AIDIMM™ & AILPBGA™ Make Global Debut | Longsys Spotlights Full-Stack Edge AI Storage Solutions at COMPUTEX 2026

TAIPEI, June 4, 2026 /PRNewswire/ -- From June 2 to 5, 2026, COMPUTEX 2026 in Taipei runs under the theme "AI Together," focusing on AI computing and next-gen technologies. Longsys (301308.SZ), under its own theme "Edge AI Storage • Integrated Implementation," showcases its full-stack edge AI storage portfolio and multi-scenario integrated application solutions. Together with its globally recognized high-end consumer brand Lexar, Longsys demonstrates a comprehensive range of innovative edge AI storage hardware and software designed to optimize local LLM performance and accelerate collaboration across the global AI hardware ecosystem.

Global Launch: AIDIMM™ & AILPBGA™ Custom-Built Memory for Edge AI Inference

At the show, Longsys highlights two dedicated memory products purpose-built for edge AI inference, precisely tailored to meet the deployment needs of AI models across various scenarios.

AIDIMM™: Single-Module Memory Enables Stable 70B+ Edge LLM Operation

Optimized specifically for intensive AI computing, AIDIMM™ delivers up to 128GB capacity, a native 256-bit bus, and industry-leading 307.2GB/s single-channel bandwidth in a compact form factor. It resolves common pain points of today's AI agent hosts—including insufficient memory, compute bottlenecks causing stuttering, thermal throttling, and costly upgrades. The module features a single-sided layout with four LPDDR5x chips, simplifying PCB routing, along with high-density, high-pin-count connectors that are natively compatible with mainstream AI agent host motherboards. This allows customers to upgrade without extensive hardware redesign, significantly reducing system costs.

Under demanding conditions such as real-time LLM inference, frequent AI agent computation, and multi-task concurrent interactions, AIDIMM™'s ultra-high bandwidth slashes data transmission latency, reduces idle compute cycles and prevents laggy responses. A single module reliably and smoothly runs edge LLMs with over 70 billion parameters. Its efficient integrated thermal design maintains stable temperatures and avoids performance throttling even in densely packed AI agent host deployments.

Supporting dynamic voltage regulation from 0.9V to 1.05V and featuring proprietary FDVFS intelligent power tuning, AIDIMM™ dynamically adjusts voltage and operating states based on varying AI workloads to deliver fine-grained power management. This superior energy efficiency reduces overall heat output, providing AI PCs and AI agent hosts with a high-performance, low-power, easy-to-upgrade memory solution that unlocks full edge AI compute potential. It also enables more power-efficient and stable long-term mass deployment.

AILPBGA™: High-bandwidth memory chip with native LPDDR compatibility for compact embedded AI inference

Developed specifically for space-constrained embedded edge AI inference environments, AILPBGA™ centers around three core advantages: outstanding cost-effectiveness, broad compatibility, and ultra-low power consumption. Built on Longsys's proprietary architecture, the chip features a native 256-bit bus delivering up to 307GB/s bandwidth and configurable capacity from 24GB to 64GB, with full native LPDDR interface compatibility. Housed in a compact 22×22mm package with high integration density, AILPBGA™ fits seamlessly into space-limited embedded terminals for AI inference and lightweight to mid-sized LLM deployment.

Compared to cloud AI high-bandwidth memory, AILPBGA™ strikes a balance between cost and power consumption, delivering superior cost-efficiency for edge AI inference and helping customers reduce costs while improving performance. Versus standard LPDDR5x solutions, it offers multiple times the bus width, performance, and capacity while remaining compatible with existing LPDDR platforms—enabling simple PCB routing and eliminating the need for SoC or system architecture redesign. This significantly shortens development cycles and lowers adaptation costs.

Its low-power architecture cuts terminal power draw to extend operational runtime for embedded inference and edge LLM hardware, reduces internal heat generation and simplifies internal thermal component layout for compact devices, mitigating hardware failure risks stemming from overheating. Over large-scale commercial deployments, the chip's energy-saving properties further lower long-term operational and maintenance costs for end users.

Hardware-Software Storage Integration for Edge AI Compute-Storage Applications

Longsys also demonstrates its vertically integrated hardware-software storage stack engineered to streamline on-device LLM operation across diversified edge scenarios.

SPU™ + iSA™ for AI agent edge deployments

In AI agent edge scenarios, HLCache™ technology is deeply integrated into Longsys's proprietary SPU™ (Storage Processing Unit), effectively reducing terminal DRAM usage and hardware costs. Serving as the brain of SPU™, the self-developed iSA™ (Intelligence Storage Agent) is a professional scheduling engine for edge AI inference. Targeting the pain points of MoE LLMs—large parameter counts, rapid KV Cache expansion, and I/O latency that hampers inference efficiency—it leverages expert offloading, intelligent cache management, and predictive prefetch algorithms to efficiently resolve storage scheduling challenges and comprehensively improve local AI inference smoothness.

A live on-site demonstration using an AI agent host powered by the AMD Ryzen AI Max+ 395 processor validated these gains: with 128GB of DRAM, a massive 397B-parameter LLM was deployed locally; with 64GB of DRAM, mid- to large-scale 80B and 122B LLMs ran smoothly with optimized long-context support. The solution effectively reduces the high memory requirements of edge AI, significantly improving runtime efficiency and cost-effectiveness.

HLCache™-enabled UFS: Accelerated AI performance for mobile edge terminals

For mobile edge AI applications, Longsys introduced UFS products integrated with HLCache™ technology, which significantly improve DRAM scheduling efficiency. Live comparative demonstrations using smartphones with different DRAM capacities clearly validated the enhancement in mobile edge AI interaction speed, enabling mobile terminals to smoothly run lightweight 13B and 20B AI models. This solution delivers a smooth experience comparable to that of devices with much larger DRAM while using lower-spec memory, effectively reducing DRAM usage in terminals. It ensures smooth operation, extends hardware lifespan, and optimizes overall device BOM costs.

Looking ahead, the two combinations—SPU™+iSA™+AIDIMM™ for AI agent edge AI storage, and UFS+HLCache™+AILPBGA™ for mobile edge AI storage—are expected to achieve unified compute-storage-acceleration synergy, better adapting to the loading, inference, and operational needs of complex local LLMs on edge devices.

High-Speed mSSD Storage Media with Integrated Thermal Management for Storage Applications

Longsys showcases its full lineup of PCIe Gen4 and Gen5 high-speed mSSD products with real-time on-site performance benchmarking. Manufactured via advanced wafer-level SiP system-in-package technology, each mSSD integrates storage controller, NAND flash and PMIC into a single compact die for robust chip-grade reliability and flexible form-factor customization into M.2 SSDs, portable PSSDs and specialized AI memory cards catering to diverse terminal design specifications.

The PCIe Gen5 mSSD variant features a compact 20×30mm footprint compatible with standard M.2 2230 dimensions, paired with an optional M.2 2280 extended thermal bracket as displayed at the booth. Equipped with premium onboard controllers, it hits peak sequential read/write speeds of 11GB/s and 10GB/s alongside maximum random 4K read/write IOPS of 2,200K and 1,800K, with single-drive capacity expandable up to 8TB to satisfy high-throughput storage and large-model repository needs of next-gen AI PCs. Incorporating proprietary VC phase-change liquid cooling paired with multi-layer stacked thermal architecture, the Gen5 SSD sustains prolonged peak performance under intense KV Cache-driven AI workloads, balancing ultra-high transfer speeds, slim device chassis design, low thermal rise and long-term operational stability.

Meanwhile, Longsys' mature PCIe Gen4 mSSD lineup has achieved commercial deployment. The products have secured collaborations with multiple PC OEMs and are widely integrated into a range of AI PCs and thin-and-light laptops. With stable performance and proven reliability, they have earned strong market and industry recognition.

Global Expansion and Brand Power Drive Edge AI Adoption

Built upon Longsys's proprietary AI Storage Core architecture derived from its high-speed mSSD storage media, Lexar debuts its next-generation AI-Grade Gen5 professional storage lineup at COMPUTEX. Powered by the exclusive Lexar AI Storage Solution, the new product family improves edge AI operational efficiency while reducing DRAM capacity requirements for end devices, and is well-suited for AI PCs, smart imaging equipment, and consumer robotics applications.

Marking Lexar's milestone 30th brand anniversary and ahead of the upcoming USA-Canada-Mexico FIFA World Cup, the iconic storage brand unveils limited-edition co-branded PSSDs and USB drives developed in partnership with Argentina National Football Team, alongside its full catalog of high-capacity SSDs, portable solid-state drives and SD/microSD memory cards to enrich its consumer storage portfolio.

Founded in 1996, Lexar maintains a robust global presence spanning six continents with established distribution channels covering tier-one global retailers including Costco and Best Buy plus mainstream cross-border e-commerce platforms. Core edge AI storage technologies and products unveiled at COMPUTEX will be gradually commercialized worldwide through Lexar's global channel advantages, delivering high-performance, reliable storage solutions for global AI content creation, edge inference, and mobile computing scenarios.

Looking ahead, Longsys will continue to deepen its focus on edge AI storage, leveraging its full-stack storage foundry capabilities to continuously iterate products and solutions, optimize the local LLM experience, and drive multi-scenario deployments—delivering mature edge AI storage solutions that are both practical and industry-benchmarking.

About Longsys

Founded in 1999, Longsys (301308.SZ) is a globally leading branded semiconductor memory enterprise, integrating R&D, design, packaging and testing, manufacturing, and sales services. For more information please visit https://www.longsys.com/, and follow Longsys on LinkedIn and Facebook.

View original content to download multimedia:https://www.prnewswire.com/apac/news-releases/aidimm--ailpbga-make-global-debut--longsys-spotlights-full-stack-edge-ai-storage-solutions-at-computex-2026-302791265.html

SOURCE Longsys