Longsys Electronics, a China-listed independent memory company, has resubmitted a listing application to the Hong Kong Stock Exchange, seeking a dual listing.
The filing comes as China’s domestic memory supply chain sees a wave of capital market activity. ChangXin Memory Technologies, the country’s leading domestic DRAM (dynamic random access memory) manufacturer, passed the Shanghai Star Market listing committee review in May and is expected to list as early as July. Yangtze Memory Technologies, China’s largest NAND flash memory manufacturer, completed its IPO counseling filing in May, formally starting its listing process.
According to its prospectus, Longsys is the world’s second largest independent memory company and the largest in China. Its revenue reached RMB 22.766 billion (USD 3.4 billion) in 2025, while its A-share market capitalization stood at RMB 220 billion (USD 32.4 billion).
The company owns three brands:
- Foresee serves B2B customers and contributed about RMB 9.066 billion (USD 1.3 billion) in revenue in 2025, covering consumer electronics, internet-of-things, automotive electronics, enterprise storage, and other categories.
- Zilia serves B2B customers in Latin America and generated RMB 2.924 billion (USD 430.3 million) in revenue in 2025.
- Lexar targets the global high-end B2C market and recorded RMB 4.741 billion (USD 697.7 million) in revenue in 2025.
From 2023–2025, 77.1%, 71.1%, and 66.8% of Longsys’ revenue came from outside mainland China. Its customers include Dell, Lenovo, Mindray, Oppo, Samsung, Transsion, and Xiaomi.
The Hong Kong listing plan is about more than fundraising and capacity expansion. Longsys’ prospectus points to a broader shift in the company’s business, from consumer-grade memory modules toward enterprise storage and customized edge artificial intelligence solutions. The goal is to build stronger customer retention and pricing power through technical solutions, rather than relying mainly on standardized hardware sales.
Demand for AI computing power has led the three major IDM (integrated device manufacturing) companies, Samsung, SK Hynix, and Micron, to shift their most advanced capacity and R&D resources toward cloud AI-related HBM (high-bandwidth memory) and enterprise DRAM, including RDIMM (registered dual in-line memory module) products. As a result, capacity for edge DRAM, including DIMM (dual in-line memory module) and LPDDR (low-power double data rate), has tightened, pushing up prices.
Longsys argues that large IDMs have reduced their direct participation in the edge AI storage market, creating structural room for non-IDM companies. Estimates by Omdia, TrendForce, Yole, and China Insights Consultancy put the non-IDM memory market at USD 275.4 billion in 2025 and project it will reach USD 875.6 billion in 2030, implying a compound annual growth rate (CAGR) of 26%. Within that market, AI storage is expected to grow from USD 13.8 billion to USD 141.5 billion, at a CAGR of 59.4%. The same estimates suggest the outsourcing ratio among IDMs will continue rising from about 20% today, with 75–90% of end-market storage products ultimately requiring completion by non-IDM companies.
Edge AI applications span phones, PCs, wearables, vehicles, robots, and other device formats. These applications require localized deployment and more differentiated product services. Longsys is expanding in AI storage by strengthening its customer customization capabilities and working with global IDMs to serve segments where large memory manufacturers have reduced direct participation.
Longsys has said its B2C base is highly sensitive to storage cost swings during industry cycles. Price increases directly affect sales and profit. Edge AI scenarios, including robotics and autonomous driving, have a stronger ability to absorb storage costs. Their core requirements are delivery assurance and product implementation, which can support both higher shipment volumes and higher prices for specialized storage products.
Longsys is using HLC (high-level cache) technology to help customers reduce cost pressure caused by rising DRAM prices.
In edge AI devices, not all data requires expensive high-speed storage such as DRAM. A large amount of warm and cold data, meaning data that is accessed less frequently than active data, can be cached by increasingly fast NAND flash memory through Longsys’ controller chips and firmware algorithms. Because DRAM costs far more per unit of capacity than NAND flash, HLC technology allows end devices to reduce DRAM usage without materially affecting user experience, lowering overall storage costs.
Another bottleneck in local inference for large models is that, as an edge AI device generates each token during operation, it accumulates KV (key-value) cache. The longer the context, the more memory the KV cache occupies. Longsys’ self-developed iSA storage agent, used with HLC, can dynamically offload cold data in the KV cache to NAND based on access frequency, freeing up DRAM space.
In ecosystem development, Longsys has completed joint optimization of HLC technology with AMD and Unisoc, demonstrating cross-platform compatibility. It is now advancing productization and market promotion. Once device makers design their system architecture around HLC, switching storage suppliers becomes more difficult. Unlike the traditional module business, HLC gives Longsys a deeper role in customer systems.
Around edge AI scenarios, Longsys has built integrated storage capabilities across chip design, firmware algorithms, packaging, and testing. Controller chips manage the intelligence of storage products. The company has independently designed eight controller chips, six of which have been commercialized. In March, it launched the WM8500 SPU (storage processing unit) controller chip for the AI era, enabling data compression, hot and cold data identification, and workload adaptation.
The company also commercialized its self-developed controller chip for UFS (universal flash storage) 4.1 products in 2026, becoming the first storage company in mainland China to do so, according to the company. UFS 4.1 is a high-performance flash memory standard and a core component for AI smartphones and other devices.
The combination of iSA and HLC has already translated into shipment capabilities, according to Longsys. Its mSSD (micro solid-state drive) products for AI PCs and other scenarios have received recognition from leading PC makers and are expected to enter large-scale commercial application this year. Its ultrathin ePOP5x and ePOP4x products have been adopted by leading global smart wearable device companies. Its automotive-grade products have entered the supply chains of leading global customers, including North American intelligent vehicle and autonomous driving technology companies, and have secured an initial market share.
This month, Longsys released a new stack of edge AI storage products at Computex, including AIDIMM and AILPBGA, dedicated memory solutions for edge AI inference. AIDIMM targets agentic AI hosts, with up to 128 gigabytes per module, while AILPBGA focuses on embedded AI scenarios through compact packaging. With these launches, Longsys’ edge AI storage solutions now cover both NAND and DRAM, giving it a more complete product portfolio.
Longsys is also using the industry-grade service network built by its Foresee brand to move from its traditional strength in consumer-grade storage into the enterprise storage supply chain.
In 2025, the company’s enterprise storage revenue reached RMB 1.783 billion (USD 262.4 million), up 93.30% year-on-year. In the first half of 2025, it ranked first among domestic brands in China’s enterprise SATA (Serial Advanced Technology Attachment) SSD (solid-state drive) market by total capacity, according to the prospectus.
In DRAM, revenue from memory module products grew from RMB 513 million (USD 75.5 million) in 2023 to RMB 2.216 billion (USD 326.1 million) in 2025, representing cumulative growth of about 335%.
In enterprise DRAM, IDMs produce standardized DRAM chips, while Longsys performs customized tuning on that base. Its focus is on differentiated RDIMM solutions for specific platforms and scenarios.
Longsys’ RDIMM products, including DDR5, have been adapted to domestic server CPU platforms such as Kunpeng, Hygon, and Phytium. Its DDR5 RDIMM has also been certified for AMD’s Threadripper Pro 9000WX series workstation CPUs. According to the prospectus, the company’s enterprise products have entered the supply chains of some leading internet companies and server manufacturers.
The company’s product roadmap is aligned with the evolution of AI server memory architecture, including demand for higher bandwidth, lower power consumption, and modularization. It has mass-produced DDR5 RDIMM products in 32-, 64-, and 96-gigabyte specifications, with capacity specifications reaching up to 256 gigabytes. Next-generation enterprise DRAM form factors, including SOCAMM, MRDIMM, LPCAMM2, and CXL 2.0 memory expansion modules, have entered its R&D reserve. The company is trying to position itself ahead of changes in AI server memory architecture.
The final piece is capacity preparation. Driven by enterprise storage and edge AI demand, multiple Longsys product lines are in ramp-up or mass-production introduction stages. Meanwhile, wafer capacity at IDMs is expected to remain tight for some time, further reducing the share available for consumer-grade and industrial-grade storage.
According to Longsys’ first-quarter 2026 report, inventory rose from RMB 11.678 billion (USD 1.7 billion) at the start of the quarter to RMB 17.961 billion (USD 2.6 billion), an increase of nearly RMB 6.3 billion (USD 927.2 million). During the same period, advance payments rose from RMB 1.369 billion (USD 201.5 million) to RMB 4.642 billion (USD 683.2 million), up 239%, as the company locked in a large amount of upstream wafer capacity.
In 2023, Longsys recorded revenue of RMB 10.125 billion (USD 1.5 billion), a gross margin of 4.7%, and a net loss of RMB 837 million (USD 123.2 million). In 2024, as the cycle reached an inflection point, revenue rose to RMB 17.464 billion (USD 2.6 billion), gross margin recovered to 15.8%, and the company returned to profitability with net profit of RMB 505 million (USD 74.3 million). In 2025, the recovery continued, with revenue reaching RMB 22.766 billion, gross margin rising to 18%, and net profit reaching RMB 1.498 billion (USD 220.5 million).
In the first quarter of 2026, Longsys benefited from continued increases in storage prices and inventory built during a period of lower prices. Net profit attributable to shareholders, excluding nonrecurring items, reached RMB 3.943 billion (USD 580.3 million), 2.8 times the comparable full-year 2025 figure. But for memory module makers, excess profit earned during an upcycle can still disappear during a downcycle.
This profit surge comes with several variables. Longsys’ advance capacity locking has shifted from a one-way bet on prices to a demand-led model, in which customer demand is identified first and upstream capacity is locked afterward. The prospectus describes this shift through two cooperation models:
- Under PTM (product technology manufacturing), Longsys delivers end-to-end customized storage solutions to major customers based on self-developed chips and vertically integrated capabilities, improving its pricing power.
- Under TCM (technology contract manufacturing), Longsys goes further by connecting wafer manufacturers and end customers, using three-party contracts to lock in capacity and demand in advance. In other words, when automakers, server manufacturers, and other customers require stable future supply, the company’s inventory buildup is not only a price judgment during an industry cycle. It is also preparation for contracted delivery obligations.
As these two cooperation models expand, contract customers could provide volume and price support even during a downcycle, reducing inventory impairment risk and stabilizing Longsys’ profit base. Meanwhile, customization demand from edge AI is replacing standardized module procurement, while enterprise customers are shifting toward long-term supply chain partnerships. Both trends create more conditions for sustainable profitability.
This article was adapted based on a feature originally written by Robin and published on IPO Zaozhidao. KrASIA is authorized to translate, adapt, and publish its contents.
Note: RMB figures are converted to USD at rates of RMB 6.79 = USD 1 based on estimates as of June 9, 2026, unless otherwise stated. USD conversions are presented for ease of reference and may not fully match prevailing exchange rates.