GSI Technology, Inc.

What they do

GSI designs and sells high-speed synchronous SRAM products and is developing a proprietary associative processing unit (APU) for compute-in-memory processing at the edge. It operates fabless, outsourcing wafer fabrication, assembly and testing, with headquarters in Sunnyvale, California and additional operations in Taiwan and Israel. Revenue today comes from SRAM sales to OEM customers including KYEC, Cadence Design Systems and Nokia, plus military/defense and aerospace customers; APU revenue has not been material to date.

Revenue drivers

• Synchronous SRAM — The established legacy business and essentially all current revenue, sold into networking and telecommunications, test and measurement, and military/defense and aerospace. Fiscal 2026 net revenue of $25.1 million rose 22% year over year on strong SRAM sales to chip design and simulation customers.
• Military/defense and aerospace SRAM — Includes radiation-tolerant and radiation-hardened space-grade SRAMs. Military/defense was 31.3% of first-quarter fiscal 2027 shipments, versus 19.1% a year earlier and 45.7% in the prior quarter.
• APU / edge AI products — Gemini APU and planned Plato chip target low-power edge AI, synthetic aperture radar, vector search and computer vision. The company states revenue from the APU line has not been material to date.
• U.S. government SBIR contracts — Four DoD SBIR contracts, including a $2.0 million Space Development Agency award and a $1.1 million AFWERX/AFRL award. Aggregate payments were approximately $557,000 in fiscal 2025 and $1.6 million in fiscal 2026, and have not yet produced follow-on production contracts.

Recent performance

First-quarter fiscal 2027 revenue was $6.3 million, flat versus both the year-ago quarter and the prior quarter and at the mid-point of guidance. Gross margin was 53.4%, down from 58.1% a year ago and up from 52.4% in the prior quarter, with the year-over-year decline attributed to product mix. Operating expenses rose to $8.8 million from $5.8 million a year ago, as R&D increased to $5.9 million from $3.1 million on Plato chip design. The net loss widened to $4.8 million, or $0.13 per diluted share, from $2.2 million, or $0.08, a year earlier. Quarter-end cash was $77.0 million, up from $67.2 million at March 31, 2026, aided by ATM proceeds and employee option exercises.

Strategy

The company is pursuing a two-pronged strategy: commercialize the APU for edge AI while the SRAM business funds development. Current priorities are completing proof-of-concept programs (Sentinel laboratory phase completed, Smart City Phase I deployment on schedule for November 2026), expanding Gemini-II into adjacent applications, and releasing an AI-assisted SDK in September. Plato development remains on schedule for a March 2027 tape-out and is intended to enable larger language models at the edge with power consumption below 10 watts. The company also completed radiation testing on a standard commercial Gemini-II device, which it says operated normally under radiation levels representative of harsh aerospace environments.

Risks

• Customer concentration — KYEC, Nokia and Cadence Design Systems account for a significant percentage of net revenues, and quarterly shipments to them fluctuate sharply — Cadence was 23.9% of revenue a year ago and 7.4% this quarter.
• Persistent losses and cash use — The company has incurred significant annual net losses every year from fiscal 2022 ($16.4 million) through fiscal 2026 ($13.2 million), with operating cash flow negative in each of those years.
• Dependence on legacy SRAM and unproven APU — The company depends on Very Fast SRAM for most revenue while transitioning to in-place associative computing; a demand downturn or failure to hit APU revenue goals could cause cash shortfalls.
• U.S. government funding exposure — The business relies on U.S. government funding for SBIR programs, and appropriations lapses or shutdowns could materially affect operations.

Outlook

For the second quarter of fiscal 2027, management guides net revenues of $5.7 million to $6.5 million with gross margin of approximately 53% to 55%. Management says its priorities are completing current proof-of-concept projects, expanding Gemini-II into adjacent applications, and delivering the AI-assisted SDK in September. Plato remains on schedule for tape-out in March 2027, which management expects to significantly expand the addressable market by enabling larger language models at the edge with power consumption below 10 watts.