GSI Technology, Inc. (GSIT) Earnings

Management highlights

### Strategic Transition - The company has completed its multi-year transition from a standalone memory chip company to an edge AI platform company centered on its custom APU (AI Processing Unit) product line. - The APU fills a gap in the edge AI market: it delivers lower power consumption than traditional GPUs and higher performance than standard mobile processors for real-time workloads, with the production Gemini 2 chip now available for customer engagements. The company's current focus is shifting from technology development to commercializing the APU business. ### Key Proof of Concept (POC) Progress - **Sentinel (Defense Drone POC):** All first-phase deliverables (chip, software) have been completed, with successful lab testing conducted in partnership with G2 Tech and defense agencies. The test recorded a time to first token (TTFT) of 2.4 seconds, confirming Gemini 2's performance-per-watt advantage. All GSI deliverables for the upcoming field demonstration phase are also complete, with timing now dependent on G2 Tech. - **Shenzhou County, Taiwan Smart City POC:** Phase 1 deliverables are on track, with Gemini 2 chips, servers, and software for 20 cameras set to ship soon. Phase 1 deployment is expected to be completed in November 2026. If selected for Phase 2, the project will expand to 80 cameras and add audio processing, generating additional hardware and software licensing revenue. The platform now supports natural language smart query for surveillance footage. - **Government-Funded SBIR Programs:** A U.S. Army Phase II program has been formally launched, with preliminary work underway on a ruggedized Gemini 2 system for Army use. A Space Development Agency SBIR completed successful radiation testing on an off-the-shelf (unmodified) Gemini 2 chip; the device remained operational at radiation levels common for harsh space environments, met typical radiation-tolerant thresholds, and experienced no single-event latchups (SELs), making it potentially suitable for space and high-altitude applications. ### Product Development Milestones - The company is developing an AI-assisted SDK (expanded from the originally planned AI compiler) targeted for alpha release in fall 2026. Early testing shows the SDK drastically reduces model porting time: from one engineer-year per adaptation to less than one month, enabling faster expansion into more customer applications and supporting faster development for the upcoming Plato next-generation APU. - Plato development remains on schedule for tape-out in early Q1 calendar 2027 (March 2027). - The company's near-term expansion strategy leverages existing work across adjacent applications to reduce redundant engineering, rather than building custom solutions for every new customer, allowing faster and more efficient market expansion. For example, capabilities developed for the Sentinel POC were reused for the Smart City project.

Guidance

- For Q2 FY2027, management guided net revenue in the range of $5.7 million to $6.5 million, with gross margin expected between 53% and 55%. - The company expects quarterly cash usage to remain at approximately $4 million (around $16 million annually) through Gemini 2 software development and Plato tape-out, with timing varying based on development milestones. - Management confirmed the existing $77 million cash position (with no debt) provides sufficient funding to complete all current strategic priorities including Gemini 2 commercialization and Plato tape-out. - The AI-assisted SDK alpha release is still targeted for fall 2026, and Plato tape-out remains on schedule for March 2027, with no revisions to these timelines. - The company plans to file a replacement for its universal shelf registration to restore its $100 million capacity, but has no current plans to raise additional capital; maintaining the shelf is for strategic and financial flexibility only.

Segment performance

GSI Technology reports only aggregated revenue and does not break out performance by individual product segments in this transcript. Aggregate Q1 FY2027 net revenue was $6.3 million, which was at the midpoint of the company's prior guidance range. Revenue was modestly lower than the year-ago Q1 FY2026. SRAM revenues remained stable overall. Combined sales to KYEC and Cadence represented 23% of net revenues for the quarter, compared to 22.2% for KYEC alone in the prior quarter (no Cadence shipments in the prior quarter). Gross margin for the quarter was 53.4%, which was slightly below the guided 54-56% range, but improved 100 basis points sequentially from the prior quarter. Gross margin was lower year-over-year primarily due to product mix fluctuations. Operating expenses were $8.8 million, up from $5.8 million in the year-ago quarter, with research and development expenses of $5.9 million (up $2.8 million year-over-year). Operating loss was $5.4 million, up from $2.2 million in the year-ago quarter, but remained flat sequentially. As of quarter-end June 30, 2026, the company held $77 million in cash and cash equivalents with no debt.

Risks & headwinds

- Customer order patterns for SRAM products vary quarter-to-quarter, which creates volatility for quarterly revenue and gross margin due to shifting product mix. - SBIR funding fluctuates quarter-to-quarter based on project milestone timing, which creates variability in reported net R&D expenses. - Software development costs for the company's Israeli-based team are exposed to foreign exchange volatility between the Israeli shekel and U.S. dollar, which can increase operating expenses when the shekel strengthens. - Commercial revenue generation from the new APU business is dependent on successful completion of current POC projects, securing design wins, and attracting system integrator partners, all of which carry execution risk.

Analyst Q&A

• Q: How transferable is technical work from existing POC programs to new applications, and can the POC pipeline expand without significant increases to engineering headcount? /

  A: Technical work from existing POCs is highly transferable across adjacent edge AI applications. Features like performance-optimized vision language model capabilities and fast time-to-first-token developed for the Sentinel drone POC were directly reused for the Taiwan Smart City project, avoiding full redevelopment from scratch. Combined with the upcoming AI-assisted SDK launching this fall, which drastically reduces model porting time, the company will be able to expand into new application areas much faster without requiring proportional increases to engineering headcount.

• Q: How mature do the SDK and application framework need to be before system integrators can adopt Gemini 2, and what is the timeline for formalizing these partnerships? /

  A: The company's current strategy is to complete initial POCs in new markets internally first, to prove that Gemini 2 delivers strong value for that use case. Once use cases are proven, system integrators will be able to take over customized deployments for individual end customers, as GSI does not have enough internal staff to support all customized projects across multiple markets. The AI-assisted SDK launching this fall is a critical prerequisite to enable system integrator adoption, as it will allow integrators to quickly adapt the platform to their specific customer requirements.

• Q: What are the implications of the successful off-the-shelf Gemini 2 radiation testing for the aerospace and defense sector, and can this open a new revenue stream for the company? /

  A: Successful radiation testing opens a significant new potential revenue stream in the space and aerospace defense market, which is currently seeking low-power AI-capable processors for new satellite and space vehicle programs. Traditional GPUs do not perform well in high-radiation space environments, and power constraints are extremely tight for space applications, which aligns perfectly with Gemini 2's core low-power advantage. Next-generation Plato will extend this advantage with a target power draw of less than 10 watts per device. The company has already begun discussions with SBIR partners about opportunities in this space, and the radiation testing success has already enabled program extensions to further explore this market.