Behind Qualcomm's Partnership with Shenzhen Longsys Electronics (09976): The First Wall AI Hits Is Not Compute Power, but Storage

Title context: Behind Qualcomm's Partnership with Shenzhen Longsys Electronics (09976): The First Wall AI Hits Is Not Compute Power, but Storage Text: During the recent 2026 Snapdragon Summit, Qualcomm announced a four-party collaboration with StepFun, Wulianghuo, and Shenzhen Longsys Electronics (09976), based on the sixth-generation Snapdragon 8 Super Ultimate mobile platform, to deeply adapt and optimize inference for the StepEdge-Omni 30B-MoE (Mixture of Experts) model on the device side, further accelerating the arrival of agentic AI experiences to more terminals and consumers. The presence of a memory manufacturer on this partnership list itself says one thing: for large models to truly sink into phones, PCs, and Siasun Robot&Automation, the first question to answer is not whether the chip computes fast enough, but whether the memory can hold it and feed it. From the Snapdragon Summit to the collective price hikes by terminal manufacturers in the second half of the year, both point to the same thing: when large models sink from the cloud to terminals, the first wall they hit is not compute power, but storage. From "Not Enough Capacity" to "Not Enough Bandwidth": AI Pushes the Storage Wall into Everyone's Pocket In the second half of this year, the pressure of storage price increases began to transmit to consumers in the most direct way. In early September, Xiaomi, Huawei, and Honor successively raised prices on multiple models on sale, covering flagships, mid-range, and even some entry-level products, with overall increases ranging from 200 to 1,000 yuan. Behind the collective price adjustments by terminal manufacturers is the systemic scarcity of upstream storage resources. The cause of this round of scarcity is not how robust consumer electronics demand is; the real variable is AI. TrendForce data shows that in 2026, HBM and RDIMM are expected to account for 51% of DRAM supply bits, with original manufacturers continuously tilting limited advanced capacity toward high-value-added products such as HBM and server memory, structurally compressing the supply elasticity of general-purpose DRAM and NAND. In August, global average DRAM and NAND prices hit new historic highs, and the institution has raised its forecast for the quarter-on-quarter increase in third-quarter PC DRAM contract prices to 18%-23%. Beyond price, there is another hurdle that money alone may not overcomephysical capacity. Running a 7B-parameter large model requires more than 14GB just to load the weights into memory, while mainstream phones can allocate only 1-2GB to a single application. Winbond Electronics also pointed out at a recent industry conference that on-device AI imposes a "triangular constraint" on storage centered on capacity, bandwidth, and power consumption, with the three mutually restraining each other, and traditional storage solutions are already clearly struggling in edge inference scenarios. In other words, what terminal manufacturers face is no longer a multiple-choice question of "whether to spend more money," but a physical ceiling of "spending more money may still not fit it." In other words, the biggest gap left by AI compute expansion has shifted from "how fast it computes" to "whether the data can be delivered." And this is exactly Shenzhen Longsys Electronics' home turf. Don't Stack Memory, Change the Architecture: Making 4GB Deliver an 8GB Experience Facing the memory wall, the industry's mainstream approach is to add capacity and boost bandwidth, at the cost of simultaneously rising BOM costs and power consumption. Shenzhen Longsys Electronics offers a different solutionnot installing more memory in terminals, but making existing memory "sufficient." The company's SPU (Storage Processing Unit) chip integrates a storage control engine and an intelligent processing engine, paired with its self-developed iSA storage agent architecture, building a closed loop of "chip hardware + intelligent scheduling" for on-device AI storage software-hardware collaboration. After joint tuning with AMD, on AI mobile workstations and AI PC platforms, the scale of on-device AI models carried increased by about 3.2 times, memory utilization efficiency improved by about 200%, and terminal DRAM usage dropped by about 40%. At the experience level, after 4GB DDR is equipped with this intelligent software-hardware architecture, terminal response time approaches the normal configuration level of 6GB/8GB DDR. At a time when memory prices have doubled in a year and phones are raising prices across the board, the industry implications of these numbers are extremely direct: still delivering the same experience with 2-4GB less memory means terminal manufacturers can offset upstream price increases with lower BOM costs. Technical advantage thus translates directly into bargaining power. At present, Shenzhen Longsys Electronics' product matrix for diverse terminals has also taken shape. AIDIMM supports up to 128GB capacity, 256-bit width, and 307.2GB/s bandwidth, supports 0.9V1.05V dynamic voltage regulation, and features the FDVFS intelligent energy-efficiency optimization mechanism, which can intelligently adjust voltage and operating states for different load scenarios such as on-device AI inference and large model operation; AILPBGA adopts a single-die native 256-bit width design, with bandwidth reaching 307GB/s and capacity covering 24GB-64GB, fully compatible with LPDDR standard interfaces without the need to rebuild the SoC and system architecture; the ePOP5x for AI/AR smart glasses and smartwatches has a package thickness as thin as 0.52mm and DRAM transfer rates up to 8533Mb/s. The underlying support is the large-scale mass production of self-developed 5nm advanced-process UFS 4.1 controller chipsthe company's self-designed controller chips have cumulatively shipped over 280 million units, and it has launched SPU chips using the same advanced process, with NAND I/O reaching 4800MT/s and supporting a maximum single-drive capacity of 128TB SSD. Building on this product matrix, Shenzhen Longsys Electronics further expands the platform ecosystem of the iSA architecture. Recently, the company took the lead in adapting its industry-unique iSA storage agent to the sixth-generation Snapdragon 8 Super Ultimate mobile platform, empowered by the software-hardware collaboration of its self-developed iSA storage agent and Huiyi Micro controller UFS 4.1 storage chip, enabling efficient loading of model weights from on-device flash memory, greatly reducing the runtime memory footprint of large models through storage-compute collaboration, and significantly improving model loading and running efficiency. In the future, Shenzhen Longsys Electronics will continue to deepen ecosystem collaboration with Qualcomm to jointly build high-performance on-device AI solutions and bring terminal users a more stable and reliable local AI experience. Premiumization Takes Over: The Structural Shift Behind a 59% Gross Margin The value of architectural capability must ultimately be verified in the revenue structure. In the first half of 2026, the company achieved operating revenue of 24.088 billion yuan, up 136.26% year-on-year; net profit attributable to shareholders of the parent company was 10.577 billion yuan, up 71,528.66% year-on-year; the gross margin of storage products reached 59.08%, up 45.80 percentage points year-on-year, with a comprehensive gross margin of 58.90% and a net margin of 44.49%. More noteworthy is the speed at which high-margin businesses are taking over. Enterprise storage revenue in the first half was 2.140 billion yuan, up 208.80% year-on-year. The company has become one of the few domestic enterprises with "eSSD+RDIMM" product design, combination, and large-scale supply capabilities, with products introduced into the supply chains of some leading internet companies and server manufacturers, and has completed compatibility adaptation with multiple domestic CPU platforms including Kunpeng, Hygon, and Phytium. DDR5 RDIMM has passed certification for the AMD Threadripper PRO 9000WX series workstations, and it has also lit up SOCAMM, combining MRDIMM and CXL 2.0 memory expansion modules to build a complete enterprise-grade product matrix. Automotive-grade UFS 4.1 is equipped with a self-developed 5nm controller, with sequential read speeds up to 4200MB/s, and has entered the introduction and validation stage with leading automakers and begun direct supply to North American smart car and autonomous driving technology giants. The company has established cooperation with more than 20 OEMs and over 50 Tier 1 suppliers. mSSD relies on the SiP system-in-package technology at its Suzhou packaging and testing base to achieve large-scale mass production, with monthly capacity exceeding one million units, and can flexibly adapt to scenarios such as PCs and laptops, gaming handhelds, drones, and VR devices. Global layout is also delivering simultaneously: Zilia achieved external sales revenue of 3.950 billion yuan, up 184.58% year-on-year; Lexar global sales revenue was 3.966 billion yuan, up 84.90% year-on-year; overseas revenue was 17.017 billion yuan, up 142.27% year-on-year. With the completion of the acquisition of the remaining 19% equity of Zilia in March 2026, the company has built a global manufacturing network covering Zhongshan, Suzhou, and Manaus and Atibaia in Brazil, with the three major brands FORESEE, Zilia, and Lexar taking shape in coordination. It is worth mentioning that this Hong Kong stock fundraising will also be mainly used to enhance independent R&D and innovation capabilities in chip design and advanced storage product development, with specific investments focused on three major directions: AI high-end memory R&D, controller chips, and advanced packaging and testing. When memory becomes the scarcest resource in the AI era, the core of industry competition also shifts: whoever can deliver stronger intelligence with less memory will master the ability to define the next generation of terminals. Today, leveraging its "A+H" dual platform, Shenzhen Longsys Electronics is aiming precisely at long-term positioning under this trendgradually building underlying capabilities suited to on-device AI needs around the collaborative efficiency of storage and compute. Only by taking this step steadily can it have the opportunity, in the evolution of the next generation of terminals, to grow from a link in the supply chain into a key technology participant.