As AI workloads increasingly shift from cloud platforms to edge devices, memory bandwidth and storage efficiency have become critical constraints for deploying large language models (LLMs). At Future of Memory and Storage (FMS) 2026, Longsys unveiled a new portfolio of edge AI storage technologies designed to improve AI inference across AI agent hosts, AI PCs, and mobile devices through optimized memory architecture, intelligent storage management, and high-performance SSD innovations.
The rapid growth of edge AI is driving renewed innovation in memory and storage technologies as enterprises seek to run increasingly sophisticated AI models closer to end users. At FMS 2026, Chinese memory and storage provider Longsys Technology Co., Ltd. introduced its latest “Edge AI Storage Fusion” portfolio, highlighting new hardware and software technologies aimed at overcoming some of the biggest performance bottlenecks in local AI deployment.
Rather than focusing solely on faster storage devices, Longsys is positioning itself as an integrated edge AI infrastructure provider by combining high-bandwidth memory, intelligent storage orchestration, AI-optimized SSDs, and embedded memory technologies into a unified ecosystem.
The company’s showcase spans three major deployment scenarios—AI agent hosts (AI BOX), AI PCs, and AI mobile devices—reflecting growing industry demand for on-device AI processing that reduces dependence on cloud infrastructure while improving privacy, responsiveness, and operational efficiency.
AI Agent Hosts Target Larger Local AI Models
One of the event’s most notable demonstrations centered on deploying large language models locally using optimized memory and storage architectures.
Longsys announced a collaboration with AMD that integrates its AIDIMM™ high-bandwidth memory, iSA™ (Intelligent Storage Agent) software, and AISSD™ storage platform into systems powered by AMD Ryzen™ AI Halo processors.
Working alongside SixUnion, the companies demonstrated systems capable of running 70-billion, 80-billion, and 122-billion parameter AI models using only 64GB of AIDIMM memory. According to Longsys, intelligent memory scheduling reduced overall memory consumption while improving inference efficiency compared with conventional deployment approaches.
Instead of relying entirely on expensive DRAM expansion, the proprietary iSA™ engine dynamically balances workloads between memory and storage using technologies including Mixture-of-Experts (MoE) offloading, predictive prefetching, and tiered cache management.
Longsys says its AIDIMM™ modules provide a native 256-bit memory interface, peak bandwidth of 307.2 GB/s, and capacities of up to 128GB per module while maintaining compatibility with standard manufacturing processes through a plug-and-play design.
The company also revealed that its AMD Ryzen AI Halo platform integrating AISSD™ and iSA™ has achieved AMD AVL (Approved Vendor List) certification, potentially simplifying commercial deployment for enterprise system builders.
AI PCs Gain Faster SSD Performance
Beyond AI servers, Longsys introduced several storage technologies aimed at AI-enabled personal computers.
The company’s new 5nm SPU™ (Storage Processing Unit) powers a DRAM-less PCIe Gen5 SSD capable of sequential read speeds of 14.8 GB/s and write speeds of 13 GB/s while operating at 6.3 watts or less. According to Longsys, this represents approximately 10% lower power consumption than comparable SSD controllers while maintaining flagship-class performance.
The SSD also incorporates built-in lossless compression technology that can expand usable storage capacity by up to 2:1 without reducing transfer speeds, a capability that could become increasingly valuable as AI applications require larger local datasets and model files.
Longsys also showcased a PCIe Gen5 mSSD delivering up to 11 GB/s read, 10 GB/s write, and 2.2 million read IOPS, supporting multiple form factors including M.2 2230, 2242, and 2280. The company says manufacturing capacity now exceeds one million units per month, with products already supplied to PC manufacturers including Lenovo and ASUS.
AI Mobile Focuses on Memory Optimization
Edge AI adoption is also expanding rapidly within smartphones and embedded systems, where physical space and battery life remain significant constraints.
Longsys introduced HLCache™, a Universal Flash Storage (UFS) technology that intelligently moves inactive application data from DRAM into storage to free system memory for AI workloads.
Internal testing on a Google Pixel 7a equipped with 4GB LPDDR5 memory showed background application capacity increasing from 17 to 28 applications, while DRAM utilization decreased by approximately 20%. According to Longsys, the optimized configuration delivers performance approaching that of native 6GB memory devices while supporting 13-billion to 20-billion parameter AI models.
Complementing the technology is AILPBGA™, an embedded memory package featuring a 256-bit interface and 307 GB/s bandwidth within a compact 22 × 22 mm footprint designed for smartphones, embedded AI hardware, and other space-constrained devices.
Growing Demand for Edge AI Infrastructure
Longsys’ latest product announcements reflect broader industry trends as enterprises increasingly deploy AI inference outside centralized cloud environments. Companies including NVIDIA, AMD, Intel, Qualcomm, Microsoft, and Google are investing heavily in edge AI hardware to support real-time AI applications across PCs, industrial systems, robotics, and mobile devices.
According to IDC, worldwide spending on edge computing continues to grow as organizations move AI workloads closer to data sources to reduce latency and improve operational efficiency. Gartner also expects AI PCs and intelligent edge devices to become major drivers of enterprise AI adoption over the coming years.
For Longsys, the FMS 2026 showcase signals a strategy that extends beyond storage hardware into broader AI infrastructure. Through its consumer brand Lexar and manufacturing operations under Zilia in South America, the company is expanding its ability to support localized production and global delivery of AI storage solutions.
As enterprises increasingly deploy larger AI models across diverse edge environments, innovations in memory architecture, intelligent storage management, and AI-optimized SSDs are expected to become key enablers of scalable, energy-efficient intelligent computing.
Market Landscape
The edge AI infrastructure market is entering a period of rapid expansion as enterprises seek alternatives to cloud-only AI deployments. IDC forecasts continued growth in edge computing investment, driven by demand for low-latency AI inference across industrial automation, intelligent PCs, and mobile devices. Meanwhile, Gartner expects AI PCs, intelligent storage, and AI-optimized memory architectures to become foundational technologies supporting enterprise AI over the next several years.
Vendors capable of integrating memory, storage, and intelligent software orchestration are increasingly positioning themselves as critical infrastructure providers within the evolving AI hardware ecosystem.
Top Insights
- Longsys introduced its Edge AI Storage Fusion portfolio at FMS 2026, combining high-bandwidth memory, intelligent storage orchestration, and AI-optimized SSDs for edge AI applications.
- A collaboration with AMD enables local deployment of LLMs up to 122 billion parameters using 64GB of optimized AIDIMM™ memory through intelligent memory scheduling.
- New PCIe Gen5 SSDs powered by Longsys’ 5nm SPU™ deliver flagship performance with lower power consumption, supporting the growing demands of AI-enabled PCs.
- HLCache™ and AILPBGA™ improve AI performance on mobile devices by optimizing DRAM utilization and embedded memory bandwidth for on-device AI inference.
- The announcements reflect growing enterprise demand for integrated edge AI infrastructure that combines memory innovation, storage optimization, and intelligent software management
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
