GSI (GSIT) 27th Annual Needham Growth Conference summary

Company overview and legacy business

• Founded in 1995, public since 2007, with manufacturing outsourced to TSMC and a strong engineering partnership.

• Legacy SRAM product line funds AI initiatives, with 136 patents and $150M invested in APU architecture over nine years.

• Revenue last fiscal year was just under $22M, with 122 employees after a recent reduction in force.

• High insider ownership at 27% and over $18M in cash, no debt, and a market cap exceeding $100M.

• SRAM products are sole-sourced in new designs, providing a competitive edge and stable revenue.

Market focus and product innovation

• Targeting radiation-hardened SRAMs for space, aiming for 15% of a $100M+ annual market with ASPs up to $30,000 and gross margins over 90%.

• Entering AI market with APU chips leveraging SRAM expertise, addressing AI and space semiconductor markets with high growth rates.

• Compute-in-memory (CIM) APU architecture offers true in-place processing, reducing power and increasing performance compared to GPUs/CPUs.

• Gemini-I and Gemini-II products focus on SAR, fast vector search, and edge data center applications.

• Roadmap includes Plato, targeting large language models (LLMs) with ultra-low power consumption.

Technology differentiation and scalability

• CIM architecture eliminates data transfer bottlenecks, enabling significant power savings and performance gains.

• Bit engine processor allows flexible bit-width configuration, supporting evolving AI model requirements.

• Scalable solution supports multiple boards and servers, adapting to growing database and AI model sizes.

• Gemini-II brings data center performance to the edge, addressing power and scalability challenges in AI.

• Plato aims to run LLMs like Llama 3.2 90B at around 10 watts, with rapid design turnaround using a 12nm process.