, this concludes our conference for today. Thank you for your participation. You may disconnect your lines and have a wonderful day.

Ladies and gentlemen thank you for standing by. Welcome to GSI Technologies First Quarter Fiscal 2026 Results Conference Call.

[Operator Instructions] Before we begin today's call, the company has requested that I read the following safe harbor statement. The matters discussed in this conference call may include forward-looking statements regarding future events and future performance of GSI Technology that involve risks and uncertainties that could cause actual results to differ materially from those anticipated.

These risks and uncertainties are described in the company's Form 10-K filed with the Securities and Exchange Commission. Additionally, I have also been asked to advise you that this conference call is being recorded today, July 31, 2025, at the request of GSI Technology. Hosting the call today is Lee-Lean Shu, the company's Chairman, President and Chief Executive Officer. With him are Douglas Schirle, Chief Financial Officer; and Didier Lasserre, Vice President of Sales. I would now like to turn the conference over to Mr. Shu. Please go ahead, sir.

Good afternoon, everyone, and thank you for joining us today. Let me begin with a few key highlights from this quarter's financial results. Fiscal 2026 is off to a strong start. In the first quarter, we achieved net revenue of $6.3 million, up 7% sequentially and 35% year-over-year.

This growth was fueled by rising demand for our SRAM chip, driven by strong market momentum for leading AI processors. Our profitability metrics also improved this quarter with a 200 basis point sequential increase in gross margin and over 1,100 basis points compared to the prior year.

We have also made meaningful progress on cost control over the last year with operating expenses declining by 15% year-over-year, excluding the gain from the sale of our headquarter in Q1 2025. Now I would like to provide an update on our product road map and customer milestones.

We have concluded the evaluation of the second spin of our Gemini-II chip. I am pleased to report that all loans have been resolved.

The Silicon is fully functional and ready for production. This week, the Leda-2 board and associated algorithm were delivered to a key offshore defense contractor for proof-of-concept work with Gemini-II for satellites and drones. This delivery keep us firmly on track with our road map and customer commitments.

And while this is a major milestone, we also believe it represent an opportunity to play in broader market with Gemini-II. Didier will provide more details on this subject in a few minutes. SI is at a pivotal point in its development. We plan to target high-growth opportunity for Gemini-II in the satellite zone and edge computing sectors. These are markets that are increasingly defined by AI-driven capabilities.

We are evaluating options to access to expand our software and application teams to then develop the platform necessary for future customer fulfillment and support. Management is actively working with the Board and our advisers to evaluate strategic options that will enable us to scale efficiently.

Our near-term priorities include funding the expansion of our software and application team and advancing the development of the platform required to support future customer development deployment of Gemini-II. Accelerating the launch of Gemini-II is key to laying the groundwork for our next-generation APU title and advancing the company's long-term product road map.

In the meantime, the ATM has provided valuable flexibility, allowing us to raise $11 million today net of fee. As a result, we ended the first quarter with a strengthen cash position of $22.7 million. Now I hand the call over to Didier, who will discuss our business development and sales activities. Please go ahead, Didier.

Thank you, Lee-Lean. Starting with our SRAM business, we had another strong quarter of sales to KYEC and Cadence Design Systems, a leading provider of AI chip emulation systems. We have experienced our third consecutive quarter of rising SRAM sales, driven by the growth with the enterprise adoption of AI and also in the generative AI by hyperscalers who are training ever larger models.

Despite continued strong demand for high-performance SRAM chips, extended lead times are impacting our second quarter of fiscal '26 sales. While customers have maintained typical ordering patterns, a portion of our backlog is not shippable this quarter due to these supply constraints. We proactively informed all of our distributors and sales representatives of the situation. It may take some time for customers to adjust to the increased lead time accordingly.

In the interim, we anticipate instances where orders cannot be fulfilled within the requested time frame. Although forecast from our largest customers remain solid, we expect SRAM revenue for the remainder of the fiscal 2026 to be stable compared to first quarter as we navigate these supply chain challenges.

Switching to deliverables for our SBIRs, as Lee-Lean mentioned, we also have completed the development of our SAR and YOLO 3 and YOLO 5 algorithms optimized for edge AI applications. In parallel, we also shipped a Leda-2 board with a low-power version of our Gemini-II chip to an offshore defense contractor with whom we have been working with for over a year. Both of these are now available for POC opportunities with other partners. Our defense work with the low-power version of Gemini-II has highlighted the chip's capability to address large models at the edge in varying capacity versions, depending on the latency and power sensitivity of the application.

This makes Gemini-II in conjunction with the SAR and YOLO 3 and YOLO 5 algorithms very well positioned for the broader market potential of applications moving to the edge and particularly for high-demand, high-volume and high mixed processing needs of drones operating in GPS-denied environments as well as next-generation satellite applications. Gemini-II is also well suited for large language models or LLMs for short, for edge applications. LLMs require a high-density, high-performance memory path from external DRAM to the internal SRAM next to the processor.

Gemini-II's compute and memory architecture provides high-density, high-performance internal SRAM to allow a high-efficiency memory path for high-speed and lower power -- I'm sorry, low-power operations required by LLMs. Gemini-II is also a bit processor that is flexible to do 1 bit to 32-bit or larger operations in the same circuit efficiently, which further enhances the capability for LLM processing. We are developing a multimodal LLM targeting edge applications and we will have benchmark results available next quarter.

To ease the adoption of the technology, we will continue to improve the AI compiler for Gemini-II, which is currently in its initial release phase. In parallel, we continue to develop ready-to-use vision, multimodal and recognition apps and libraries. Our software team is also developing dynamic, low-precision software libraries that support larger models, enabling high accuracy at low powers and edge devices. This is a major enabler for efficient edge AI as a bit engine, we are uniquely capable of addressing these edge needs where compute, memory and power resources are far limited.

As Lee-Lean mentioned, we are eager to advance our software development team to pursue drone and satellite AI chip applications with Gemini-II. Let me switch now to our first quarter customer and product breakdown. In the first quarter of fiscal 2026, sales to KYEC were $267,000 or 4.3% of net revenues compared to $1 million or 21.9% of net revenues in the same period a year ago and $1.7 million or 29.5% of net revenues in this prior quarter.

Sales to Nokia were $536,000 or 8.5% of revenues compared to $998,000 or 21.4% of net revenues in the same period a year ago and $444,000 or 7.5% of net revenues in the prior quarter. Sales to Cadence Design Systems were $1.5 million or 23.9% of net revenues compared to 0 in the same period a year ago and $642,000 or 10.9% of net revenues in the prior quarter.

Defense and military sales were 19.1% of first quarter shipments compared to 31.9% of shipments in the comparable quarter a year ago and 30.7% of shipments in the prior quarter.

SigmaQuad sales were 62.5% of first quarter shipments compared to 36.3% in the first quarter of fiscal 2025 and 39.3% in the prior quarter. Regarding our SRAM business outlook, our largest customer is currently navigating supply chain constraints. However, we expect their order volume to remain stable for the rest of this fiscal year. Meanwhile, other SRAM customers have largely normalized their inventory levels, and we anticipate continued order activity from them as well. I'd like to hand the call over to Doug. Go ahead, Doug.

Thank you, Didier. We reported net revenues of $6.3 million for the first quarter of fiscal 2026 compared to $4.7 million for the first quarter of fiscal 2025 and $5.9 million for the fourth quarter of fiscal 2025.

Gross margin was 58.1% in the first quarter of fiscal 2026 compared to 46.3% in the first quarter of fiscal 2025 and 56.1% in the preceding fourth quarter of fiscal 2025. The increase in gross margin in the first quarter of 2026 was primarily due to product mix and benefits of scale from higher revenue on the fixed cost of revenues.

Total operating expenses in the first quarter of fiscal 2026 were $5.8 million compared to $6.8 million in the year ago quarter, excluding a onetime gain of $5.7 million on the sale and leaseback of the company's corporate headquarters and $5.6 million in the prior quarter.

Research and development expenses were $3.1 million compared to $4.2 million in the prior year period and $3 million in the prior quarter. Selling, general and administrative expenses were $2.7 million compared to $2.6 million in both the prior year and previous quarter. First quarter fiscal 2026 operating loss was $2.2 million compared to an operating loss of $4.7 million in the year ago quarter, excluding the $5.7 million onetime gain previously mentioned related to the company's corporate headquarters and an operating loss of $2.3 million in the prior quarter.

First quarter fiscal 2026 net loss included interest and other income of $13,000 and a tax provision of $54,000 compared to $55,000 in interest and other income and a tax provision of $57,000 for the same period a year ago. In the preceding fourth quarter, net loss included interest and other income of $52,000 and a tax provision of $6,000 Net loss in the first quarter of fiscal 2026 was $2 million or $0.08 per diluted share compared to net income of $1.1 million or $0.04 per diluted share for the first quarter of fiscal 2025.

Net income for the year ago period reflects a $5.7 million onetime gain on the sale and leaseback transaction of the company's headquarters. For the prior fourth fiscal quarter of 2025, net loss was $2.2 million compared to a $0.09 loss per share. Total first quarter pretax stock-based compensation expense was $341,000 compared to $658,000 in the comparable quarter a year ago and $512,000 in the prior quarter.

At June 30, 2025, the company had $22.7 million in cash and cash equivalents compared to $13.4 million at March 31, 2025. Working capital was $25.7 million at June 30, 2025, compared to $16.4 million at March 31, 2025. Stockholders' equity as of June 30, 2025, was $37.4 million compared to $28.2 million as of the fiscal year ended March 31, 2025.

On an earnings conference call in May 2024, we announced that the company had initiated a comprehensive strategic view established a special committee of the Board to evaluate strategic alternatives and engaged Needham & Company as our strategic and financial adviser to assist in the process.

As Lee-Lean mentioned, we are actively evaluating potential strategic opportunities to secure the necessary capital to advance the development of our APU products. In the interim, we may choose to draw on the remaining balance of the ATM during upcoming trading windows to support near-term funding needs related to Gemini-II development depending on market conditions and other factors.

Finishing with the outlook for the second quarter of fiscal 2026, we expect net revenues in the second fiscal quarter to range between $5.9 million and $6.7 million, with gross margin in the range of 56% to 58%. We remain focused on disciplined execution to bring Gemini-II to market, advancing our road map for Plato while developing long-term shareholder value. Operator, at this point, we will open the call to Q&A.

[Operator Instructions] The first question is from [inaudible], a private investor.

So can you provide a little more color on the supply chain issues?

Sure. Yes. So I'm sure you're aware of all the tariffs that are being thrown around by the U.S. government, and a lot of these are directed at China. And so a lot of the folks who have been doing assembly in China are moving some of their assembly to Taiwan. And so it's really affecting the capacity in Taiwan. And as you know, we do all of our back end in Taiwan. So it's thrown out the lead times pretty much overnight to us because of that transition.

Will that end up making the customers possibly order earlier?

Correct. Yes. So that's something I mentioned a little earlier, which is this came about very quickly and then customers have been used to their ordering patterns based off of lead times we've quoted. So we have gone back to them via our reps and our distributors to make sure they understand they need to get more backlog coverage in place so that their future orders will not be late or delayed.

And so in the future, we anticipate this won't be a problem. But for the current quarter end and possibly into next, there will be some backlog that would have been shippable that will be delayed a bit just because of these lead times.

Yes. And it will just make the further quarters probably even stronger then once we get out of that? Possibly. Yes. So sales to KYEC seemed a little weak this quarter. Can you comment on that a little bit?

Sure. Yes. So part of that was the inventory levels that I mentioned in mind, which have seemed to stabilize along with the lead time as well. Yes, unfortunately, those orders come in within lead time, and we've been able to react in the past, and we weren't able to this past quarter.

And like with Cadence, those orders were pretty strong this quarter. What type of product are you shipping to them?

Yes. So there are emulation systems. This is kind of what we've talked about that even though we don't sell our SRAMs directly into AI applications, we do a lot of support. KYC is supporting the manufacturing of AI chips. The cadence systems or emulations to emulate the design of some of these GPUs and other devices. So it's emulation systems in the front-end design.

Okay. And last question. As far as the ATM is concerned, what are the trading windows for the company for that?

Well, typically, our trading window starts 2 days after our earnings call. So in the case of this quarter, the trading window will open on Tuesday, and it closes on the 15th of last month of the quarter. So that would mean in this case, September 15 or the last trading day up till 15th of the month.

There are no further questions at this time. I would like to turn the floor back over to Lee-Lean Shu for closing comments.

Thank you all for joining us. Please join us on August 20 at upcoming Needham Virtual Semiconductor Conference.

Actually, operator, there is one more question that just popped up.

I see that now. We have a question now from Anna Chapman from 25 Productions.

Yes. My background has always been in sales and capital equipment. I want to know how you're incentivizing your sales force because to me, you make one of the best products out there. It has an excellent portfolio how are you incentivizing these people? It seems like your sales should be more in the pipeline. That's my question.

I'm sorry, was that an advertising? Or was that a question? I'm not sure I got the question. So...

To your sales to the distributors, to your sales force, are they -- is your product in the -- one of their #1 things in their bag? Or is it like #12 or maybe an afterthought? How are you incentivizing these people to go out there and tell your story and get sales?

Yes. So our independent sales reps are paid on commission. So they're paid on shipment of product and distributors are paid on margin. And so with our independent sales reps, there's no competing lines, and they understand that our products are door openers. And so certainly, they're important lines for them. And again, with distributors, we do have large distributors. As you know, we have Avnet, which carries most of the lines. And so with them, the incentivization is in the margin and GSI generally pays them above corporate average for the margins.

Okay. All right. I think they need to do better, quite frankly. Just my opinion...

Great. There are no further questions at this time. This concludes today's teleconference. You may disconnect your lines at this time. Thank you for your participation.