TaipeiJune 3, 2026 /PRNewswire/ — From June 2 to 5, 2026, Longsys showcased its full-stack edge AI storage new products and comprehensive application solutions at COMPUTEX 2026 in Taipei. With the theme “AI Together,” the exhibition focused on core directions such as AI and computing, and next-generation technologies. Centered around “Edge AI Storage • Comprehensive Applications,” Longsys highlighted new AI memory products, full-chain technical solutions, and multi-scenario combined applications. Leveraging the global brand advantages of its subsidiary Lexar, the company comprehensively presented innovative achievements in the edge AI storage field, aiming to optimize the local model experience of edge AI and promote collaborative development within the industry ecosystem.
AIDIMM™ & AILPBGA™ Two New AI Memory Products Debut, Precisely Tailored for Edge AI Inference Applications
At this exhibition, Longsys focused on launching two specialized memory products designed for edge AI inference, precisely matching the deployment needs of AI models in various scenarios.
AIDIMM™: Stably Supports 70B+ Edge AI Large Models with a Single Module
AIDIMM™, as a memory deeply optimized for AI computing, effectively addresses common challenges faced by current AI agent hosts, such as insufficient memory capacity, lagging computing tasks, heat dissipation difficulties, and inconvenient upgrades. It offers up to 128GB capacity, 256-bit bus width, 307.2GB/s single-channel ultra-high bandwidth, and a compact form factor. The product adopts a design with four LPDDR5x chips on the same side, simplifying wiring. The high-speed, high-density, high-pin-count connector is compatible with the motherboard architecture of mainstream AI agent hosts, requiring no major hardware modifications, thereby reducing customers’ hardware upgrade costs.
Under high-intensity workloads involving frequent AI agent computations, real-time large model inference, and multi-scenario synchronous interactions, AIDIMM™’s high-speed bandwidth quickly responds to computing resource scheduling demands, significantly reducing data transmission latency. This helps alleviate issues like idle computing power, task lag, and delayed model responses. A single module can stably support the smooth operation of 70B+ level edge large models.
Additionally, AIDIMM™ is paired with an efficient heat dissipation structure, enabling precise temperature control in high-density deployment scenarios of AI agent hosts to avoid performance throttling, ensuring both high performance and operational stability. The product supports dynamic voltage adjustment from 0.9V to 1.05V and features an FDVFS intelligent energy efficiency optimization mechanism. This allows intelligent voltage and operating state adjustments based on different workloads, such as edge AI inference and large model operation, achieving fine-grained dynamic power management under AI loads. It effectively improves the overall energy efficiency ratio of edge AI scenarios, significantly reduces heat generation during operation, and provides AI PCs and AI agent hosts with high-performance, low-power, and easily upgradeable AI memory solutions, fully unleashing terminal AI computing power for more energy-efficient and stable long-term batch deployments.
AILPBGA™: High-Bandwidth Memory Chip Compatible with LPDDR Interface
AILPBGA™ focuses on embedded AI inference scenarios requiring a compact form factor, offering core advantages of “high efficiency, high adaptability, and low power consumption.” The product adopts proprietary technical standards and an innovative architecture, featuring a native 256-bit bus width per chip, bandwidth up to 307GB/s, and capacity ranging from 24GB to 64GB, fully compatible with the LPDDR interface. It uses a compact 22×22mm package design, offering small size and high integration, making it flexibly adaptable to compact terminals for AI inference and medium-to-lightweight model deployment.
Compared to cloud AI high-bandwidth memory, AILPBGA™ prioritizes a balance between cost and power consumption, meeting edge AI inference needs with higher cost-effectiveness, helping customers reduce costs and improve efficiency. Compared to standard LPDDR5x, it offers several times higher bus width, performance, and capacity, while remaining compatible with existing LPDDR platforms. Its simple wiring eliminates the need to redesign SoC and system architectures, significantly shortening terminal R&D and deployment cycles and reducing adaptation costs.
AILPBGA™ adopts a low-power design, which not only substantially reduces device energy consumption, effectively extending the battery life of AI inference terminals and edge large model devices, but also significantly lowers heat generation, simplifying heat dissipation structure design to meet the spatial requirements of compact devices. It also enhances long-term operational stability, preventing high-temperature-related failures. As AI applications become widespread, its low-power characteristics enable efficient energy savings, helping users effectively reduce overall electricity and maintenance costs.
Software and Hardware Storage Technology Layout, Building Comprehensive Edge AI Storage and Computing Applications
In terms of technical applications, Longsys showcased storage solutions ranging from chip hardware to intelligent software, comprehensively optimizing the local model operation experience for edge AI.
SPU™ + iSA™ Storage Agent Application
In agent-side AI application scenarios, the HLCache™ technology is deeply integrated into the SPU™ Storage Processing Unit, effectively reducing terminal DRAM usage and hardware costs. The iSA™ Storage Agent, serving as the brain of the SPU™, is a specialized scheduling engine for edge AI inference. It addresses pain points such as large parameter counts in MoE large models, rapid KV Cache expansion, and I/O latency affecting inference efficiency. By utilizing expert offloading, intelligent cache management, and smart prefetching algorithms, it efficiently solves storage scheduling challenges for running large models on edge devices, comprehensively improving the smoothness of local AI inference. With this optimization solution, a live demonstration was conducted on an AI agent host based on the AMD Ryzen AI Max+ 395 processor: 128GB of memory enabled local deployment of a 397B-parameter ultra-large AI model, while 64GB of memory smoothly ran medium-to-large models like 122B and 80B. It also optimized the long-context user experience, effectively alleviating high memory usage issues in edge AI, significantly enhancing edge AI operational efficiency and cost-effectiveness.
UFS + HLCache™ Technology Application
In the mobile edge AI field, Longsys introduced UFS products equipped with HLCache™ technology, which significantly improves DRAM scheduling efficiency. Comparative demonstrations on phones with different DRAM capacities clearly validated its effectiveness in enhancing mobile edge AI interaction efficiency, enabling mobile terminals to smoothly run lightweight AI models like 13B and 20B. This solution achieves a smooth operation experience comparable to larger memory configurations using lower-spec memory, effectively reducing terminal DRAM usage. While ensuring smooth device operation and extending hardware lifespan, it optimizes the overall BOM cost of terminal devices.
In the future, the two combinations—agent-side edge AI storage SPU™+iSA™+AIDIMM™ and mobile edge AI storage UFS+HLCache™+AILPBGA™—are expected to achieve integrated “storage-computing-acceleration” synergy, better adapting to the loading, inference, and operational needs of complex local large models on edge devices.
mSSD High-Speed Storage Media, Integrated Heat Dissipation Storage Applications
At this exhibition, Gen4 and Gen5 full-series mSSD high-speed storage media were showcased with live performance testing. All mSSDs adopt wafer-level SiP (System-in-Package) technology, integrating the controller, NAND, and power management chip into one unit. They offer chip-level reliability, a compact form factor, and flexible form factor expandability, enabling derivatives such as M.2 SSDs, PSSDs, and AI storage cards, suitable for diverse terminal design needs.
Among them, the PCIe Gen5 mSSD features a 20×30mm small size, natively compatible with the M.2 2230 specification. A live display included an M.2 2280 heat dissipation expansion card and an overall heat dissipation structure. The product is equipped with a high-performance controller, achieving sequential read/write speeds of up to 11GB/s and 10GB/s, 4K random read/write of up to 2200K and 1800K IOPS, and a single-drive maximum capacity of 8TB, precisely matching the high-throughput and large model storage needs of AI PCs. In terms of thermal engineering, the product features a dedicated VC (Vapor Chamber) phase-change liquid cooling solution, combined with a multi-layer thermal conduction structure, significantly extending the duration of peak performance output. This is fully suitable for high-load KV Cache operation scenarios in AI PCs, balancing extreme high-speed read/write with slim device design, achieving high performance, low temperature rise, and high stability.
Alongside the Gen5 iteration, Longsys’s mature PCIe Gen4 mSSD has achieved commercial deployment. The product has already partnered with several PC host manufacturers and is widely used in various AI PCs and thin-and-light laptops, gaining broad market and industry recognition for its stable performance and reliability.
Global Layout Empowerment, Brand Driving Application Adoption
Based on the AI Storage Core technology architecture derived from mSSD high-speed storage media, Lexar focused on launching a new generation of AI-Grade Gen5 professional storage products at this exhibition. By incorporating the Lexar AI Storage Solution, these products significantly enhance the operational efficiency of edge AI applications and effectively reduce terminal DRAM capacity requirements, widely adapting to cutting-edge application scenarios such as AI PCs, intelligent imaging, and smart robots.
Coinciding with the 30th anniversary of the Lexar brand and the upcoming World Cup, co-branded products with the Argentine national team have also garnered attention. The brand partnered with the Argentine national team to launch a co-branded series of PSSDs and USBs, and presented a full product line including high-capacity SSDs, PSSDs, and memory cards, further enriching its consumer storage portfolio.
In terms of global expansion, Longsys, through its international high-end consumer storage brand Lexar, continues to expand the global influence of edge AI storage technology. As an internationally renowned brand founded in 1996, Lexar boasts a full-category product portfolio and global channels spanning six continents, having entered top global retail channels such as Costco and BestBuy, as well as major e-commerce platforms. Some of the core edge AI storage technologies and products launched at this exhibition will be simultaneously introduced to the global market through Lexar’s global channel advantages, providing high-performance, high-reliability storage solutions for global AI content creation, edge inference, and mobile computing scenarios.
Longsys will continue to deeply cultivate the edge AI storage field, leveraging its full-chain storage Foundry capabilities to continuously iterate products and solutions, optimize the local large model operation experience, and create mature edge AI storage application solutions that are both practical and industry-referable through diversified scenario implementation practices.