GSI Technology, Inc.
Item 1. Business
Overview
GSI Technology, Inc. (“GSI” or the “Company”) is a semiconductor company pursuing a two-pronged business strategy. Our growth strategy centers on the commercialization of our proprietary associative processing unit (“APU”) technology, which enables high-performance, low-power, compute-in-memory processing for artificial intelligence, high-performance computing, and search applications at the edge. We fund this development through our established legacy business designing and selling high-speed synchronous static random access memory (“SRAM”) products, primarily for the networking and telecommunications, test and measurement, and military/defense and aerospace markets. We operate under a fabless manufacturing model and are headquartered in Sunnyvale, California, with additional operations in Taiwan and Israel.
Our APU family of products delivers in-place associative computing capabilities in a compact, low-power form factor, making them well suited for the expanding market for physical artificial intelligence (“AI”) at the edge. The APU has demonstrated industry-leading time-to-first-token performance in multi-modal vision language models (“VLMs”), a capability that is critical for real-time situational awareness in physical AI applications. We believe this performance advantage, combined with our proven low-power architecture, positions the APU for broad applicability as edge AI adoption accelerates. In addition, we have demonstrated APU use cases in synthetic aperture radar (“SAR”) image processing, fast vector search of very large databases, computer vision, drug
3
discovery, and cybersecurity. Together, these capabilities enable multi-workload processing within size, weight, and power constrained environments. Revenue from our APU line has not been material to date.
We remain committed to our established synchronous SRAM business, which generates revenue that supports our APU development. We offer what we believe is the broadest portfolio of high-density, high-performance synchronous SRAM products in the market. These products are used in test and measurement equipment, high-performance networking and telecommunications infrastructure, and military/defense and aerospace applications. We maintain long-term relationships with leading original equipment manufacturer (“OEM”) customers, including KYEC, Cadence Design Systems, and Nokia. In addition, we serve the military/defense and aerospace markets with radiation-tolerant and radiation-hardened space-grade SRAMs, as well as APU-based solutions for applications such as SAR image processing.
We operate under a fabless business model, outsourcing wafer fabrication, assembly, and testing. This model allows us to focus our resources on research and development, product design, and marketing while gaining access to advanced process technologies with modest capital investment and fixed costs.
GSI’s fiscal year 2026 net revenue increased by 22% compared to net revenue in fiscal year 2025, reflecting strong SRAM sales to chip design and simulation customers. GSI’s gross margin increased by 5% compared to the prior fiscal year primarily reflecting a favorable mix weighted toward higher-margin SRAM products.
We have been awarded four contracts under the U.S. Department of Defense Small Business Innovation Research (“SBIR”) program to develop and demonstrate applications of our APU compute-in-memory architecture for military and space customers. All milestones under these contracts were completed as of March 31, 2026, except for certain amended milestones under the Space Development Agency agreement described below and the US Army xTECH agreement described below. Aggregate payments received under these contracts totaled approximately $557,000 and $1.6 million in fiscal 2025 and 2026, respectively. While these contracts validated the performance of our APU architecture in defense applications, they have not yet resulted in follow-on production contracts or material commercial revenues from the underlying technology.
Space Development Agency — SBIR Direct to Phase II (Prototype Agreement) — $2.0 million.
Development of next-generation APU2 compute-in-memory integrated circuit for space-based edge processing, including radiation-hardened capability assessment. The agreement was originally valued at $1.25 million and was amended in September 2025 to increase the total award to $2.0 million. All original milestones were completed as of March 31, 2026. Milestone payments of $435,000, $318,000, and $496,000 were received in fiscal 2024, 2025, and 2026, respectively.
AFWERX / U.S. Air Force Research Laboratory (“AFRL”) — SBIR Direct to Phase II — $1.1 million.
Development and demonstration of high-data computation use cases using the Gemini APU for AFRL, including in-aircraft search and rescue, object detection, moving target indication, change detection, and GPS-denied navigation. All milestones were completed as of March 31, 2026. Milestone payments of $157,000 and $983,000 were received in fiscal 2025 and 2026, respectively.
U.S. Army (DoD SBIR) — SBIR Phase II — Up to $250,000.
Development of edge computing AI solutions using Gemini-II, including feasibility assessment of integrating 1-bit Large Language Models for low-power, low-latency military applications. All milestones were completed as of March 31, 2026. Milestone payments of $82,000 and $165,000 were received in fiscal 2025 and 2026, respectively.
4
U.S. Army — xTech SBIR Phase II — $2.0 million.
Development of a ruggedized edge-processing platform based on the Gemini-II APU, with testing in representative operational environments intended to validate performance across real-time AI workloads, such as sensor data processing, object detection, and command-and-control analytics. This SBIR was awarded to GSI in April 2026.
In addition to the four SBIR’s discussed above, in January 2026, we announced a new proof-of-concept (“POC”) engagement with two government agencies. GSI is partnering with G2 Tech, an Israel Deep tech AI company, on Sentinel, a program to develop an autonomous perimeter security system that manages drones and cameras in real time for advanced monitoring, detection, and response. The project is jointly backed by the U.S. Department of War (“DoW”), formerly known as the Department of Defense (“DoD”), and a foreign government agency.
In May 2026, we announced that we were awarded Phase I of a Smart City project by a local government agency in Taiwan. The completion of Phase I will mark our first smart city deployment of the Gemini-II APU.
Our APU technology is implemented in the Gemini series of AI chips. The Gemini-I part is in full production, and the Gemini-II device is in pre-production and already in the market with sales delivery of PCIe boards and chips. There have not been substantial sales of Gemini-I parts to date. We support customers with prebuilt application program interfaces (“APIs”) and libraries to support the parallel processing capabilities of the Gemini-I and Gemini-II parts. The software stack accelerates development by providing an integrated framework environment for the compute-in-memory as well as host and management code modules. Our compiler stack framework allows customers to optimize their applications by editing APIs provided by GSI, or write their own APIs. Benchmarking on the Gemini-II has shown an industry leading capability on time-to-first-token which is being used to provide situational awareness for physical edge AI products.
In March 2025, we secured an initial production order for our radiation-hardened SRAM from a North American prime contractor, with follow-on orders expected in fiscal 2027. This sale carries a significantly higher gross margin than our traditional SRAM chips. In parallel, we are actively pursuing heritage status for this chip, which will improve our market readiness and open important new sales channels.
We were incorporated in California in 1995 under the name Giga Semiconductor, Inc. We changed our name to GSI Technology in December 2003 and reincorporated in Delaware in June 2004 under the name GSI Technology, Inc. Our principal executive offices are located at 1213 Elko Drive, Sunnyvale, California, 94089, and our telephone number is (408) 331-8800.
5
Recent Developments
On October 21, 2025, we entered into a securities purchase agreement (the “Purchase Agreement”) with an investor (the “Purchaser”) pursuant to which we agreed to issue and sell, in a registered direct offering (the “Registered Direct Offering”) an aggregate of (i) 1,508,462 shares (the “Shares”) of our common stock, $0.001 par value per share (the “Common Stock” or the “common stock”) at a price of $10.00 per Share and (ii) pre-funded warrants to purchase 3,491,538 shares of Common Stock (the “Pre-Funded Warrants”). Each of the Pre-Funded Warrants is exercisable for one share of Common Stock at the exercise price of $0.01 per Pre-Funded Warrant, immediately exercisable, and may be exercised at any time. The Purchaser’s ability to exercise its Pre-Funded Warrants in exchange for shares of Common Stock is subject to certain beneficial ownership limitations set forth therein. The gross proceeds to us from the Registered Direct Offering were approximately $50 million, before deducting offering expenses payable of approximately $3.1 million. The Registered Direct Offering closed on October 22, 2025. All of the Pre-Funded Warrants were exercised in October 2025.
Industry and Market Strategy
Associative Processing Unit Computing Market Overview
The markets for associative processing computing solutions are significant and growing rapidly. These markets include embedded physical AI edge products, on-premises small-medium business servers, and remote servers. The total addressable market for APU in AI, search applications, and HPC which is where GSI is focusing its APU commercialization efforts, has been determined by GSI to be approximately $247 billion in 2025, and growing at a compound annual growth rate (“CAGR”) of 27% to $708 billion by 2028. GSI has similarly determined that the Serviceable Available Market for APU in edge AI deployments in those markets is approximately $7 billion in 2025, and anticipated to grow at a CAGR of 18%-22% to $16 billion by 2030.
The growth in demand for associative processing computing solutions is being driven by the increasing market adoption and usage of graphics processing unit (“GPU”) and central processing unit (“CPU”) farms for AI processing of large data collections, including parallel computing in scientific research. However, the large-scale usage of GPU and CPU farms for AI processing of data is demonstrating the limits of GPU and CPU processing speeds and resulting in ever higher energy consumption. The amounts of data being processed, which is coming from increasing numbers of users and continuously increasing amounts of collected data, has resulted in efforts to split and store the processed data among multiple databases, through a process called sharding. Sharding can substantially increase processing costs and worsen the power consumption factors associated with processing so much data if the underlying architecture is inefficient to begin with.
The APU has been demonstrated to outperform CPUs and GPUs in the market for AI search of large data collections by providing lower latency and increased capacity in a smaller form-factor and achieve such results with lower power consumption. In addition, our compute-in-place technology has wide application. The APU has several benefits that are particularly useful to overcome the high power challenges of GPUs. First, the APU does not have the word size limitation of traditional CPU and GPU processors. Because traditional data processors move data around to various parts of a system, they need to select or duplicate resources of particular word sizes, be they 8-bit, 16-bit, 32-bit or 64-bit. The APU is based on a memory line structure, which means that it can operate on legacy instruction widths of 8 or 16-bits, or just as seamlessly operate on instructions of arbitrary widths of 1 bit, 768-bits or 2048-bits. APUs can operate on any word width at interim processing steps. This dynamic flexibility is a tremendous advantage for non-linear processing used in high performance compute workloads. Second, the APU is also an associative machine, which means that data that is resident in the device can be applied to a function only if it is deemed associated (for example, with a meta-tag) to the processing. Such processing is like a person looking for his car in a parking lot, but ignoring all cars that are not the color of his car. An additional benefit of the Gemini APU designs is that they are multi-threaded. One sensor or query input can be simultaneously applied to multiple
6
functions or searches in the device or to sequential different functions on a single device towards a single edge solution.
Our associative computing technology utilizes in-memory associative processor structures to address the bottlenecks that limit performance and increase power consumption in CPUs, GPUs, and Field Programmable Gate Arrays (“FPGAs”). By constantly having to move operands and results in and out of devices with ever increasing processing speeds and bus speeds, current solutions are focused on memory transfers rather than addressing the basic computation problem. By changing the computational framework to parallel processing and having search functions conducted directly in a processing memory array, the APU can greatly expedite computation and response times in many “big data” applications. We are creating a new category of computing products that are expected to have substantial target markets and a large new customer base in those markets. We are seeing adoption of this direction in research by at least one hyperscaler and by researchers. Being on our second generation device and architecting our third generation device puts us in a strong position to address target markets as the methodology becomes more mainstream.
Our commercialization efforts for the APU product are focused on markets where the APU shows factors of improvement against CPU or GPU systems. The APU differentiates itself most for similarity search, multi-modal vector search, real-time very large database search, and several scientific high-performance computing-workloads processing sensor data. The APU’s improved performance over CPU or GPU systems provides a paradigm-shifting ability to process data in real-time. As a result, we see applications for the APU in artificial intelligence applications, including approximate nearest neighbor searches, cryptography, and synthetic aperture radar as well as other fields whose processing can benefit from the APU’s smaller footprint, superior productivity, and low system power consumption. GSI has solutions to accelerate multimodal vector search as an on-prem or SaaS solution for OpenSearch and general Fast Vector Search, and for processing large area SAR images in real-time at high resolution.
Similarity search uses a technique called distance metric learning, in which learning algorithms measure how similar related objects are to each other. The APU is well suited for very fast similarity search because its design determines distance metric at fast computation speeds with high degrees of accuracy. Our APU is further differentiated from other solutions in the market by its scalability for very large datasets. The APU has demonstrated its ability to increase the rate of computation for visual search by orders of magnitude with greater accuracy and reduced power consumption. The APU also adds multi-modal search capability to this computational performance. For instance, the ability to search on a picture of a product on an ecommerce website, with pricing and specific filters, does not impede the performance of the in-memory search versus a traditional text only search. This kind of performance has the potential to transform online retailers’ capabilities to run search queries and improve customers’ online shopping experience.
As we continue our efforts to simplify use of the Gemini devices in the markets discussed, we also see opportunities in the edge applications of these markets. In the edge segment the high power and small database coverage of single GPUs is not suitable. While some edge products are coming to market, they do not have the capacity to provide large database support. The APU capabilities and the larger capacity of the Gemini-II chip are well suited for this growing segment. In an attempt to address greater density in processed data, the market is working to reduce bit widths used in models. As the APU technology is ideally suited for smaller bit widths, and including even 1-bit, we are undertaking an effort to adopt several AI models to 1-bit and ternary (1.58-bit) optimization for Gemini-II application. We see this effort as furthering the density and value of our parts for the edge market.
Our commercialization efforts for the Gemini-II APU are focused on markets where low latency, deterministic response, and system efficiency provide clear advantages over conventional CPU and GPU architectures. Gemini-II extends the capabilities of the prior generation beyond high-throughput vector and
7
similarity search into real-time inference domains, enabling fast time-to-first-token (“TTFT”) for vision-language models (“VLMs”) and vision-language-action systems. These capabilities are increasingly critical in emerging physical AI applications, where systems must interpret sensor data and respond within operationally relevant time constraints. Gemini-II delivers these capabilities in a compact, low power form factor, enabling deployment in size, weight, and power constrained environments where traditional accelerators are impractical.
Gemini-II’s architecture is optimized for memory-resident computation, allowing large models and datasets to be processed without the memory movement latency penalties associated with GPUs. This enables rapid initial inference response, or TTFT, which is a key performance metric for interactive and autonomous systems. In contrast to conventional GPU pipelines that prioritize throughput, Gemini-II is designed to minimize inference latency at the point of decision, making it well suited for edge-deployed systems requiring real-time situational awareness and control with continuously new image data. These include robotics, autonomous platforms, industrial sensing systems, and defense applications.
The device maintains strong applicability for similarity search, multi-modal vector search, and large-scale database operations, while extending these capabilities into streaming and real-time inference workflows. For example, Gemini-II can support VLM-based interpretation of video, imagery, command and control, and sensor inputs with response times enabling systems to extract semantic meaning and act on that information in near real time. Gemini-II enables both rapid feature extraction and higher-level inference for awareness-driven decisions within constrained edge environments.
Loading financial statements...
Financial statements
data from SEC XBRL filings. Values are as-reported; restatements supersede originals. Values reported in .
| Line item |
|---|
| Period ending |
Item 7. Management’s Discussion and Analysis of Financial Condition and Results of Operations
The following discussion contains forward-looking statements that involve risks and uncertainties. Our actual results could differ substantially from those anticipated in these forward-looking statements as a result of many factors, including those set forth under “Forward Looking Statements”, “Risk Factors” and elsewhere in this report. The following discussion should be read together with our consolidated financial statements and the related notes included elsewhere in this report.
This discussion and analysis generally covers our financial condition and results of operations for the fiscal year ended March 31, 2026, including year-over-year comparisons versus the fiscal year ended March 31, 2025. Our Annual Report on Form 10-K for the fiscal year ended March 31, 2025 includes year-over-year comparisons versus the fiscal year ended March 31, 2024 in Item 7 of Part II, “Management’s Discussion and Analysis of Financial Condition and Results of Operations.”
Overview
We are a provider of high-performance semiconductor memory solutions for in-place associative computing applications in high growth markets such as artificial intelligence and high-performance computing, including natural language processing and computer vision. Our initial APU products are focused on applications using similarity search, but have not resulted in material revenues to date. Similarity search is used in visual search queries for ecommerce, computer vision, drug discovery, cybersecurity and service markets such as NoSQL, Elasticsearch, and OpenSearch. We have solutions to accelerate multimodal vector search for OpenSearch and general Fast Vector Search, and for processing large area SAR images in real-time at high resolution. Our revenue is currently generated from the design, development and marketing of static random access memories, or SRAMs, that operate at speeds of less than 10 nanoseconds, which we refer to as Very Fast SRAMs, primarily for the networking and telecommunications, test equipment and the military/defense and aerospace markets. We are subject to the highly cyclical nature of the semiconductor industry, which has experienced significant fluctuations, often in connection with fluctuations in demand for the products in which semiconductor devices are used. Our revenues have been substantially impacted by significant fluctuations in sales to our largest end user customers, Nokia, KYEC and Cadence Design Systems. We expect that future direct and indirect sales to Nokia, KYEC and Cadence Design Systems will continue to fluctuate significantly on a quarterly basis. The networking and telecommunications market has accounted for a significant portion of our net revenues in the past and has declined during the past several years and is expected to continue to decline. In anticipation of the decline of the networking and telecommunications market, we have been using the revenue generated by the sales of high-speed synchronous SRAM products to finance the development of our new in-place associative computing solutions and the marketing and sale of new types of SRAM products such as radiation-hardened and radiation-tolerant SRAMs.
As of March 31, 2026, we had cash and cash equivalents of $67.2 million, with no debt. We have a team in-place with tremendous depth and breadth of experience and knowledge, with a legacy business that is providing an ongoing source of funding for the development of new product lines. Our balance sheet and liquidity position was strengthened by the sale of our Sunnyvale, California property in June 2024. In addition, between May and August 2025, we sold 4,508,350 shares of common stock pursuant to an At-the-Market offering, at an average price of $3.29 for net proceeds of $14.3 million. On October 21, 2025, we entered into a securities purchase agreement with an institutional investor pursuant to which we agreed to issue and sell, in a registered direct offering (the “Registered Direct Offering”) an aggregate of (i) 1,508,462 shares of our common stock, $0.001 par value per share, at a price of $10.00 per share and (ii) Pre-Funded Warrants to purchase 3,491,538 shares of Common Stock. Each of the Pre-Funded Warrants is exercisable for one share of Common Stock at the exercise price of $0.01 per Pre-Funded Warrant, immediately exercisable, and may be exercised at any time. The Purchaser’s ability to exercise its Pre-Funded Warrants in exchange for shares of Common Stock is subject to certain beneficial ownership limitations set
44
forth therein. The gross proceeds to the Company from the Registered Direct Offering were $50.0 million, before deducting the placement agents’ fees and other offering expenses payable by the Company of $3.1 million. The Registered Direct Offering closed on October 22, 2025. All of the Pre-Funded Warrants were exercised in October 2025.
Revenues. Substantially all of our revenues are derived from sales of our Very Fast SRAM products. Sales to networking and telecommunications OEMs accounted for 16% to 34% of our net revenues during our last three fiscal years. We also sell our products to OEMs that manufacture products for military and aerospace applications such as radar and guidance systems and satellites, for test and measurement applications such as high-speed testers, for automotive applications such as smart cruise control, and for medical applications such as ultrasound and CAT scan equipment.
The average selling price of our products has increased or remained unchanged in recent years. However, as is typical in the semiconductor industry, the selling prices of our products has historically declined over the life of the product. If prices decline, our ability to increase net revenues, therefore, is dependent upon our ability to increase unit sales volumes of existing products and to introduce and sell new products with higher average selling prices in quantities sufficient to compensate for the anticipated declines in selling prices of our more mature products. Our ability to increase unit sales volumes is dependent primarily upon increases in customer demand but, particularly in periods of increasing demand, can also be affected by our ability to increase production through the availability of increased wafer fabrication capacity from TSMC, our wafer supplier, and our ability to increase the number of good integrated circuit die produced from each wafer through die size reductions and yield enhancement activities.
We may experience fluctuations in quarterly net revenues for a number of reasons. Historically, orders on hand at the beginning of each quarter are insufficient to meet our revenue objectives for that quarter and are generally cancelable up to 30 days prior to scheduled delivery. Accordingly, we depend on obtaining and shipping orders in the same quarter to achieve our revenue objectives. In addition, the timing of product releases, purchase orders and product availability could result in significant product shipments at the end of a quarter. Failure to ship these products by the end of the quarter may adversely affect our operating results. Furthermore, our customers may delay scheduled delivery dates and/or cancel orders within specified timeframes without significant penalty.
We sell our products through our direct sales force, international and domestic sales representatives and distributors. Our customer contracts, which may be in the form of purchase orders, contracts or purchase agreements, contain performance obligations for delivery of agreed upon products. Delivery of all performance obligations contained within a contract with a customer typically occurs at the same time (or within the same accounting period). Transfer of control occurs at the time of shipment, title and the risks and rewards of ownership have passed to the customer, and we have a right to payment. Thus, we will recognize revenue upon shipment of the product for direct sales and sales to our distributors.
Historically, a small number of OEM customers have accounted for a substantial portion of our net revenues, and we expect that significant customer concentration will continue for the foreseeable future. Many of our OEMs use contract manufacturers to manufacture their equipment. Accordingly, a significant percentage of our net revenues is derived from sales to these contract manufacturers. In addition, a significant portion of our sales are made to foreign and domestic distributors who resell our products to OEMs, as well as their contract manufacturers. Direct sales to contract manufacturers accounted for 4.9%, 7.9% and 20.5% of our net revenues for fiscal 2026, 2025 and 2024, respectively. Sales to foreign and domestic distributors accounted for 93.3%, 91.7% and 76.4% of
45
our net revenues for fiscal 2026, 2025 and 2024, respectively. The following direct customers accounted for 10% or more of our net revenues in one or more of the following periods:
| | | | | | | |
| | Fiscal Year Ended | | ||||
| | March 31, | | ||||
| | 2026 | | 2025 | | 2024 |
|
Contract manufacturer: | | | | | | | |
Flextronics Technology | | 2.3 | % | 2.7 | % | 13.5 | % |
Distributors: | | | | | | | |
Avnet Logistics | | 63.7 | | 49.6 | | 50.6 | |
Holystone | | 14.2 | | 22.6 | | 2.5 | |
Nexcomm | | 8.9 | | 9.8 | | 9.3 | |
KYEC was our largest end user customer in fiscal 2026 and 2025. Nokia was our largest end user customer in fiscal 2024. KYEC purchases product through contract manufacturers and distributors. Based on information provided to us by KYEC’s contract manufacturers and distributors, purchases by KYEC represented approximately 14%, 23% and 3% of our net revenues in fiscal 2026, 2025 and 2024, respectively. Nokia purchases products directly from us and through contract manufacturers and distributors. Based on information provided to us by its contract manufacturers and our distributors, purchases by Nokia represented approximately 6%, 12% and 21% of our net revenues in fiscal 2026, 2025 and 2024, respectively. Cadence Design Systems purchases products through contract manufacturers and distributors. Based on information provided to us by its contract manufacturers and our distributors, purchases by Cadence Design Systems represented approximately 12%, 8% and 8% of our net revenues in fiscal 2026, 2025 and 2024, respectively. Our revenues have been substantially impacted by significant fluctuations in sales to Nokia, KYEC and Cadence Design Systems, and we expect that future direct and indirect sales to Nokia, KYEC and Cadence Design Systems will continue to fluctuate substantially on a quarterly basis and that such fluctuations may significantly affect our operating results in future periods. To our knowledge, none of our other OEM customers accounted for more than 10% of our net revenues in fiscal 2026, 2025 or 2024.
Cost of Revenues. Our cost of revenues consists primarily of wafer fabrication costs, wafer sort, assembly, test and burn-in expenses, the amortized cost of production mask sets, stock-based compensation and the cost of materials and overhead from operations. All of our wafer manufacturing and assembly operations, and a significant portion of our wafer sort testing operations, are outsourced. Accordingly, most of our cost of revenues consists of payments to TSMC and independent assembly and test houses. Because we do not have long-term, fixed-price supply contracts, our wafer fabrication, assembly and other outsourced manufacturing costs are subject to the cyclical fluctuations in demand for semiconductors. In recent years we have experienced increased costs as a result of supply chain constraints for wafers and outsourced assembly, burn-in and test operations. We review our manufacturing costs on a regular basis and pass on any cost increases to our customers when it makes sense to do so. Cost of revenues also includes expenses related to supply chain management, quality assurance, and final product testing and documentation control activities conducted at our headquarters in Sunnyvale, California and our branch operations in Taiwan.
Gross Profit. Our gross profit margins vary among our products and are generally greater on our radiation hardened and radiation tolerant SRAMs, on our higher density products and, within a particular density, greater on our higher speed and industrial temperature products. We expect that our overall gross margins will fluctuate from period to period as a result of shifts in product mix, changes in average selling prices and our ability to control our cost of revenues, including costs associated with outsourced wafer fabrication and product assembly and testing.
Research and Development Expenses. Research and development expenses consist primarily of salaries and related expenses for design engineers and other technical personnel, the cost of developing prototypes, stock-based compensation and fees paid to consultants. We charge all research and development expenses to operations as incurred. We charge mask costs used in production to cost of revenues over a 12-month period. However, we charge
46
costs related to pre-production mask sets, which are not used in production, to research and development expenses at the time they are incurred. These charges often arise as we transition to new process technologies and, accordingly, can cause research and development expenses to fluctuate on a quarterly basis. We incurred charges of $2.4 million for a pre-production mask set for our APU2 during the quarter ended December 31, 2023. We incurred charges of $3.2 million for intellectual property rights that we purchased for our Plato project during the quarter ended December 31, 2025. We believe that continued investment in research and development is critical to our long-term success, and we expect to continue to devote significant resources to product development activities. In particular, we are devoting substantial resources to the development of our in-place associative computing products. Accordingly, we expect that our research and development expenses will continue to be substantial in future periods and may lead to operating losses in some periods. Such expenses as a percentage of net revenues may fluctuate from period to period.
Selling, General and Administrative Expenses. Selling, general and administrative expenses consist primarily of commissions paid to independent sales representatives, salaries, stock-based compensation and related expenses for personnel engaged in sales, marketing, administrative, finance and human resources activities, professional fees, costs associated with the promotion of our products and other corporate expenses. We expect that our sales and marketing expenses will increase in absolute dollars in future periods if we are able to grow and expand our sales force but that, to the extent our revenues increase in future periods, these expenses will generally decline as a percentage of net revenues. We also expect that, in support of any future growth that we are able to achieve, general and administrative expenses will generally increase in absolute dollars.
Acquisition
On November 23, 2015, we acquired all of the outstanding capital stock of privately held MikaMonu Group Ltd. (“MikaMonu”), a development-stage, Israel-based company that specialized in in-place associative computing for markets including big data, computer vision and cyber security. MikaMonu, located in Tel Aviv, held 12 United States patents and had a number of pending patent applications.
The allocation of the purchase price to acquired identifiable intangible assets and goodwill was based on their estimated fair values at the date of acquisition. The fair value allocated to patents was $3.5 million and the residual value allocated to goodwill was $8.0 million.
The acquisition agreement provided for potential “earnout” payments to the former MikaMonu shareholders in cash or shares of GSI Technology’s common stock, at our discretion, during a period of up to ten years following the closing of the acquisition if certain revenue targets for products based on MikaMonu technology were achieved. December 31, 2025 was the final date during which revenues from the sale of qualifying products were measured for purposes of calculating earnout consideration under the acquisition agreement. None of the revenue targets have been achieved, the amount of revenues recognized during the measurement period was not sufficient to create an earnout payment obligation, and no revenue-based earnout payments have been paid.
47
Results of Operations
The following table sets forth statement of operations data as a percentage of net revenues for the periods indicated:
| | | | | | |
| | Year Ended March 31, | | |||
| | 2026 | | | 2025 | |
Net revenues | | 100.0 | % | | 100.0 | % |
Cost of revenues | | 45.5 | | | 50.6 | |
Gross profit | | 54.5 | | | 49.4 | |
Operating expenses: | | | | | | |
Research and development | | 79.4 | | | 78.0 | |
Selling, general and administrative | | 44.7 | | | 52.5 | |
Gain from sale of assets | | — | | | (28.2) | |
Total operating expenses | | 124.1 | | | 102.3 | |
Loss from operations | | (69.6) | | | (52.9) | |
Interest and other income, net | | 16.3 | | | 1.6 | |
Loss before income taxes | | (53.3) | | | (51.3) | |
Provision (benefit) for income taxes | | (0.6) | | | 0.6 | |
Net loss | | (52.7) | | | (51.9) | |
Fiscal Year Ended March 31, 2026 Compared to Fiscal Year Ended March 31, 2025
Net Revenues. Net revenues increased by 22.4% from $20.5 million in fiscal 2025 to $25.1 million in fiscal 2026. The overall average selling price of all units shipped in fiscal 2026 increased by 16.3% in fiscal 2026 compared to the prior fiscal year. Units shipped increased by 5.7% in fiscal 2026 compared to fiscal 2025. KYEC, which is a leading provider in the test and measurement market, was our largest end user customer in fiscal 2026 and 2025. Direct and indirect sales to KYEC decreased by $1.0 million from $4.6 million in fiscal 2025 to $3.6 million fiscal 2026. Direct and indirect sales to Nokia decreased by $1.0 million from $2.5 million in fiscal 2025 to $1.5 million fiscal 2026. Direct and indirect sales to Cadence Design Systems increased by $1.5 million from $1.6 million in fiscal 2025 to $3.1 million in fiscal 2026. The decrease in Nokia’s purchases in the past several fiscal years is due in part to Nokia’s decision to replace SRAM with alternative memory solutions. The test and measurement markets represented 38% and 32% of shipments in fiscal 2026 and in fiscal 2025, respectively. The networking and telecommunications markets represented 16% and 19% of shipments in fiscal 2026 and in fiscal 2025, respectively. Shipments to KYEC, Nokia and Cadence Design Systems will continue to fluctuate on a quarterly basis as a result of demand and shipments to their end customers. While recent customer order patterns have been particularly variable, these fluctuations are related to economic and external factors, which include worldwide inflationary pressures, increased or new tariffs, export controls and other trade barriers and trade disputes, increasing geopolitical tensions and the challenging global economic environment.
Cost of Revenues. Cost of revenues increased by 10.1% from $10.4 million in fiscal 2025 to $11.4 million in fiscal 2026. The increase in cost of revenues was primarily related to the increase in net revenues in fiscal 2026 compared to fiscal 2025 and changes in the mix of products and customers. Cost of revenues included a provision for excess and obsolete inventories of $301,000 and $305,000 in fiscal 2026 and in fiscal 2025, respectively. Cost of revenues included stock-based compensation expense of $231,000 and $199,000, respectively, in fiscal 2026 and fiscal 2025. Cost of revenues in fiscal 2025 includes $204,000 in severance related payments related to our August 2024 cost reduction initiative.
48
Gross Profit. Gross profit increased by 35.1% from $10.1 million in fiscal 2025 to $13.7 million in fiscal 2026. Gross margin increased from 49.4% in fiscal 2025 to 54.5% in fiscal 2026. The change in gross profit is primarily related to the change in net revenues discussed above. The increase in gross margin was primarily related to change in the mix of products and customers and also reflects the impact of fixed overhead on higher shipment levels compared to the prior year. Gross margin in fiscal 2025 was also impacted by the severance related payments related to our August 2024 cost reduction initiative discussed above.
Research and Development Expenses. Research and development expenses increased 24.6% from $16.0 million in fiscal 2025 to $19.9 million in fiscal 2026. The increase in research and development spending was primarily related to charges of $3.2 million for intellectual property rights that we purchased for our Plato project during the quarter ended December 31, 2025, increases in outside consulting expenses, also related to our Plato project and lesser increases in payroll related expenses, partially offset by a lesser decrease in software maintenance expenses. Research and development expenses in fiscal 2026 and fiscal 2025 were also offset by $1.0 million and $1.2 million, respectively, of funding received under the government contracts. Research and development expenses included stock-based compensation expense of $1.0 million and $945,000 in fiscal 2026 and fiscal 2025, respectively.
Selling, General and Administrative Expenses. Selling, general and administrative expenses increased 4.3% from $10.8 million in fiscal 2025 to $11.2 million in fiscal 2026. Increases of $574,000 stock-based compensation expense and $573,000 in facility related expenses were partially offset by a decrease of $465,000 in professional fees and a lesser decrease in payroll related expenses. In fiscal 2025, the value of contingent consideration liability resulting from our prior acquisition of the MikaMonu Group Ltd. decreased by $160,000. Selling, general and administrative expenses included stock-based compensation expense of $1.6 million and $1.1 million in fiscal 2026 and fiscal 2025, respectively.
Gain from Sale of Assets. Gain from sale of assets includes the gain from the sale of our headquarters building located at 1213 Elko Drive in Sunnyvale, California. The sale and leaseback transaction was completed on June 6, 2024. For further discussion of the sale and leaseback transaction, see Note - 8 Leases to the consolidated financial statements contained elsewhere in this report.
Interest Income and Other (Expense), Net. Interest income and other (expense), net increased from income of $326,000 in fiscal 2025 to income of $4.1 million in fiscal 2026. Interest income increased by $481,000 primarily due to higher cash balances invested in money market funds. The foreign currency exchange loss increased from ($119,000) in fiscal 2025 to ($206,000) in fiscal 2026. The exchange loss in each period was primarily related to our Taiwan branch operations and operations in Israel. Other income in fiscal 2026 included a gain on the change in the fair value of warrants of $6.2 million and costs associated with the Registered Direct Offering of $2.8 million.
Provision (benefit) for Income Taxes. The provision (benefit) for income taxes was $130,000 in fiscal 2025 to ($132,000) in fiscal 2026. Because we recorded a cumulative three-year loss on a U.S. tax basis for the year ended March 31, 2026 and the realization of our deferred tax assets is questionable, we recorded a tax provision reflecting a valuation allowance of $25.4 million in net deferred tax assets in fiscal 2026. Reductions in uncertain tax benefits due to lapses in the statute of limitations were $767,000 in fiscal 2025 and were not significant in fiscal 2026.
Net Loss. Net loss was $10.6 million in fiscal 2025 compared to a net loss of $13.2 million in fiscal 2026. This increase in net loss was primarily due to the changes in net revenues, gross profit and operating expenses discussed above.
49
Liquidity and Capital Resources
As of March 31, 2026, our principal sources of liquidity were cash and cash equivalents of $67.2 million compared to $13.4 million of cash, cash equivalents and short-term investments as of March 31, 2025. Cash and cash equivalents totaling $16.0 million were held in foreign locations as of March 31, 2026.
Net cash used in operating activities was $15.9 million and $13.0 million for fiscal 2026 and fiscal 2025, respectively. Cash from operations in fiscal 2026 was adjusted for the non-cash gain on the change in fair value of warrants in the amount of $6.2 million. The primary uses of cash in fiscal 2026 were the net loss of $13.2 million and increases of $1.5 million in accrued expenses and other liabilities, $1.2 million in accounts receivable and $1.0 million in prepaid expenses and other assets. The primary source of cash in fiscal 2026 was an increase in accounts payable of $2.6 million. The uses of cash in fiscal 2026 were offset by non-cash items including stock-based compensation of $2.8 million and depreciation and amortization expenses of $628,000.
The primary uses of cash in fiscal 2025 were the net loss of $10.6 million and an increase of $1.1 million in prepaid expenses and other assets. The increase in prepaid expenses and other assets was primarily related to a production mask set for our APU2. The primary source of cash in fiscal 2025 was a reduction in inventories of $781,000. Cash from operations in fiscal 2025 was adjusted for the non-cash gain on the sale of assets in the amount of $5.8 million. The uses of cash in fiscal 2025 were offset by non-cash items including stock-based compensation of $2.3 million and depreciation and amortization expenses of $665,000.
Net cash used by investing activities was $486,000 in fiscal 2026 and net cash provided by investing activities was $11.3 million in fiscal 2025. Investment activities in fiscal 2026 consisted of the purchase of property and equipment of $486,000. Investment activities in fiscal 2025 primarily consisted of the net proceeds of $11.4 million from a sale and leaseback transaction, partially offset by the purchase of property and equipment of $45,000.
Cash provided by financing activities was $70.2 million and $633,000 in fiscal 2026 and fiscal 2025, respectively. Cash provided by financing activities in fiscal 2026 primarily consisted of the proceeds from the issuance of common stock and warrants of $49.7 million, proceeds from the sale of common stock pursuant to an At-the-Market offering of $14.3 million and the proceeds from the sale of common stock pursuant to our employee stock plans of $6.2 million. Net cash provided by financing activities in fiscal 2025 consisted of the proceeds from the sale of common stock pursuant to our employee stock plans of $633,000.
At March 31, 2026, we had total minimum lease obligations of approximately $10.9 million from April 1, 2025 through May 30, 2034, under non-cancelable operating leases for our facilities.
While higher interest rates, worldwide inflationary pressures, tariffs and trade disputes, increasing geopolitical tensions and the decline in the global economic environment have created significant uncertainty as to general economic and capital market conditions for fiscal 2027 and beyond, we believe that our existing balances of cash and cash equivalents, and cash flow expected to be generated from our future operations will be sufficient to meet our cash needs for working capital and capital expenditures for at least the next 12 months. Our future capital requirements will depend on many factors, including revenue growth, if any, that we experience, any additional manufacturing cost increases resulting from supply constraints and the continuation of the impact of higher interest rates and inflation may have on our business, the extent to which we utilize subcontractors, the levels of inventory and accounts receivable that we maintain, the timing and extent of spending to support our product development efforts as well as potentially additional funding to complete the commercialization and development of Gemini-II and Plato, and the expansion of our sales and marketing team. Additional capital may also be required for the consummation of any acquisition of businesses, products or technologies that we may undertake. On June 28, 2023, we filed a registration statement on Form S-3, which was declared effective by the SEC on July 19, 2023. On August 1, 2023, we commenced a registered securities offering pursuant to a Sales Agreement (the “Sales
50
Agreement”) with Needham & Company, LLC (“Needham”). The Sales Agreement provides that we may offer and sell our common stock having an aggregate offering price of up to $25.0 million from time to time (the “Offering”) through Needham, acting as our sales agent. We sold 133,000 shares pursuant to the Offering at an average price of $4.20 for proceeds of $542,000, less offering costs of $389,000 during the quarter ended September 30, 2023. In May and June 2025, we sold 3,380,773 shares pursuant to the Offering at an average price of $3.32 for proceeds of $11.2 million, less offering costs of $411,000. We cannot assure that additional equity or debt financing, if required, will be available on terms that are acceptable or at all.
On October 21, 2025, we entered into the Purchase Agreement with the Purchaser pursuant to which we agreed to issue and sell, in the Registered Direct Offering an aggregate of (i) 1,508,462 Shares of our Common Stock at a price of $10.00 per Share and (ii) the Pre-Funded Warrants to purchase 3,491,538 shares of Common Stock. Each of the Pre-Funded Warrants is exercisable for one share of Common Stock at the exercise price of $0.01 per Pre-Funded Warrant, immediately exercisable, and may be exercised at any time. The Purchaser’s ability to exercise its Pre-Funded Warrants in exchange for shares of Common Stock is subject to certain beneficial ownership limitations set forth therein. The gross proceeds to us from the Registered Direct Offering were approximately $50 million, before deducting offering expenses payable of approximately $3.1 million. The Registered Direct Offering closed on October 22, 2025. All of the Pre-Funded Warrants were exercised in October 2025.
As of March 31, 2026, we had $18.5 million in purchase obligations for facility leases, wafers, software maintenance and chip design service purchase obligations that are binding commitments, of which $9.1 million are payable in the next twelve months and $9.4 million are committed in the long term.
Critical Accounting Estimates
The preparation of our consolidated financial statements and related disclosures in conformity with accounting principles generally accepted in the United States (“GAAP”) requires us to make estimates and assumptions that affect the reported amounts of assets and liabilities at the date of the financial statements and the reported amounts of revenue and expenses during the reporting period. Significant estimates are inherent in the preparation of the consolidated financial statements and include estimates affecting obsolete and excess inventory. We believe that we consistently apply these judgments and estimates and that our financial statements and accompanying notes fairly represent our financial results for all periods presented. However, any errors in these judgments and estimates may have a material impact on our balance sheet and statement of operations. Critical accounting estimates, as defined by the Securities and Exchange Commission, are those that are most important to the portrayal of our financial condition and results of operations and require our most difficult and subjective judgments and estimates of matters that are inherently uncertain. Our critical accounting estimates include those regarding the valuation of inventories.
Valuation of Inventories. Inventories are stated at the lower of cost or net realizable value, cost being determined on a weighted average basis. Our inventory write-down allowance is established when conditions indicate that the selling price of our products could be less than cost due to physical deterioration, obsolescence based on changes in technology and demand, changes in price levels, or other causes. We consider the need to establish the allowance for excess inventory generally based on inventory levels in excess of 12 months of forecasted customer demand for each specific product, which is based on historical sales and expected future orders. At any point in time, some portion of our inventory is subject to the risk of being materially in excess of our projected demand. Additionally, our average selling prices could decline due to market or other conditions, which creates a risk that costs of manufacturing our inventory may not be recovered. These factors contribute to the risk that we may be required to record additional inventory write-downs in the future, which could be material. In addition, if actual market conditions are more favorable than expected, inventory previously written down may be sold to customers resulting in lower cost of sales and higher income from operations than expected in that period.
51
Recent insider activity
| Date | Insider | Role | Action | Shares | Price | Value |
|---|---|---|---|---|---|---|
| 2026-06-02 | CHUANG PATRICK T | Senior VP, Memory Design | Sell | -40,000 | $10.98 | -$439,064 |
| 2026-06-01 | CHUANG PATRICK T | Senior VP, Memory Design | Sell | -40,000 | $11.31 | -$452,584 |
| 2026-06-01 | Wu Bor-Tay | VP, Taiwan Operations | Sell | -20,000 | $10.93 | -$218,658 |
| 2026-05-26 | Wu Ping Tak | VP, U.S. Operations | Sell | -30,000 | $11.01 | -$330,396 |
| 2026-05-22 | Wu Bor-Tay | VP, Taiwan Operations | Sell | 0 ×2 | $10.93 | -$2 |
| 2026-05-21 | Lasserre Didier | VP, Sales | Sell | -30,000 | $9.31 | -$279,180 |
| 2026-05-14 | Schirle Douglas | CFO | Sell | -40,000 | $11.32 | -$452,928 |
| 2026-05-13 | Shu Lee-Lean indirect | Pres., CEO and Chairman | Sell | -10,313 | $12.51 | -$128,977 |
| 2026-05-12 | Wu Ping Tak | VP, U.S. Operations | Sell | -11,763 | $10.80 | -$127,040 |
| 2026-05-13 | Akerib Avidan | VP, Associative Computing | Sell | -19,653 | $11.47 | -$225,326 |
| 2026-05-12 | Akerib Avidan | VP, Associative Computing | Sell | -347 | $12.03 | -$4,174 |
| 2026-05-12 | CHUANG PATRICK T | Senior VP, Memory Design | Sell | -40,000 | $10.68 | -$427,088 |
| 2026-05-11 | Shu Lee-Lean indirect | Pres., CEO and Chairman | Sell | -10,313 | $10.51 | -$108,420 |
| 2026-05-11 | Shu Lee-Lean | Pres., CEO and Chairman | Sell | -132,749 ×2 | $11.54 | -$1,531,788 |
Source: SEC Form 4 filings.
Next expected filings
- ~2026-08-11 10-Q expected by 2026-08-14 (in 48 days)
- ~2026-11-10 10-Q expected by 2026-11-13 (in 139 days)
- ~2027-02-09 10-Q expected by 2027-02-12 (in 230 days)
- ~2027-06-04 10-K expected by 2027-06-08 (in 345 days)
Predicted from historical filing cadence; not an SEC commitment.
Recent SEC filings
- 2026-06-05 10-K Annual Report
- 2026-05-28 8-K Officer/Director Change; Financial Statements and Exhibits
- 2026-05-22 8-K Earnings Release; Other Events
- 2026-05-07 8-K Earnings Release; Financial Statements and Exhibits
- 2026-03-18 8-K Other Events; Financial Statements and Exhibits
- 2026-02-06 10-Q Quarterly Report
- 2026-01-29 8-K Earnings Release; Financial Statements and Exhibits
- 2025-11-07 10-Q Quarterly Report
- 2025-10-30 8-K Earnings Release; Financial Statements and Exhibits
- 2025-10-21 8-K Material Agreement Entered; Other Events; Financial Statements and Exhibits
- 2025-10-21 8-K Earnings Release; Other Events
- 2025-08-22 8-K Officer/Director Change; Shareholder Vote Results
- 2025-08-08 10-Q Quarterly Report
- 2025-07-31 8-K Earnings Release; Other Events; Financial Statements and Exhibits
- 2025-06-18 10-K Annual Report