GSI (GSIT) 16th Annual LD Micro Invitational Conference summary

Company overview and financials

• Over 30 years of experience in high-performance SRAM, with a strong partnership with TSMC for manufacturing.

• Self-funded $175 million in R&D for AI product development, with $47 million raised in October last year.

• 126 employees, primarily design and software engineers; $67 million in cash and no debt.

• Insider ownership exceeds 20%, including management and original angel funders.

• Year-over-year revenue growth of 20-22%, with trailing 12-month revenue at $25 million, mostly from SRAM.

AI product line and technology

• Focused on edge AI markets, not data centers, targeting applications like autonomous drones, satellites, and smart cities.

• Developed an Associative Processing Unit (APU) using compute-in-memory (CIM) architecture, reducing data movement and power consumption.

• Cornell study showed Gemini-I board used 98% less power than an NVIDIA GPU for RAG applications.

• Won POCs for drone surveillance and smart city deployments, demonstrating superior performance per watt.

• Aggressively filing patents, with 147 total and 88 related to the APU.

Product roadmap and applications

• Gemini-II chip powers current POCs, including a Department of Defense drone surveillance system and a smart city project in Taiwan.

• Smart city POC phases: 20 cameras (phase I), 80 cameras with audio (phase II), and a potential 6,000 camera deployment (phase III) in 2025-2027.

• Next-gen Plato chip, designed for LLMs, will be smaller, lower power (10W), and ready by March/April next year.

• Plato targets true edge applications, not data centers, with further miniaturization and power reduction.

• Future products may address data center needs, with discussions underway with funding partners.