GSI Technology (NASDAQ:GSIT) is positioning its associative processing unit, or APU, technology for artificial intelligence workloads at the edge rather than in data centers, Vice President of Sales and Investor Relations Didier Lasserre said during a company presentation.
Lasserre, who said he has been with the company for nearly 28 years, described GSI as a long-standing SRAM memory supplier that is using its memory expertise to develop AI-focused compute-in-memory products. He said the company has self-funded $175 million in research and development for its APU development, with its profitable SRAM business helping offset those costs.
GSI reported trailing 12-month revenue of about $25 million, with most of that coming from SRAM, representing a roughly 20% to 22% year-over-year increase, according to Lasserre. He said the company has 126 employees worldwide, just over $67 million in cash and no debt. GSI also raised net proceeds of $47 million in October, which Lasserre said removed the need for a previously considered strategic funding initiative.
Edge AI Focus and Compute-in-Memory Strategy
Lasserre said GSI is not targeting the data center market currently dominated by companies such as NVIDIA. Instead, he said the company is focused on edge applications where power budgets are limited and data movement can create latency and power-consumption challenges.
GSI’s APU approach is based on compute-in-memory, which Lasserre said differs from “near memory compute” approaches that move processing and memory closer together but still require data transfers. In GSI’s architecture, he said data resides where the processing elements are and computations are performed in the memory array itself.
He cited a Cornell study comparing GSI’s Gemini-I board with an NVIDIA GPU in a retrieval-augmented generation, or RAG, application. Lasserre said that at the same performance level, Gemini-I used 98% less power than the GPU.
“Because we do the compute and memory, we are not moving data,” Lasserre said.
Drone, Smart City and Defense Applications
Lasserre described several proof-of-concept projects and government-funded programs that GSI is pursuing with its Gemini-II product. One drone surveillance application, he said, involved a customer that had evaluated NVIDIA’s Jetson family and Qualcomm’s Snapdragon platform. According to Lasserre, Jetson met a three-second “time to first token” performance threshold but used 160 watts, above the system’s sub-50-watt requirement. Snapdragon met the power target but took 12 seconds.
GSI’s solution met both requirements, Lasserre said, initially achieving three seconds and later improving to 2.7 seconds, with the company aiming to get below 2.5 seconds. He said GSI won the bake-off and is now the partner for the drone activity.
Lasserre also discussed a smart city proof of concept in Taiwan that GSI announced recently. Phase I will use 20 existing cameras in a county to identify events such as fires, riots or car accidents and recommend actions. Phase II would expand to 80 cameras and add audio, with possible school deployments to detect physical and verbal abuse. Lasserre said a potential Phase III in 2027 could involve a full-scale 6,000-camera system, and noted that one Gemini-II chip controls four cameras.
GSI also has three active Small Business Innovation Research, or SBIR, awards. Lasserre highlighted a $2 million Phase II award from the U.S. Army to create a ruggedized edge node for object detection, synthetic aperture radar and other applications. Unlike prior SBIRs that served mainly as R&D offsets, he said this program could become a revenue-generating product.
Plato Chip Roadmap
GSI’s next-generation product, Plato, is being designed for lower-power edge AI and large language model applications, Lasserre said. The company began design work last year and expects completion around March or April of next year.
While Lasserre said Gemini-II is being used for some LLM applications, he said it was not designed specifically for that use case. Plato is intended to support LLM workloads by increasing bandwidth to DRAM, while lowering maximum chip power to 10 watts and reducing the chip size to about one-quarter the size of Gemini-II.
In response to an analyst question, Lasserre said Plato is not intended for data centers. He said the chip is meant to push GSI “further away to the edge,” including robotics and smaller drones. However, he added that GSI has begun discussions with funding partners about what comes after Plato, and one area of interest from those partners is a data center part.
SRAM Business and Radiation-Hardened Products
Lasserre said GSI’s SRAM business continues to fund the company and that the company has the highest performance and density in its market segment. He said GSI has frozen its SRAM roadmap to focus R&D on the APU family, but competitors have also frozen their roadmaps, leaving GSI with what he described as a one- to two-generation lead.
GSI has also developed radiation-tolerant and radiation-hardened SRAM products for space applications. Lasserre said the company’s highest-end standard SRAM sells for a few hundred dollars, while a radiation-tolerant version of the same density sells for a few thousand dollars and a radiation-hardened version can sell for tens of thousands of dollars, up to $30,000.
He said GSI expects its first production orders for these space-focused products during the current calendar year, though he noted that satellite deployment timelines can be slow.
Revenue Timing
During the question-and-answer session, an analyst asked when GSI expects revenue from its AI initiatives. Lasserre said the company expects some minor revenue this year from drone and smart city proof-of-concept work, with possible additional revenue if the smart city project moves to Phase II by year-end.
“Really, it’s 2027,” Lasserre said, pointing to the potential 6,000-camera deployment and a Department of Defense field demonstration expected at the end of this year. He characterized 2026 as a period of prototyping, with volume revenue expected in 2027.
About GSI Technology (NASDAQ:GSIT)
GSI Technology, Inc is a fabless semiconductor company specializing in the design and development of high-performance memory products. Headquartered in Sunnyvale, California, the company was founded in 1995 and has focused its efforts on content addressable memory (CAM) and high-speed SRAM (static random-access memory) solutions. As a publicly traded company listed on NASDAQ under the ticker GSIT, GSI Technology leverages advanced architectures to meet demanding data-processing requirements.
The company's core product portfolio includes ternary CAM (TCAM) devices, binary CAM (BCAM) devices and high-speed synchronous SRAM.
This instant news alert was generated by narrative science technology and financial data from MarketBeat in order to provide readers with the fastest reporting and unbiased coverage. Please send any questions or comments about this story to contact@marketbeat.com.
The article "GSI Technology Bets on Edge AI as APU Projects Point to 2027 Revenue Ramp" first appeared on MarketBeat.
