Executive Summary / Key Takeaways

• GSI Technology is undergoing a strategic transformation, leveraging its profitable legacy SRAM business to fund the development and commercialization of its high-potential, but currently non-revenue-generating, in-place associative computing (APU) technology for high-growth AI and HPC markets.
• The APU, particularly the Gemini-II and the new Plato chip, offers significant performance advantages over traditional architectures for specific workloads like similarity search and edge AI/LLMs, demonstrated by quantifiable benefits such as up to 85% faster index building and a target of >10x improvement in 1-bit/2-bit processing.
• Recent financial performance shows a revenue decline in fiscal 2025 driven by SRAM market softness, but significant progress in reducing net loss and cash burn through cost-cutting initiatives and a headquarters sale/leaseback, strengthening the balance sheet.
• Near-term outlook is supported by anticipated stable demand from a key SRAM customer in the AI chip manufacturing supply chain, expected follow-on orders for high-margin radiation-hardened SRAM, and progress on government SBIR contracts which provide funding and validation for the APU.
• The core investment thesis hinges on successful APU commercialization, securing strategic partnerships and funding for future development (like Plato), and navigating risks including intense competition from larger players, reliance on single suppliers, and the inherent challenges of bringing disruptive technology to market.

Setting the Scene: A Semiconductor Company in Transition

GSI Technology, Inc. (NASDAQ: GSIT) is a semiconductor company with roots stretching back to 1995, initially establishing itself as a provider of high-speed synchronous static random access memory (SRAM). For decades, its SRAM products served critical applications in networking, telecommunications, test equipment, and the demanding military and aerospace sectors. However, recognizing the evolving landscape of computing and the emergence of data-intensive workloads, GSI embarked on a transformative strategic pivot.

This pivot was catalyzed by the 2015 acquisition of MikaMonu Group Ltd., which brought in-place associative computing technology into GSI's portfolio. This technology represents a fundamental departure from the traditional Von Neumann architecture that bottlenecks conventional processors (CPUs, GPUs) by requiring data to constantly move between memory and processing units. GSI's strategy is now dual-pronged: maintain and optimize the legacy SRAM business to generate the necessary capital, while simultaneously investing heavily in and commercializing its innovative associative processing unit (APU) technology for high-growth markets like Artificial Intelligence (AI) and High-Performance Computing (HPC). The success of this transformation is the central narrative for investors.

The Technological Core: Unlocking Performance with Compute-In-Memory

At the heart of GSI's future lies its in-place associative computing technology, embodied in the Gemini family of APU chips. This compute-in-memory architecture allows processing to occur directly within the memory array, drastically reducing data movement and its associated latency and power consumption penalties.

The APU is particularly well-suited for workloads involving similarity search and Boolean processing, critical functions in areas like large database search, computer vision, drug discovery, and cybersecurity. Unlike traditional processors with fixed word sizes, the APU can operate on data of arbitrary widths, offering dynamic flexibility for non-linear processing. Its associative nature allows it to process only relevant data, akin to searching for a specific item in a large collection by filtering based on attributes directly in memory. The Gemini designs are also multi-threaded, enabling simultaneous application of a single input to multiple functions.

Specific, quantifiable benefits highlighted by the company underscore the APU's potential:

• Demonstrated ability to outperform CPUs and GPUs in AI search of large data collections, offering lower latency and increased capacity in a smaller, lower-power footprint.
• Reduced HNSW index build time by approximately 85% compared to traditional CPU-based solutions, a critical advantage for fast vector search applications in AI and e-commerce.
• Targeting a demonstration of greater than 10 times improvement in 1-bit and 2-bit processing with the Gemini-II, a key focus for efficient edge AI applications.

The Gemini family includes Gemini-I, currently in full production and being marketed for specific applications and as-a-service offerings. The second-generation Gemini-II, with first silicon received in January 2024, is designed for edge applications, promising an order of magnitude improvement in performance and lower power consumption. Software workarounds have been identified for minor bugs on the first silicon, allowing firmware and software library development to proceed, with a target completion by December 2024 and customer sampling in early calendar 2025.

Building on the Gemini-II architecture, GSI has begun development of a new chip named Plato. This chip is specifically designed to address the growing market for Large Language Models (LLMs) at the edge, emphasizing ultra-low power consumption while aiming to deliver data center performance levels. Recent enhancements to Plato include the integration of a camera interface and other connectivity features, making it particularly well-suited for AI agents requiring object recognition directly on the device.

For investors, the "so what" of this technology is its potential to create a significant competitive moat in specific, high-growth AI/HPC niches where traditional architectures struggle. If GSI can successfully commercialize these performance advantages, it could capture meaningful market share and establish new, higher-margin revenue streams, fundamentally changing its financial profile.

Competitive Landscape: Niche Innovation Against Giants

GSI operates in a highly competitive semiconductor market, facing both large, diversified players and smaller, specialized firms. In the traditional memory space, competitors include giants like Micron Technology (MU), as well as other providers like Infineon Technologies AG, Integrated Silicon Solution (ISSI), and Renesas Electronics Corporation. These companies often have substantially greater resources, broader product portfolios, and in some cases, their own manufacturing facilities, which can offer cost and capacity advantages. Micron, for example, holds a dominant market share (around 25.88% in semiconductor memory) and benefits from massive scale and R&D investment, enabling lower per-unit manufacturing costs and strong pricing discipline.

In the emerging associative computing and AI acceleration markets, GSI competes with established players like NVIDIA Corporation (NVDA) and Intel Corporation (INTC), as well as smaller companies like Quicklogic Corp (QUIK) and Peraso Inc (PRSO), and expects new entrants. NVIDIA and Intel offer powerful GPU and CPU-based accelerators that dominate the broader AI market. However, GSI's APU differentiates itself by focusing on specific workloads where its compute-in-memory architecture provides a distinct advantage. While competitors might offer broader solutions or greater processing power for general AI tasks, GSI's technology can offer superior performance metrics (like latency, capacity, power efficiency) for tasks such as similarity search and certain edge AI applications. For instance, the APU's ability to accelerate HNSW index building by 85% is a direct quantifiable advantage over traditional CPU methods. Similarly, the focus on low-power edge LLMs with Plato targets a segment where the high power consumption of large GPUs is a significant limitation.

GSI's competitive strategy leverages its niche technological strengths and a fabless business model. The fabless model allows focus on design and access to advanced process technologies with lower capital investment, providing a degree of cost efficiency (e.g., potentially 15% lower manufacturing overhead compared to vertically integrated models). Its proprietary radiation-hardened SRAM technology offers a significant advantage in military and aerospace markets, providing lower failure rates (potentially 20-30% lower) in extreme environments, which translates into higher gross margins (mentioned as "north of 90%" for RadHard SRAM) and customer loyalty in these specialized, high-reliability segments.

However, GSI's smaller scale (market share less than 0.05%) and lower production volumes compared to giants like Micron result in higher per-unit costs for standard products and less pricing power. Its financial performance, characterized by recent net losses and negative free cash flow, also lags behind more profitable competitors like Micron and Quicklogic (which has recently achieved positive operating margins). Dependence on single-source suppliers like TSMC (TSM) for wafer fabrication and key distributors like Avnet Logistics (AVT) also presents operational risks that larger, more diversified competitors may mitigate more effectively.

To compete, GSI must continue to innovate rapidly in its niche areas, effectively demonstrate the quantifiable performance advantages of its APU technology, and build strong customer relationships, particularly in the military/defense and emerging edge AI markets, where its unique capabilities can provide a compelling value proposition that offsets its scale disadvantages.

The SRAM Business: Funding the Future

Historically, GSI's revenue has been almost entirely derived from its Very Fast SRAM products. This segment continues to be the financial engine of the company, providing the revenue necessary to fund the significant research and development investments required for the APU technology.

In fiscal year 2025, net revenue decreased by 5.7% to $20.5 million compared to $21.8 million in fiscal 2024. This decline was attributed to cautionary customer spending, macroeconomic pressures (inflation, interest rates), geopolitical tensions, and a general decline in the global economic environment impacting demand for SRAM products. The overall gross margin also decreased in fiscal 2025 to 49.4% from 54.3% in fiscal 2024, primarily due to product mix, the impact of lower revenue on fixed manufacturing costs, and severance payments.

However, recent performance shows signs of improvement. The fourth quarter of fiscal 2025 saw solid revenue growth (14% year-over-year, 9% sequentially) to $5.9 million, driven by strong demand for SRAM chips. This uptick was partly due to existing customers depleting channel inventories and resuming orders. Crucially, GSI has secured a new SRAM design win with a customer whose systems are integral to manufacturing leading AI chips. Shipments of GSI's 144Mb SRAM to this customer began in calendar 2024, and this customer is anticipated to become GSI's largest SRAM customer in fiscal year 2025 and potentially the number one customer overall in the near future, driving substantial demand.

Another significant opportunity in the SRAM segment is the radiation-hardened (RadHard) and radiation-tolerant (RadTolerant) product line for military/defense and aerospace. In March 2025, GSI secured an initial production order for its RadHard SRAM from a North American prime contractor, with follow-on orders expected in fiscal 2026. These RadHard products carry significantly higher gross margins (mentioned as "north of 90%") than traditional SRAM, offering a strong financial lever. Pursuing heritage status for this chip is a key strategic goal to improve market readiness and open new sales channels.

While the networking and telecommunications market for external SRAM is expected to continue its decline due to the trend of embedding more memory into ASICs, the military/defense and aerospace markets, requiring high densities and reliability in harsh environments, present ongoing opportunities that GSI is well-positioned to serve with its specialized products.

APU Commercialization: The Path Forward

The long-term growth story for GSI hinges on the successful commercialization of its APU technology, which has not yet generated material revenue. The company is actively pursuing multiple avenues to bring its APU products to market and establish revenue streams.

For the Gemini-I APU, efforts are focused on specific applications and service offerings. GSI is preparing to launch the GXL platform for Fast Vector Search, available as cloud-based or on-premise solutions. This platform leverages Gemini-I's ability to accelerate HNSW index building, offering significant time and cost savings for applications like vector similarity search in AI and e-commerce. GSI also offers its SAR processing capabilities as a SaaS product.

The Gemini-II chip is central to the near-term APU strategy, particularly for edge AI applications. Following the receipt of first silicon, the software team is working to finalize firmware and complete the software library by December 2024, enabling customer sampling in early calendar 2025. A key focus is demonstrating the chip's performance advantages, specifically targeting a validation of greater than 10 times improvement in 1-bit and 2-bit compute-in-memory processing by the December quarter. Successfully showcasing this capability is seen as crucial for engaging in more productive discussions with potential strategic partners.

Government contracts, particularly through the Small Business Innovation Research (SBIR) program, are a critical component of the APU commercialization strategy. These contracts provide non-dilutive funding for development and offer valuable validation and exposure within the Department of Defense (DoD). GSI is currently executing Phase II SBIR contracts with the Space Development Agency ($1.25 million, over 50% complete) and the Air Force Research Labs ($1.1 million, over 25% complete), and recently secured a Phase I contract with the U.S. Army (up to $250,000) focused on evaluating Gemini-II for edge computing AI solutions. The company is actively writing multiple SBIR proposals, with a total pipeline valued at $6 million, including a potential Phase II U.S. Army contract worth up to $2 million. These SBIRs are expected to build a pipeline of future revenue and help offset Gemini-II development costs.

Looking further ahead, the development of Plato for edge LLMs is a strategic priority, but this $50 million development program is contingent on securing significant funding through strategic and financial partners. Preliminary discussions are underway, and the recent enhancement of Plato with an integrated camera interface is aimed at increasing strategic interest.

Overall, the APU commercialization path is focused on demonstrating tangible performance benefits, securing government funding and validation, launching initial product platforms (like GXL), and seeking strategic partnerships to scale development and market reach.

Financial Performance and Liquidity

GSI Technology has faced financial challenges, reporting significant net losses in recent years ($10.6 million in FY2025, $20.1 million in FY2024, $16.0 million in FY2023). However, fiscal year 2025 showed meaningful progress in reducing the net loss, which decreased by 47% compared to fiscal 2024. This improvement was largely driven by a 35% reduction in operating expenses, reflecting strategic cost-cutting measures implemented in August 2024, including workforce reductions across all departments, projected to generate annualized savings of about $3.5 million.

Research and development expenses decreased significantly in fiscal 2025 to $16.0 million from $21.7 million in fiscal 2024, partly due to lower pre-production mask costs and payroll expenses, offset by government funding received under SBIR programs ($1.2 million in FY2025 vs. $440,000 in FY2024). Selling, general, and administrative expenses remained relatively stable at $10.8 million in FY2025 compared to $10.6 million in FY2024.

Liquidity was strengthened by the sale and leaseback of the company's headquarters in Sunnyvale, California, completed in June 2024. This transaction generated net cash proceeds of $11.2 million and resulted in a $5.7 million gain recorded in the first quarter of fiscal 2025. As of March 31, 2025, GSI had $13.4 million in cash and cash equivalents, compared to $14.4 million at March 31, 2024.

Net cash used in operating activities improved to $13.0 million in fiscal 2025 from $17.4 million in fiscal 2024. The company also raised $11.2 million in proceeds (net of costs) in May and June 2025 through an At-the-Market (ATM) offering.

Management believes that existing cash and cash equivalents, combined with expected cash flow from future operations, will be sufficient to meet cash needs for working capital and capital expenditures for at least the next 12 months. However, future capital requirements, particularly for scaling APU development (like the Plato program) or potential acquisitions, may necessitate additional funding.

Strategic Outlook and Key Risks

GSI's outlook for fiscal year 2026 is cautiously optimistic, balancing the potential of the APU with the realities of the SRAM market and the need for external funding. For the first quarter of fiscal 2026, GSI expects revenues between $5.5 million and $6.3 million, with a gross margin projected in the range of 56% to 58%. This guidance reflects anticipated continued momentum in SRAM demand, particularly from the AI chip manufacturing customer, and potential early APU commercialization milestones.

The company plans to build on APU development progress, drive continued growth in SRAM sales (especially RadHard), and advance strategic initiatives in both commercial and government markets. Maintaining quarterly operating expenses at current reduced levels is crucial to minimize cash burn until new funding sources are secured. The ongoing strategic review, initiated in May 2024 with Needham & Company as advisor, explores a range of options including financing, divestiture, licensing, or a sale, with a primary focus on securing funding for the next phase of APU development.

However, the investment thesis is subject to significant risks:

• APU Commercialization Risk: The APU has not yet generated material revenue, and market adoption may take longer or be less successful than anticipated, impacting the ability to offset development costs.
• Funding Risk: The development of next-generation APU products like Plato requires substantial capital ($50 million estimated), and there is no assurance that strategic partnerships or external funding will be secured on favorable terms or at all.
• SRAM Market Dependence: The company remains heavily dependent on SRAM sales, and a significant downturn in demand or fluctuations in purchases by key customers (KYEC, Nokia (NOK), or the new AI chip manufacturing customer) would severely impact financial results and the ability to fund APU development.
• Development Risk: Bringing complex, cutting-cutting technology like the APU to market is subject to technological problems, delays, and unanticipated costs.
• Supplier Concentration: Reliance on single-source suppliers like TSMC for wafer fabrication exposes GSI to risks of supply constraints, manufacturing failures, or price increases that could harm the business.
• Competition: GSI faces intense competition from larger, better-resourced companies that could offer alternative solutions or aggressively price products, potentially limiting GSI's market share and profitability.
• Geopolitical Risks: Operations in Taiwan (manufacturing/testing) and Israel (software development) expose the company to risks from political instability, military conflict, or restrictions on transportation logistics.

Conclusion

GSI Technology is at a critical juncture, executing a bold strategic transformation from a legacy SRAM provider to a potential leader in niche compute-in-memory solutions for AI and HPC. While the SRAM business continues to provide essential funding, recent performance highlights both its vulnerability to market cycles and new growth opportunities, particularly in the AI chip manufacturing supply chain and high-margin RadHard products. The core investment narrative centers on the successful commercialization of the APU technology, which offers compelling, quantifiable performance advantages for specific workloads like similarity search and edge AI/LLMs compared to traditional architectures.

Progress on the APU front, including the development of Gemini-II and Plato and securing government SBIR contracts, demonstrates technological advancement and strategic execution. However, the path to material APU revenue is long and uncertain, requiring significant further investment and successful market adoption. The company's ability to secure strategic partnerships and funding, effectively navigate intense competition from larger players, and manage operational risks will be paramount. Investors should closely monitor progress on Gemini-II software development and benchmarking, the launch of the GXL platform, the securing of additional SBIR contracts, and developments from the strategic review as key indicators of whether GSI can successfully transition and unlock the potential value of its compute-in-memory technology.