GSI Technology, Inc.

# Business Overview: GSI Technology, Inc. (NASDAQ: GSIT)

---

Executive Summary

GSI Technology, Inc. is a fabless semiconductor company best known for decades of leadership in very-fast synchronous SRAM (static random-access memory), now using the cash flow from that legacy business to fund a higher-risk, higher-reward pivot into Associative Processing Unit (APU) chips — an "in-place associative computing" architecture the company brands Gemini, aimed at AI search, edge inference, and defense applications.

As a fabless designer, GSI relies on TSMC for wafer fabrication and ASE for most packaging. The company's SRAM business remains profitable-adjacent and well-established in defense/aerospace and networking niches, while Gemini-II represents a speculative bet on associative computing becoming a meaningful alternative to GPU-based AI inference for specific workloads.

---

1. Core Business Model & How They Work

GSI's core logic: mature, defensible SRAM revenue subsidizes R&D on a novel compute architecture that, if it gains design wins, could open a much larger addressable market in AI/edge inference.

[ SRAM Product Sales (SyncBurst, NBT, SigmaQuad, SigmaDDR; ~10,000 part numbers) ] ➡️ [ Funds R&D ] ➡️ [ Gemini APU Development (Gemini-I in production, Gemini-II sampling) ] ➡️ [ Hardware Sales + Cloud/SaaS + Future IP Licensing + Government SBIR Contracts ]

Key Operational Drivers

1. Legacy SRAM Franchise: Four SRAM product families, plus radiation-hardened (RadHard) and radiation-tolerant (RadTolerant) space-grade variants, serve radar, guidance, satellite, networking, telecom, and test-equipment customers who need extremely fast, reliable memory.
2. Gemini APU Pivot: Gemini-I ships today as the Leda-E (PCIe card) and Leda-S (1U card); Gemini-II reached first silicon and is sampled as the Leda-E2 card, targeting general-market launch and, per FY2026 results, early customer design-win conversations.
3. Government Funding as a De-Risking Tool: SBIR milestone-payment contracts from the Space Force, Air Force, and Army — including a ~$2M Army SBIR Phase II award for a ruggedized Gemini-II edge AI platform won after FY2026 year-end — fund APU development with non-dilutive capital (over $5.0M to date tied to edge AI applications).
4. Distribution-Heavy Sales Model: About 91.7% of fiscal 2025 net revenue flowed through distributors (Avnet, Holystone prominent), with direct sales to contract manufacturers making up the remainder.
5. High Customer Concentration: KYEC, a Taiwanese chip-testing/assembly customer, represented roughly 22-23% of recent quarterly/annual revenue; Nokia has also been a significant direct customer historically.

---

2. Business Segments

GSI does not report formal financial segments, but Item 1 of its 10-K organizes the business cleanly around two product lines.

┌─────────────────────────────────┐
│         GSI Technology, Inc.        │
└──────────────┬──────────────────┘
               │
   ┌───────────┴────────────┐
   ▼                         ▼
┌──────────────────┐  ┌───────────────────────────┐
│  SRAM Products      │  │  APU / Gemini Products      │
│  (SyncBurst, NBT,    │  │  (Gemini-I in production,   │
│   SigmaQuad,         │  │   Gemini-II sampling,       │
│   SigmaDDR, RadHard)  │  │   cloud/SaaS, future IP     │
│                      │  │   licensing)                │
└──────────────────┘  └───────────────────────────┘

---

3. Product Portfolio

---

4. Competitive Landscape

      SEMICONDUCTOR NICHE POSITIONING
┌──────────────────────────────────────────────────────┐
│ High │                                                   │
│  ▲   │                           [NVIDIA] [Intel]         │
│  A   │                           (AI/HPC compute,          │
│  I   │                            vastly greater           │
│  /   │                            resources & scale)        │
│  C   │        [GSI Gemini]                                 │
│  O   │        (niche associative                           │
│  M   │         computing architecture)                      │
│  P   │                                                     │
│  U   │  [Infineon] [ISSI] [Renesas]                         │
│  T   │  (SRAM competitors — GSI's                           │
│  E   │   established, defensible niche)                      │
│ Low  │                                                     │
│      └───────────────────────────────────────────────────► │
│       Low            SCALE / RESOURCES              High   │
└──────────────────────────────────────────────────────┘

• SRAM competitors: Infineon, Integrated Silicon Solution (ISSI), and Renesas — larger, often vertically integrated competitors, though GSI has decades of specialization in very-fast SRAM.
• APU competitors: GSI explicitly names NVIDIA and Intel as principal rivals in AI compute, acknowledging both have "greater resources, their own fabs, and longer OEM relationships." GSI expects more entrants into associative computing as the category matures.
• Basis of competition: product performance, power efficiency, quality, price, software tooling (compilers), speed to market, and standards conformance.
• GSI itself acknowledges in its filing that it has "limited experience developing products beyond Very Fast SRAMs" — a candid admission of execution risk in the Gemini pivot.

---

5. Strategic Strengths & Risks

Strengths (The Moat)

• Decades of SRAM design expertise: A ~10,000-part-number catalog and established qualification in defense/aerospace (RadHard/RadTolerant) represents real, hard-to-replicate engineering depth and customer trust built over decades.
• Non-dilutive government funding for R&D: SBIR milestone contracts from the Space Force, Air Force, and Army let GSI de-risk Gemini-II development without diluting shareholders as heavily as pure equity funding would.
• Differentiated compute architecture: Associative computing targets power-efficiency advantages ("performance-per-watt," per third-party validation cited by the company) relevant to edge and defense applications where NVIDIA/Intel GPUs are often too power-hungry.

Risks

• Execution risk on an unproven architecture: Gemini-II remains pre-revenue in any meaningful sense — FY2026 results describe "initial customer engagements" rather than material design wins, and management itself frames FY2027 as the year to convert engagements into revenue.
• Customer concentration: KYEC alone represented roughly 22-23% of recent revenue; loss of this or Nokia's business would materially affect top-line results.
• Persistent net losses: GSI posted a $13.2M net loss in FY2026 (on just $25.1M of revenue) and a $17.5M operating loss, with R&D spending ($19.9M, largely Gemini/Plato chip design) exceeding total revenue.
• Competing against vastly better-resourced rivals: NVIDIA and Intel dwarf GSI in R&D budget, fab access, and OEM relationships — GSI's APU bet depends on carving out specific workloads (power-constrained edge/defense AI) where its architecture has a genuine edge rather than competing head-on.

---

6. Financial Overview

---

7. Summary Conclusion

GSI Technology is, in effect, two businesses in one: a mature, defensible SRAM franchise generating steady demand from defense, aerospace, and networking customers, and an early-stage, high-R&D bet on associative computing (Gemini) as a power-efficient alternative to GPU-based AI inference at the edge. The SRAM business funds the company's survival and credibility; the Gemini-II architecture, still pre-revenue in any material sense but gaining non-dilutive government funding and early design-win conversations, represents the call option that could redefine the company's growth trajectory if it converts "initial customer engagements" into real orders during fiscal 2027 — the explicit milestone management itself has set.