At COMPUTEX 2026 in Taipei, held from June 2 to 5 under the theme “AI Together,” Longsys showcased its latest advancements in edge AI storage technologies, presenting a full-stack portfolio designed to support next-generation AI computing workloads.
Together with its high-end consumer brand Lexar, the company demonstrated integrated hardware and software solutions aimed at optimizing local large language model (LLM) performance and accelerating ecosystem collaboration across global AI infrastructure development.
Positioning its exhibition around “Edge AI Storage • Integrated Implementation,” Longsys introduced two newly developed memory technologies—AIDIMM™ and AILPBGA™—both designed specifically for edge AI inference scenarios. The company emphasized that these innovations target growing demands for higher bandwidth, lower latency, and improved energy efficiency in AI agent devices, embedded systems, and compact computing platforms.
AIDIMM™ was presented as a high-performance single-module memory solution capable of supporting stable operation of edge LLMs exceeding 70 billion parameters. With up to 128GB capacity, a native 256-bit bus architecture, and reported single-channel bandwidth reaching 307.2GB/s, the module is engineered to reduce computational bottlenecks in AI workloads. Longsys highlighted its compact design, simplified PCB integration, and compatibility with mainstream AI host motherboards, enabling easier system upgrades without major hardware redesign. The module also incorporates dynamic voltage regulation and proprietary power management technology to optimize energy efficiency, thermal control, and sustained performance under continuous AI inference and multitasking conditions.
Alongside AIDIMM™, the company introduced AILPBGA™, a compact high-bandwidth memory chip tailored for space-constrained embedded AI applications. Featuring a 22×22mm package, configurable capacity from 24GB to 64GB, and a native 256-bit bus delivering up to 307GB/s bandwidth, the solution is designed for lightweight and mid-scale edge LLM deployments. Longsys positioned the chip as a cost-efficient and power-optimized alternative for embedded systems, offering broad LPDDR compatibility while significantly improving performance density compared with conventional LPDDR5x solutions. Its architecture is intended to shorten development cycles and reduce integration complexity for device manufacturers.
Overall, Longsys framed its COMPUTEX 2026 presentation as part of a broader push to advance edge AI computing infrastructure, where storage and memory systems are increasingly critical to enabling real-time intelligence, local inference, and scalable AI agent applications.