GSI Technology 2027 Q1 Earnings Call

Thank you for joining us. I would like to begin by talking about where GSI is today and where we are headed. Didier will get into the details, but let me start with my view of the business. Over the past several years, we have worked hard to move GSI from an SoC chip company into an edge AI platform company to address a large and growing market for our technology. With the APU, we fill an important gap in the edge AI market. Traditional GPU delivers strong AI performance, but consume too much power for many edge applications. Mobile processors offer lower power but cannot deliver the AI performance required for many real-time workloads. The APU use less power than traditional GPUs while delivering higher performance than today's mobile processors. We believe that our APU delivers the performance and low power usage we set out to achieve.

Today, we have a Gemini-II production chip ready for customers. We are now in the next phase of GSI, turning the APU technology into a business. The first quarter showed disciplined execution as we continue advancing our ongoing projects while pushing ahead with the Plato development. For the next 18 months, our priority are clear. Successfully completing our current proof-of-concept projects, expanding Gemini-II into adjacent applications, and releasing our AI-assisted SDKs this fall. At the same time, we currently expect to keep Plato on schedule for tape-out in March 2027. With that, I turn the call over to Didier, who will provide further detail on this point. Please go ahead, Didier. Thank you, LiLin.

As LiLin said, we've reached a pivotal point in the evolution of our APU business. We have a Gemini-II solution for proof-of-concept projects across a variety of edge AI applications. Our focus has shifted from proving our technology to turning customers into proof-of-concept projects. Ultimately, our goal is to secure design wins that could lead to commercial deployments and meaningful APU revenue. Our near-term strategy is to work closely with a small number of customers and system integrators on proof-of-concept projects and applications that are well-suited to the APU's advantages. Rather than developing a unique solution for every customer, we found that we can adapt the software and development tools created for one application to an adjacent one. This will allow us to expand into new edge AI applications more quickly and efficiently. We've already demonstrated this, excuse me.

We've already demonstrated this by applying many of the capabilities developed for the Sentinel POC to our Smart City project, and we expect to build on that foundation as we pursue new opportunities in other edge AI applications. During the first quarter, we made successful, meaningful progress on our projects, POCs, and SBIRs. Each of these position Gemini-II for broader commercial adoption. I'll begin with Sentinel, which is a good example of our customer adoption strategy. GSI has successfully completed all of its deliverables for the first phase, including providing both the chip and the software. A laboratory test of the Sentinel POC was recently conducted in collaboration with G2 Tech and the participating defense agencies. Sentinel is an application where Gemini-II's performance per watt provide a clear advantage, and the platform's results in the lab test confirm this.

With a time to first token, also known as TTFT, of 2.4 seconds. Preparation for the field demonstration continues under G2 Tech's leadership. We have completed all of GSI's deliverables for this next part of the POC as well. The timing of the field demonstration is now dependent on G2 Tech. Turning to our Smart City project in Hsinchu County, Taiwan, we continue to meet our phase 1 deliverables, which means we will soon be shipping the Gemini-II chips, servers, and software supporting the rollout for 20 cameras. We remain on track to complete the phase 1 deployment in November. If selected for phase 2, the project would expand to approximately 80 cameras and would add audio to the video from phase 1. This next phase would provide both additional hardware and software licensing revenue.

The Smart City platform now includes a smart query capability that allows users to search surveillance footage using natural language. For example, an operator can simply ask, "Did you see a person matching this description yesterday?" The system can quickly analyze recorded video, identify relevant events, and locate potential matches without requiring manual review of hours of footage. Switching to our government-funded SBIRs, we're pleased to have launched our U.S. Army phase 2 program with a formal kickoff meeting. We have begun preliminary work on developing a ruggedized Gemini-II system for Army applications. In another SBIR for the Space Development Agency, we successfully completed radiation testing on a standard commercial Gemini-II device with no device optimization for withstanding radiation. In the test, the off-the-shelf chip continued to be operational under radiation levels found in harsh space environments.

The testing achieved the radiation tolerant threshold typically required for radiation-hardened application and showed no single event latch-ups, also known as SELs. The absence of SELs is important because that reduces technical risk for companies evaluating the device and shows that our standard chip may be suitable for some space and high-altitude applications. Before I conclude, I'd like to briefly update you on an important software milestone this fiscal year. In the past, you've heard us mention plans to release an AI compiler. We have expanded on that vision and are developing an AI-assisted SDK, as Lene mentioned. This is expected to significantly accelerate AI model development and the deployment on Gemini-II. The alpha release is currently targeted for this fall. One big advantage found in early testing shows the AI-assisted SDK significantly reduces application development time.

In one representative case, adapting an AI model to run on Gemini-II previously took one engineer about a year. With the AI-assisted SDK, that work was completed in less than a month. This will help us support a broader range of customer applications while enabling us to work with more partners and system integrators. The AI-assisted SDK will also accelerate software development for Plato by allowing algorithms and models to be ported more quickly, allowing us to bring final applications to the market sooner. To summarize, we're pleased with the progress made in the first quarter on our POCs and SBIRs. Our focus for the remainder of this fiscal year remains on executing our Gemini-II strategy, specifically completing our current proof of concept projects, expanding into adjacent applications, and completing the software tools needed to support broader commercialization.

Together, these initiatives will make Gemini-II easier to deploy, enabling us to support more customers and over time, scale our business through system integrators. In parallel, we are heads down on the Plato design and remain on track to tape-out in early calendar 2027. With that, I'd like to hand the call over to Doug.

Go ahead, Doug. In the earnings release we issued today after the market close, you'll find a detailed summary of our first quarter fiscal 2027 financial results.

In my comments to follow, I'll focus on the key drivers behind our performance and provide some additional context around the quarter. Revenue in the first quarter of fiscal year 2027 of $6.3 million was at the midpoint of our guidance range, in line with the prior quarter, and up modestly from the first quarter of fiscal year 2026. Overall, our SRAM revenues remain stable. As is typical in our business, customer demand can vary from quarter to quarter, which can impact both our revenue and gross margin.

In the first quarter of fiscal 2027, combined sales to KYEC and Cadence represented approximately 23% of net revenues, compared with the prior quarter, in which KYEC represented 22.3% of net revenues and there were no shipments to Cadence. We expect continued shipments to customers supporting AI chip design and semiconductor manufacturing. However, given the nature of our customer base, order patterns may continue to vary from quarter to quarter. Gross margin for the quarter was 53.4%, almost slightly below our guided range of 54%-56%, gross margin improved 100 basis points sequentially. Compared to the year ago quarter, gross margin was lower, primarily due to product mix, which could fluctuate based on the timing and composition of customer shipments. Operating expenses for the quarter rose to $8.8 million compared to $5.8 million in the prior year period.