GSI Technology, Inc.

Welcome to GSI Technologies' third quarter fiscal 2026 results conference call. At this time, all participants are in listen-only mode. Later, we will conduct a question and answer session. At that time, we will provide instructions for those interested in entering the queue for the Q&A. Before we begin today's call, the company has requested that I read the following Safe Harbor statement. The matters discussed in this conference 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. Leland Hsu, the company's chairman, president, and chief executive officer will be hosting the call today. With him are Douglas Shirley, chief financial officer, and Didier Lassere, vice president of sales. I would now like to turn the conference over to Mr. Hsu. Please go ahead, sir.

Good afternoon, and thank you for joining us to review our third quarter physical 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 physical year-to-day basis. Demand for our excellent products remains solid, and we expect strong sales from our largest customers in the first half of calendar 2026. After completing our financing in fall 2025, we will defend our APU roadmap. We began Plato hardware development after purchasing the required IP. We have also added contract engineers to support our hardware design team, keeping us on track to take out Plato in early 2027. We finalized an agreement with G2 Tech, an Israel-based AI company, for our recently announced proof of concept. We are partnering with G2 Tech on Sentinel, a program for autonomous perimeter security using drones and cameras. The project is backed by the U.S. Department of War and a foreign government agency. This government funding will offset all costs to build the software stack and the libraries needed for this project. Delia will share more details shortly. Another recent milestone is Gemini 2's time-to-first token benchmark, announced in the press release earlier today. For those who have not reviewed it, please see today's release for the full detail on the benchmark results. and the methodology. Accordingly, we reported three-second time-to-first token, or TTFT, performance for AOLN with text and video input, consuming approximately 30 watts of system power. Compared to third-party testing of competitive platforms, Gemini 2's TTFT delivered up to three times faster first token at a lower power than the competitive chip on the same workload. We believe these test results validate Gemini 2's fast response for age use cases that need low power and a low latency. In his comments, EDL will depend on the benchmark results. We are making steady progress for the benchmark threshold continue to improve Gemini 2 performance, and completing the Sentinel project. We are also pursuing early proof-of-concept and prototyping opportunities for Gemini 2 in systems in the defense programs, including drones and unmanned systems, and in selected commercial light deployments. In parallel, we continue to pursue long-diluted R&D funding through government defense programs and strategic partners. With that, I send the call over to Li-Ling.

Thank you, Li-Ling. I'll start by expanding on some of Li-Ling's 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 project, including software development for Gemma 3-12B on our Gemini 2, 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 2'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 2 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 2 TTFT benchmarks we discussed preliminary results showing a three-second time-to-first token for a multimodal model at the edge using video and text inputs at approximately 30 watts 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 three seconds. That means we are sampling the video image every three 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 five months to further reduce Gemini 2's first time to response. What's exciting for GSI about these Gemini 2 preliminary benchmark results is that they demonstrate what computer 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 in edge AI and physical AI is real-world 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 2. 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. Sigma quad 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.