Longsys Showcases End-to-End AI Storage Solutions at FMS 2026

San Jose, Calif., August 4, 2026 — Longsys(301308.SZ) is presenting its "Edge AI Storage Fusion" showcase at the Future of Memory and Storage (FMS) 2026, spanning three core scenarios: AI Agent Hosts (AI BOX), AI PCs, and AI Mobile embedded devices—highlighting how tightly integrated memory and storage technologies are accelerating real-world Edge AI deployment and commercialization.

Co-Optimized with AMD: Integrated Hardware-Software Architecture Unlocks Greater Edge AI Performance

As large language models (LLMs) continue to move from the cloud to local devices, Edge AI systems face increasing challenges, including limited memory capacity, insufficient bandwidth, KV cache expansion, high inference latency, costly hardware upgrades, and reduced responsiveness during long-context interactions. To address these challenges, Longsys has collaborated with AMD to jointly optimize an integrated solution combining AIDIMM™ high-bandwidth memory, the iSA™ (Intelligent Storage Agent), and AISSD™. By tightly integrating hardware and software, the solution eliminates memory and storage bottlenecks for Edge AI inference while improving overall system performance and reducing deployment cost.

At FMS 2026, Longsys is demonstrating a live AI Agent Host optimized for the AMD platform. Jointly developed with SixUnion, the system delivers impressive real-world performance. With just 64GB of AIDIMM™ memory, the system runs mainstream large language models—ranging from 70B and 80B up to 122B parameters—locally, smoothly, and reliably. The result is a marked improvement in long-context generation, multi-turn dialogue, and complex content creation, alongside a significant reduction in memory usage and a sharp gain in inference efficiency. At the same time, it lowers hardware investment and operating costs, providing an optimal balance of performance, reliability, and cost efficiency for Edge AI deployment.

Longsys is also showcasing an AMD Ryzen™ AI Halo developer platform integrated with AISSD™ and the iSA™ (Intelligent Storage Agent). Through close collaboration between the two companies, the platform delivers seamless hardware-software integration and has successfully achieved AMD Approved Vendor List (AVL) compatibility certification. With proven compatibility and stable performance, the solution is ready for large-scale commercial deployment, enabling OEMs and system manufacturers to accelerate product development and expand the Edge AI ecosystem.

AIDIMM™: High-Bandwidth Memory Purpose-Built for AI Computing

Designed specifically for Edge AI inference, AIDIMM™ is Longsys' next-generation high-performance memory solution engineered to meet the demanding requirements of locally deployed large language models. The module integrates four LPDDR5X memory packages in a compact single-sided layout that minimizes signal loss while enabling a native 256-bit memory interface with peak bandwidth of up to 307.2 GB/s. Supporting capacities of up to 128GB per module, AIDIMM™ delivers significant improvements in bandwidth, interface width, and capacity compared with conventional memory modules. To simplify system integration, AIDIMM™ features a dedicated high-speed, high-pin-count connector compatible with mainstream AI Agent Host motherboard architectures. This allows customers to integrate the solution with minimal hardware redesign, reducing both development costs and time to market. Compared with the SOCAMM2 architecture, AIDIMM™ also offers a plug-and-play design, making it better suited for large-scale manufacturing, deployment, and field maintenance of Edge AI devices.

AISSD™ + iSA™ (Intelligent Storage Agent): Edge AI Intelligent Scheduling Engine

The iSA™ (Intelligent Storage Agent) is Longsys' proprietary scheduling engine purpose-built for Edge AI inference. Deeply optimized to work with AISSD™, it forms a tightly integrated hardware-software storage solution. Powered by three proprietary technologies—dynamic Mixture-of-Experts (MoE) offloading, intelligent hierarchical cache management, and predictive data prefetching—the solution dynamically allocates memory and storage resources in real time. By intelligently redistributing idle computing resources and cached data, it significantly reduces DRAM utilization while improving the responsiveness of local AI workloads, including conversational AI, content generation, and long-context text completion. The result is faster, more efficient, and more reliable on-device AI inference.

Next-Generation SSD Innovations Accelerate the Mainstream Adoption of High-Performance AI PCs

The rapid adoption of AI PCs is driving demand for local large language model (LLM) deployment and offline AI computing, placing increasingly stringent requirements on storage performance. Conventional SSDs often struggle to meet the sustained high-performance demands of Edge AI workloads due to limitations in throughput, thermal management, and effective capacity utilization.

To address these challenges, Longsys is showcasing a new generation of AI-optimized storage solutions built around its proprietary SPU™ (Storage Processing Unit) and high-speed PCIe Gen4 and Gen5 mSSD storage media. Combined with multiple in-house AI storage technologies, these innovations deliver comprehensive improvements in storage performance, efficiency, and scalability, enabling the mainstream, large-scale adoption of high-performance AI PCs..

SPU™ (Storage Processing Unit): Flagship Performance with Built-In Lossless Compression

At FMS 2026, Longsys is demonstrating a DRAM-less PCIe Gen5 SSD powered by its proprietary 5nm SPU™ (Storage Processing Unit). Live performance testing shows sequential read speeds of up to 14.8 GB/s and sequential write speeds of 13 GB/s, delivering flagship-class performance comparable to leading DRAM-equipped SSDs currently on the market.

In addition to high performance, the SPU™ architecture offers exceptional power efficiency. Operating at 6.3 W or below, it consumes approximately 10% less power than comparable high-performance SSD controllers that typically exceed 7 W. This combination of high throughput and lower power consumption makes it particularly well suited for demanding AI PC workloads such as local LLM loading, concurrent AI inference, and AI-assisted content creation, achieving an optimal balance between performance and energy efficiency.

Longsys is also showcasing the in-memory lossless compression technology integrated into the SPU™ architecture. Unlike conventional compression methods that consume CPU and system memory resources, the SPU™ performs intelligent, hardware-based compression directly within the SSD. It automatically identifies and compresses AI model files, cache data, office documents, multimedia content, and other storage-intensive workloads with a lossless compression ratio of up to 2:1, preserving both data integrity and storage performance. By significantly improving effective storage capacity without compromising read/write speeds, the technology helps reduce storage expansion costs while addressing the growing storage requirements of AI PCs running increasingly large AI models.

mSSD High-Speed Storage Media for Edge AI: Scalable Manufacturing with Global Supply Capability

To meet the storage requirements of mainstream high-performance AI PCs, Longsys is showcasing both PCIe Gen5 and PCIe Gen4 mSSD storage media solutions. Built on a next-generation high-speed interface and advanced SiP (System-in-Package) technology, the PCIe Gen5 mSSD delivers sequential read and write speeds of up to 11 GB/s and 10 GB/s, respectively, while achieving 4K random read/write performance of up to 2200K and 1800K IOPS. The solution is capable of sustaining peak performance for 181 seconds under continuous workloads, fully unleashing the bandwidth potential of the PCIe Gen5 interface.

Designed for maximum flexibility, the mSSD natively supports the M.2 2230 form factor and can be readily adapted to additional formats including M.2 2242, M.2 2280, AI storage cards, solid-state storage cards, and portable SSDs (PSSDs). This flexibility enables deployment across a broad range of AI PCs, ultra-thin notebooks, and high-performance creator laptops, supporting workloads such as offline AI productivity, AI-assisted image generation, AI video rendering, and local LLM inference. To support large-scale commercialization, Longsys has established a stable manufacturing capacity of more than one million units per month. The company has also built strategic partnerships with leading PC manufacturers, including Lenovo and ASUS, ensuring reliable, high-quality supply for global OEM customers.

High-Bandwidth Memory and Storage Synergy Unlocks Greater Performance for AI Mobile

Smartphones and portable embedded AI devices are constrained by limited board space, power budgets, and BOM costs, making it difficult to integrate large-capacity DRAM. These limitations often restrict on-device AI model deployment, reduce system responsiveness, and increase hardware resource utilization. To address these challenges, Longsys has introduced an integrated solution combining HLCache™ UFS with AILPBGA™ high-bandwidth embedded memory. Leveraging software-defined storage optimization, HLCache™ intelligently reduces DRAM requirements while AILPBGA™ provides the high-bandwidth memory performance required for AI workloads. Together, they lower the hardware barrier for Edge AI deployment and accelerate the adoption of AI capabilities across mobile and embedded devices.

HLCache™ UFS: Redefining Mobile Memory Efficiency

Longsys' HLCache™ technology enhances UFS storage by intelligently optimizing DRAM utilization. Frequently accessed ("hot") data remains in memory, while infrequently accessed ("cold") data is dynamically migrated to UFS storage, reducing DRAM usage without compromising overall system responsiveness.

At FMS 2026, Longsys demonstrated the technology using a Google Pixel 7a. With HLCache™ enabled, a device equipped with 4 GB of LPDDR5 memory delivered significantly improved real-world performance. Application launch times were noticeably reduced across 30 commonly used applications, while the number of applications that could remain active in the background increased from 17 to 28, representing an improvement of approximately 64%. Meanwhile, DRAM utilization dropped from 3.1 GB to 2.5 GB, a reduction of approximately 20%, resulting in much smoother multitasking and application switching. Overall user experience approached that of a native 6 GB LPDDR5 configuration.

Based on internal validation, HLCache™ enables mobile devices equipped with just 4 GB of DRAM to deliver performance comparable to devices with 6–8 GB of memory. The solution supports smooth local execution of 13B- and 20B-parameter large language models while reducing DRAM requirements, lowering overall BOM costs, decreasing memory wear caused by frequent read/write operations, and extending device lifespan.

AILPBGA™: High-Bandwidth Embedded Memory Purpose-Built for AI Devices

AILPBGA™ is Longsys' high-bandwidth embedded memory solution designed specifically for embedded AI inference platforms and portable intelligent devices. Built on a proprietary architecture and innovative package design, the solution delivers a native 256-bit memory interface with peak bandwidth of up to 307 GB/s per package. Available in capacities ranging from 24 GB to 64 GB, AILPBGA™ is fully compatible with the industry-standard LPDDR5X interface, simplifying integration into existing system designs. The solution adopts a compact 22 × 22 mm BGA1764 package, providing high integration while minimizing motherboard footprint. Optimized for space-constrained AI devices and portable inference platforms, AILPBGA™ combines high bandwidth, low power consumption, system stability, and design flexibility to overcome the key limitations of embedded AI systems, including constrained computing performance, limited board space, and stringent power requirements.

A Global Multi-Brand Strategy & Localized Delivery Capabilities

With a global operational footprint, Longsys has built a multi-brand strategy and localized supply chain network to address diverse customer needs worldwide. Longsys focuses on core storage technologies through continuous innovation, while Lexar serves the global consumer storage market with an established brand and channel network. Through its wholly owned subsidiary Zilia, Longsys provides localized storage manufacturing, testing, and delivery capabilities in global markets. As Longsys' South America manufacturing and operations center, Zilia enables localized production, faster delivery, and enhanced technical support, helping accelerate the global commercialization of Longsys' Edge AI storage solutions.

Looking ahead, Longsys will continue advancing its proprietary technologies and collaborating with global ecosystem partners to deliver efficient, scalable, and customized Edge AI storage solutions, accelerating the adoption of intelligent computing.

Product performance data is based on Longsys internal testing. Actual performance may vary depending on system configuration and operating conditions.

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.

Longsys media inquiries:

Richard Chang: richard.chang@lexarcorp.com

Vivian Wen: vivian.wen@longsys.com