GSI Technology

What GSI Technology does

GSI Technology, Inc. (NASDAQ: GSIT) is a Sunnyvale, California-based semiconductor company founded in 1995 that operates two complementary product franchises. The legacy franchise consists of high-performance SRAM and Low Latency DRAM (LLDRAM) memory products, including SigmaQuad, SigmaDDR, SyncBurst, NBT, and Asynchronous SRAM families at densities from 1Mb to 288Mb, available in radiation-hardened variants for aerospace and defense applications. The newer franchise is the Gemini Associative Processing Unit (APU) family, a compute-in-memory architecture that performs computation directly within SRAM arrays to deliver GPU-class AI performance at significantly lower energy consumption than conventional accelerators. The company holds over 120 granted patents covering associative computing and memory technologies.

The Gemini APU targets AI inference, vector search, retrieval-augmented generation (RAG), molecular similarity search, Synthetic Aperture Radar (SAR) processing, and edge AI workloads, with performance validated by Cornell University at the MICRO 2025 conference showing 54-98% lower energy consumption than NVIDIA A6000 GPUs for RAG workloads. Product form factors include Leda-E 2U servers, Leda-S 1U edge servers, PCIe cards, and a SaaS/hosted service offering through searchium.ai. The company complements hardware with software including the Smart Eye Guardian video analytics platform, the Sentinel autonomous security system (co-developed with G2 Tech), an OpenSearch neural search plugin, BIOVIA Pipeline Pilot integration, and the Copperhead SDK. Software R&D is centered in Tel Aviv.

GSI distributes through regional sales offices in the US (Sunnyvale, Austin), Israel (Natanya), and Hong Kong, plus a global network of North American and international distributors spanning 50+ countries. Customers include U.S. defense agencies (Air Force Research Laboratory, Army xTech SBIR, Department of War), academic institutions (Cornell University, Weizmann Institute), and Taiwanese government (Hsinchu County smart city deployment). Reported FY2026 revenue was approximately $25.1 million (22.4% YoY growth), with the SRAM business growing ~22%; commercial APU revenue is not expected until 2027, with current APU-related revenue primarily sourced from government SBIR contracts and proof-of-concept engagements.

Name

GSI Technology

Legal name

GSI Technology, Inc.

Website

https://gsitechnology.com

Company type

Public

Founded year

1995

Operating status

Operating

Headcount range

101–250 employees

Short description

GSI Technology (NASDAQ: GSIT) designs and sells high-performance SRAM memory and Gemini Associative Processing Units (APUs), a compute-in-memory architecture for AI inference and vector search, serving defense, aerospace, networking, and edge AI customers globally.

Ownership category

akta.pro rank

Product category

Semiconductor Memory and Compute-in-Memory AI Accelerators

NAICS

Semiconductor and Related Device Manufacturing (334413), Computer Storage Device Manufacturing (334112)

SIC

Semiconductors & Related Devices (3674)

akta.pro primary industry

Compute & Acceleration ICs (CPU/GPU/AI/FPGA/ASIC) (HDAHABAC)

akta.pro secondary industries

Edge AI Hardware & NPUs (SoCs, modules, accelerators) (HDAAAJAC), AI Memory & Storage for Training (HBM/DDR/SSD/Object Storage) (HDAAAAAF)

Keywords

• High-performance SRAM memory
• Compute-in-memory AI accelerators
• Associative processing units
• Radiation-hardened SRAM
• Low latency DRAM

Headquarters

HQ city

Sunnyvale

HQ country

United States

HQ region

North America

Offices5 records

Markets served

GTM type

B2B

Offering type

Hardware or Manufacturing

Cost components

Technology or R&D, Personnel, Supply Chain, Marketing or Sales, Operations

Revenue model

1. SRAM Memory Products: Sales of high-performance SRAM products including SigmaQuad, SigmaDDR, SyncBurst, NBT, and Asynchronous SRAMs, along with Low Latency DRAMs (LLDRAMs). These are sold directly and through distributors to networking, telecom, aerospace, defense, industrial, and medical customers. Revenue grew 22.4% year-over-year in FY2026 to approximately $25.1 million, driven by strong SRAM sales to chip design and simulation customers.
2. APU (Associative Processing Unit) Products: Sales of Gemini APU-based systems including Leda-E servers, Leda-S servers, and PCIe cards for AI inference, vector search, molecular search, SAR processing, and defense applications. The company does not expect commercial APU revenues until 2027, though government-funded proof-of-concept engagements and SBIR contracts provide near-term revenue.
3. SaaS / Hosted Services: Hosted vector search and neural search services delivered through the company's SaaS platform (searchium.ai), offering usage-based hourly pricing for APU resources. Also offered as an on-premises plugin solution.
4. Government R&D Contracts: Funding from U.S. government SBIR programs including a $1.1M contract from U.S. Air Force Research Laboratory, a $2M Phase II Army xTech SBIR contract, and approximately $1M in government funding for the Sentinel autonomous perimeter security POC with G2 Tech.

Pricing tiers

Go-to-market motion4 records

Distribution channels5 records

Marketing channels6 records

Core offering

GSI Technology designs and sells two complementary semiconductor product lines. The first is a portfolio of high-performance SRAM and Low Latency DRAM (LLDRAM) memory products (SigmaQuad, SigmaDDR, NBT, SyncBurst, Asynchronous SRAMs, and radiation-hardened variants) used in networking, aerospace, defense, and industrial equipment. The second is its Associative Processing Unit (APU) compute-in-memory AI accelerators (Gemini-I, Gemini-II), packaged into Leda-E/Leda-S servers, PCIe cards, and hosted SaaS services, that deliver GPU-class AI inference and vector-search performance at a fraction of the energy consumption.

Product overview

GSI Technology offers a dual product portfolio: (1) Gemini Associative Processing Units (APU) - compute-in-memory AI accelerators comprising Gemini-I and Gemini-II chips, deployed in Leda-E/Leda-S servers for edge AI applications including vector search, smart city video analytics, and defense systems; and (2) High-performance SRAM memory solutions including SigmaQuad, SigmaDDR, NBT, SyncBurst, Asynchronous, and LLDRAM product families at 1Mb-288Mb densities, available in radiation-hardened and radiation-tolerant variants for aerospace, defense, networking, and industrial applications. The APU products are complemented by software platforms including Smart Eye Guardian for video analytics, Sentinel for autonomous security, and SaaS offerings, supported by development tools including Copperhead SDK and Metaware Compiler.

Differentiator

Problem solved

Functional benefit

Brands

• Gemini APU: Associative Processing Unit technology for AI and compute-in-memory applications
• SigmaQuad SRAM
• SigmaDDR SRAM
• SyncBurst SRAM
• NBT SRAM
• Leda
• Smart Eye Guardian

Products and services

• Gemini-I Associative Processing Unit (APU) First-generation associative processing unit featuring compute-in-memory architecture for AI inference workloads. Delivers GPU-class performance with significantly lower energy consumption for retrieval-augmented generation and similarity search applications. Targeted at enterprise and research customers.
• Gemini-II Associative Processing Unit (APU) Next-generation APU targeting edge AI applications with 3-second time-to-first-token performance for multimodal LLM inference at approximately 30 watts. Powers Smart Eye Guardian video analytics, Sentinel autonomous perimeter security, and ruggedized tactical edge AI deployments.
• Leda-E Server High-performance 2U server built around Leda-E boards for enterprise-scale vector search, molecular search, facial recognition, and high-performance computing. Available in dual-board and 8-board configurations, used by enterprise customers.
• Leda-S Server Compact 1U APU server with 16 Leda-S boards designed for edge deployment in space-constrained, high-density AI installations such as telecom sites and smart-city edge nodes.
• Smart Eye Guardian AI video analytics platform powered by Gemini-II APU that correlates inputs across cameras, sensors, and infrastructure monitoring to deliver real-time situational awareness for smart-city and public-safety deployments.
• Sentinel Autonomous Perimeter Security Autonomous perimeter security system jointly developed with G2 Tech, integrating GSI's Gemini-II APU for real-time management of drones and cameras in mission-critical tactical environments.
• Searchium.ai Hosted Vector Search SaaS Hosted SaaS service delivering APU-powered vector and neural search with flexible usage-based hourly pricing. Beta trials offered to developers and smaller customers through the Searchium.ai platform.
• SAR Fast Back Projection (FBP) Software Synthetic Aperture Radar processing software that runs on GSI's APU and applies each input pulse simultaneously to construct all pixels in parallel, enabling real-time SAR image generation in aircraft and satellites where datacenter resources are unavailable.
• SigmaQuad SRAM High-performance synchronous SRAM with simultaneous read/write dual quad data rate interface, offered in 18Mb to 288Mb densities and in radiation-hardened and radiation-tolerant variants for space, military, and commercial applications.
• SigmaDDR SRAM Double data rate SRAM for applications with infrequent read/write alternation, operating at up to 833 MHz in 18Mb to 288Mb densities for networking and industrial customers.
• NBT (No Bus Turnaround) SRAM Synchronous burst SRAM with no-bus-turnaround functionality for zero-wait read-write-read bus utilization, offered in 4Mb to 288Mb densities for networking, industrial, automotive, and medical-imaging customers.
• SyncBurst SRAM Synchronous burst SRAM that delivers 2-4 word bursts in response to a single clock signal, used in networking, industrial, automotive, and medical-imaging equipment.
• Asynchronous SRAM Easy-to-use asynchronous SRAMs with fast 7-12 ns access times, suitable as main memory for small cache-less embedded processors in industrial electronics and cost-sensitive networking equipment.
• LLDRAM (Low Latency DRAM) Low-latency DRAM for advanced networking with near-100% bus utilization, ideal for 10GbE, 40GbE, and 100GbE packet buffering and inspection, offered in 288Mb to 576Mb densities.
• OpenSearch Neural Search Plugin (Fast Vector Search) Production-grade neural-search plugin for OpenSearch that searches billions of items in milliseconds with high recall, enabling neural search without migrating to a new vector database. Customers include Muves Ltd for multilingual and multimodal search.
• Radiation-Hardened / Radiation-Tolerant SRAMs High-performance SRAMs offered in radiation-hardened and radiation-tolerant formats for space and military environments, including SigmaQuad-II+ and other SRAM families with densities from 1Mb to 288Mb and operating frequencies up to 1333 MHz.

Quantifiable outcome

• 54-98% lower energy consumption vs. NVIDIA A6000 GPU for RAG AI workloads (Cornell University validation, MICRO 2025)
• +5 more outcomes

Named customers7 records

Segments4 records

Ideal customer profiles5 records

Technology focussed Yes

API detail

Has API

No

API docs

API detail

Core technology

AI maturity

App detail

Integration4 records

AI capability12 records

Feature7 records

Partnerships

Four partnerships are on record, tiered core and minor.

• Hsinchu County Government (Taiwan)coreStrategic or Co-development Partner · 18 May 2026GSI Technology was awarded Phase I of a Smart City project by the Hsinchu County government in Taiwan, marking the first deployment of its Gemini-II APU-powered Smart Eye Guardian AI video analytics platform. Phase I runs for approximately six months, developing an open-architecture software platform and deploying a pilot system with about 20 cameras. Follow-on phases could expand to 80 cameras with multi-million-dollar revenue potential.
• G2 TechcoreStrategic or Co-development Partner · 14 January 2026GSI Technology partnered with Israel-based G2 Tech to develop Sentinel, an autonomous perimeter security system for drones and cameras, jointly funded by the U.S. Department of War and a foreign government agency. GSI expects to receive approximately $1 million in government funding to support software optimization and integration of its Gemini-II APU into the Sentinel system for real-time edge AI processing in mission-critical tactical environments.
• Cornell UniversitycoreStrategic or Co-development Partner · 19 October 2025Cornell University researchers validated GSI's Gemini-I APU in a paper presented at MICRO 2025, demonstrating GPU-class AI performance with 54-98% lower energy consumption compared to NVIDIA A6000 GPU for RAG workloads. The academic validation has been a key marketing milestone for GSI's compute-in-memory technology.
• Weizmann Institute of ScienceminorStrategic or Co-development PartnerWeizmann Institute of Science used GSI's Gemini APU for molecular search applications, reducing search times from minutes to seconds for drug discovery. The collaboration validates APU capabilities in pharma/cheminformatics.

Scale indicators12 records

Recent moves9 records

Expansion highlights6 records

Direct peers

• Mythic AI: Compute-in-memory AI inference chip company using analog in-memory computing for edge AI. Closest architectural peer to GSI's APU, targeting similar edge AI workloads with low-power neural network inference. Both pursue the compute-in-memory paradigm for energy-efficient AI.
• Groq: AI inference accelerator company providing low-latency, high-throughput inference hardware. Competes with GSI's APU in the AI inference market with a different architectural approach (LPU vs. compute-in-memory) but targeting similar use cases including LLM inference and vector search.
• Hailo: Edge AI processor company designing low-power AI accelerators for computer vision and edge inference. Targets similar edge AI applications in smart cities, automotive, and industrial IoT that GSI's Gemini-II APU is pursuing, with emphasis on low power consumption.
• Syntiant: Edge AI company producing ultra-low-power neural decision processors for always-on audio and sensor applications. Competes in the edge AI inference market targeting battery-powered devices where GSI's compute-in-memory also pursues differentiation on energy efficiency.
• Integrated Silicon Solution Inc (ISSI): Fabless semiconductor company specializing in SRAM, DRAM, and NOR flash memories for automotive, industrial, and networking applications. Direct competitor to GSI's SRAM business with similar niche focus on high-performance and specialty memory products.

Broad incumbents

• Renesas Electronics: Major semiconductor manufacturer with broad SRAM and memory product portfolio serving networking, industrial, and automotive markets. Competes with GSI's legacy SRAM business as a much larger incumbent, though Renesas' portfolio spans MCUs and SoCs beyond memory.
• NVIDIA: Dominant AI compute platform provider whose GPUs are the benchmark against which GSI positions the APU's energy efficiency. Not a direct product competitor at the edge AI inference level but defines the competitive landscape for AI compute, including edge offerings through Jetson.
• AMD: Major semiconductor company with broad AI accelerator portfolio including Instinct GPUs, Ryzen AI, and Xilinx FPGAs. GSI explicitly integrates with AMD/Xilinx UltraScale/UltraScale+/Versal FPGAs via IP ports. Represents both a partner and broad incumbent competing for AI compute market share.
• Intel Corporation: Semiconductor incumbent with broad AI accelerator portfolio (Intel Gaudi, Movidius VPU for edge AI) and SRAM heritage. Competes broadly in both GSI's legacy SRAM niche and the emerging edge AI accelerator market, with massive R&D scale advantage.

Emerging players

• Cerebras Systems: AI accelerator company providing wafer-scale compute systems for training and inference. While targeting datacenter-scale AI rather than GSI's edge focus, both pursue novel compute architectures to break memory-processor bottlenecks — Cerebras via wafer-scale integration, GSI via compute-in-memory.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat5 records

Key risks5 records

Key highlights6 records

Customer concentration

• Xx.com/gsitechnology
• LinkedInlinkedin.com/company/gsi-technology

Compliance5 records

Revenue estimate

Valuation estimate

Number of profiles

Profiles8 records

Funding overview

Funding rounds2 records

Investors1 record

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about GSI Technology