Author: ICVIEWS Editorial Department
On October 12, the 2026 China International Industry Fair (hereinafter referred to as CIIF) officially kicked off at the National Exhibition and Convention Center in Shanghai. As a key supporting forum of this year's CIIF, the 2026 Edge Computing and Storage Application Industry Forum was held concurrently in the Guoxin Exhibition Hall.
This forum gathered experts from key enterprises in the industry chain, including AMD, Alibaba DAMO Academy, Rockchip, Advantech, ChiTuo Technology, DaPuWei Technology, Dongtu Technology, and Guixin Technology. The "2026 Edge Computing and Storage Application Industry Research Report" was also officially released on-site to promote industry exchange and collaboration, and to help China's IC (Integrated Circuit) industry advance towards high-end development.
The forum was highly popular, with the venue packed to capacity and even the aisles filled with attendees. The participating audience covered multiple fields such as chip IP, general-purpose processors, industrial motherboards, edge-side AI chips, industrial operating systems, new storage media, enterprise-grade SSDs, and advanced packaging EDA. It included not only executives, technical experts, and R&D engineers from various enterprises, but also representatives from manufacturing plants, research institutes, and investment institutions who came to exchange ideas and learn.
01. The Dawn of the AI Agent Era: Edge Intelligent Computing Becomes the Main Battlefield for Industry Transition
Currently, the process of new industrialization in China continues to accelerate. The digital transformation of industrial scenarios has evolved from single-point equipment modification to the intelligent upgrading of entire production lines. Large models and AI Agent technologies are rapidly penetrating the industrial sector. Cloud computing power struggles to fully handle the massive real-time workloads on the factory side. Industrial edge computing, highly reliable industrial storage, and heterogeneous integrated chip technologies have become the key enablers to bridge the last mile of intelligent manufacturing, making this forum a subject of high attention in the industry.
Chen Lihua, General Manager and Editor-in-Chief of Semiconductor Industry Insights, stated in his speech that driven by AI Agent technology, the pace of AI iteration continues to accelerate, and computing power and storage have become the core foundation for the intelligent transformation of industry and the independent and controllable development of high-end manufacturing. The shift of computing power to the edge and the local processing of data nearby have become an inevitable trend in industrial digitalization.
Ji Zhaohui, Vice President of Marketing for Greater China at AMD, delivered a keynote sharing titled "Innovating Together, Winning in the AI Agent Era". He mentioned that the global AI industry is rapidly moving from the large model inference stage to a new stage of AI Agent development. AI Agents possess the capabilities of autonomous planning, tool invocation, and closed-loop execution, and are reshaping thousands of industries such as manufacturing, energy, and transportation. The computing power infrastructure is also undergoing structural changes. Today, AMD has entered a new stage of development. Relying on EPYC server processors and the Ryzen AI Max series products, it fully covers all levels of computing power scenarios including cloud data centers, enterprise-level computing power, professional workstations, and AI PCs.
Ji Zhaohui introduced the new-generation Ryzen AI Max+ Pro 495 processor on-site. This product can locally support large models with a scale of 320 billion parameters, and can even run multiple agents and multiple models in parallel simultaneously.
Ji Zhaohui emphasized that the industrial dividends in the AI Agent era do not belong to any single enterprise. The upstream and downstream of the industry chain must adhere to open collaboration, break the barriers between chips, software, and industry applications, transform general computing power capabilities into usable solutions for industrial scenarios, and jointly explore the huge market opportunities brought by edge intelligence.
Cai Qinan, General Manager of the Industrial IoT Business Group at Advantech, echoed this judgment. In his speech, he systematically reviewed the complete evolution path of industrial AIoT. In his view, the first half of the AI industry focused on cloud-based large model training, while in the second half where industry value is released, the core battlefield is at the edge. Industrial AIoT as a whole will go through five major development stages: "Internet of Everything, data fusion, visual insights, assisted decision-making, collaboration and full autonomy", ultimately moving towards a new model of multi-AI Agent collaborative work. He believes that although consumer-side AI Agent applications have become popular, there are still many practical thresholds for landing in industrial sites; the core foothold of industrial AI lies in scenarios, requiring facing customers' real business pain points rather than simply pursuing technical concepts.
Combining a large number of practical landing cases, Cai Qinan listed scenarios that have been successfully implemented: in the field of smart manufacturing, edge computing power supports visual inspection of production line defects and dynamic tuning of process parameters, helping factories reduce the defect rate; smart energy scenarios rely on edge nodes to complete real-time dispatch of microgrids, ensuring stable energy supply; in the field of smart transportation, edge low-latency computing is used to achieve rapid decision-making for roadside perception.
The "2026 Industrial Edge Computing and Storage Application Industry Research Report" was officially released at the forum. Yang Yuanhang, Senior Analyst at Semiconductor Industry Insights, provided an in-depth interpretation of the core data and industry trends in the report. The report quantitatively demonstrates the growth momentum of the domestic industrial edge market. Data shows that in 2026, the shipment volume of China's industrial PC market is expected to reach 4.489 million units, corresponding to a market size of about CNY 12 billion; the compound annual growth rate from 2026 to 2030 can reach 13.1%, and by 2030, the overall shipment volume is expected to grow to 7.355 million units.
Yang Yuanhang analyzed that this round of market growth is driven by dual factors. On the one hand, emerging applications such as AI quality inspection, mobile robots AGV/AMR, and machine vision bring substantial incremental demand for computing power; on the other hand, expectations of price fluctuations in upstream core components prompt downstream manufacturers to stock up moderately in advance. For manufacturers in the industry chain, it is not enough to simply compete on hardware parameters; they must also delve deep into industrial segmented scenarios, conduct differentiated product design for different industries, integrate hardware, software, and industry algorithms, and build complete solution capabilities in order to establish their own advantages in market competition.
02. Computing Power Infrastructure Moves Towards Openness, and Computing-Control Integration Reconstructs the Industrial Site
There are two completely different types of businesses on the industrial site simultaneously: real-time motion control and equipment logic operation, and AI inference analysis. The former requires microsecond-level deterministic latency, while the latter requires sufficient AI computing power. At the same time, industrial equipment generally has a long life cycle of 5-10 years, while AI algorithms iterate extremely fast. The mismatch between software and hardware is a common pain point in the industry. Facing this series of practical contradictions, processor IP, edge-side AI chips, and domestic industrial controller enterprises have provided implementation paths for domestic computing-control integration from multiple dimensions such as underlying IP, hardware modules, and control systems.
Wang Zhiyu, Deputy General Manager of Sales at Alibaba DAMO Academy, introduced the full-stack capabilities of the XuanTie RISC-V ecosystem for industrial edge under the theme "Open Cores, Intelligent Edge Connection". He pointed out that the open RISC-V architecture provides an excellent foundation for the independent innovation of domestic industrial chips. XuanTie has built a complete product matrix including E, R, C, etc., covering multi-level computing power needs such as high-performance computing, AI acceleration, automotive electronics, and intelligent industrial control, forming a complete capability spectrum. With cumulative shipments exceeding 6 billion units, a single ecosystem can cover various devices from simple I/O control to large model edge gateways.
Wang Zhiyu also pointed out the common pain points in the industry: after obtaining processor IP, many enterprises face the dilemma of complex software adaptation and long product mass production cycles, resulting in the problem of "having IP but struggling to achieve mass production and landing". In addition to computing power challenges, edge scenarios also face difficulties in storage controllers—the PCIe interface is iterating rapidly, and the controller not only needs to run through high bandwidth but also ensure deterministic latency under high concurrency; different manufacturers have a large number of customization demands for NAND scheduling and error correction strategies. The open instruction set architecture is of great value to the supply chain security and iteration rhythm of storage controllers. In response to the above pain points, XuanTie has created the FLEX customization platform and edge-side AI solutions, equipped with a complete three-layer software stack of RTOS, Linux, and Android. Through DSL description, RTL, toolchains, and verification environments can be automatically generated, encapsulating processor IP capabilities into a directly landable productized platform. At the same time, multiple solutions have been launched for SSD controllers, ranging from consumer-grade PCs to all-flash data centers, adapting to diversified storage tasks, and widely serving four major product directions: SSD controller chips, large model edge gateways, programmable edge-side NPUs, and low-power industrial terminals, lowering the development threshold for industrial chips.
Liang Zhilei, Senior Manager of Electronic Product Market at Rockchip, shared the product idea of computing power decoupling from the perspective of terminal chip productization practice. He stated that the life cycle of industrial equipment is as long as several years, but large models and machine vision algorithms are iteratively updated almost every quarter. If AI inference computing power is completely integrated inside the main control chip, once the whole equipment is finalized, subsequent algorithm upgrades will require re-doing hardware revisions. The transformation cost is high and the cycle is long, making it difficult to keep up with the pace of algorithm iteration.
Liang Zhilei analyzed the four core values of landing edge-side large models: data privacy and security, low real-time latency, controllable inference costs, and offline availability without network; at the same time, he objectively analyzed the practical challenges of landing, including bandwidth and performance pressure, ecosystem adaptation, power consumption constraints, cost and hardware area. To this end, Rockchip proposed a dual-track R&D route of "main control + AI co-processor", separating the AI inference computing power, making it into an M.2 form factor co-processing module, and externally connecting it to existing industrial motherboards through PCIe and USB interfaces. The original whole machine hardware does not need to be redesigned. It only needs to replace or add the co-processor module to adapt to the latest LLM large models, multi-modal vision, and speech recognition algorithms.
Yu Hongping, Sales Director of East China Region at the Semiconductor Business Division of Beijing Dongtu Technology, focused his perspective on the high-end industrial scenario of semiconductor equipment, demonstrating the practical achievements of the localization core control system solution. Semiconductor equipment has extremely high standards for control accuracy, system stability, and fault prediction capabilities, and has long been highly dependent on overseas control systems. Relying on its independently developed Intewell real-time operating system, MaVIEW industrial control software, and AUTBUS industrial bus, Dongtu Technology has built a complete domestic software-defined control technology stack, achieving localization for the chamber control system of semiconductor machines.
He pointed out that through virtualization technology, this system achieves the isolated coexistence of non-real-time Windows/Linux environments and hard real-time PLC environments. Even if the Windows/Linux system crashes or experiences a blue screen, it will not interfere with the real-time control link, and high-speed data interaction is completed through shared memory. Through simplified architecture design, it achieves the integrated integration of IPC+PLC and system memory sharing, ultimately reducing the volume of the whole equipment by more than 30%, and synchronously reducing energy consumption and hardware costs by more than 20%. Combined with high-frequency data acquisition and industrial edge AI algorithms, this control system can converge the temperature control fluctuation of the semiconductor chamber to within 0.5 degrees Celsius. The AI fault prediction function can output warnings about 30 minutes in advance of actual anomalies, with a fault prediction accuracy rate reaching 90%.
03. Innovations in Storage Media and Advanced Packaging Consolidate the Edge Computing Power Foundation
The full release of computing power is inseparable from the support of the storage foundation. The industrial edge environment features high temperatures, vibrations, and strong electromagnetic interference. At the same time, AI large models bring KV Cache and massive small file read/write pressures, and traditional storage solutions are gradually exposing their shortcomings. In this forum, three types of technologies—new storage media, innovative SSD architectures, and Chiplet advanced packaging EDA—were discussed on the same stage to explore how to make up for the storage shortcomings of edge computing power from the underlying hardware level.
Dai Qiang, Director of the Marketing Department of Zhejiang ChiTuo, stated that the practical advantages of MRAM in industrial sites mainly include four points: First, high-speed writing. Traditional Flash and EEPROM writing speeds require millisecond levels, while the write cycle of MRAM is only 35 nanoseconds, meaning that at the moment of power failure, MRAM can complete the landing of critical data in an extremely short time. It does not require capacitors or batteries, truly achieving maintenance-free operation and improving system reliability; Second, high durability. The erase/write lifespan of Flash is about 10,000 to 100,000 cycles, EEPROM is 1 million cycles, while MRAM can reach 10^14 cycles. Even with high-frequency recording of 100,000 cycles/second, MRAM can support a service life of over 10 years. For high-frequency data recording and fault recording scenarios, it is not only an improvement in parameters but also a qualitative change from "usable" to "full life cycle guarantee"; Third, wide temperature range. MRAM can work stably in the range of -40°C to 150°C, and its data retention capability exceeds 10 years at a high temperature of 150°C. For scenarios such as automobile engine compartments, industrial high-temperature environments, and downhole equipment, this high-temperature reliability is unattainable by Flash; Fourth, radiation resistance. The storage array of MRAM is immune to single-event radiation, making MRAM an ideal choice for high-radiation scenarios such as commercial aerospace and medical electronics.
With these four advantages superimposed, MRAM is no longer "just another memory", but a technical solution that can simultaneously meet the requirements of speed, lifespan, temperature, and high reliability in critical data storage scenarios.
Lin Bin, Technical Director of Shenzhen DaPuWei, focused on the innovation of enterprise-grade SSD architecture, explaining how dual-mode SSDs reshape the AI and edge computing power foundation. In the traditional deployment mode, high-performance SLC drives and large-capacity QLC drives are independent of each other. The slot resources inside industrial edge cabinets are very precious. If two types of hard drives are deployed simultaneously, it will occupy a large number of hardware slots, increasing equipment procurement and spare parts operation and maintenance costs, and it is difficult to flexibly adjust the performance and capacity ratio. Lin Bin pointed out that in the era of AI inference, storage has become a performance bottleneck. Edge scenarios are constrained by multiple factors such as slots, power consumption, and operation and maintenance. The traditional "SLC+QLC" heterogeneous deployment mode has pain points such as slot waste, complex operation and maintenance, and rigid ratio.
The dual-mode enterprise-grade SSD launched by DaPuWei integrates a high-performance pSLC partition inside a single SSD disk body, enabling a single hard drive to have both a high-performance metadata partition and a large-capacity data storage partition. This architecture effectively solves the pain points of traditional solutions such as high slot occupation, solidified performance bottlenecks, and complex expansion ratios. Under high-concurrency business scenarios, the file metadata operation performance is improved several times. This product is highly adaptable to typical scenarios such as tiered storage of AI inference KV Cache, massive small file read/write in industry, and limited space in edge cabinets. It balances high performance and large capacity within limited hardware space, simplifies the hardware architecture of edge nodes, and reduces the complexity of operation and maintenance.
Dr. Zhao Yi, General Manager of Guixin Technology, shared the innovative practice of 2.5D/3D IC STCO co-design in response to the industry challenge of high-density heterogeneous integration in the post-Moore era. He mentioned that the demand for AI computing power is currently exploding, but the speed of single-chip performance improvement is gradually slowing down, and Moore's Law is gradually reaching its bottleneck. The industry is turning to 2.5D/3D Chiplet stacked heterogeneous integration, relying on multi-chiplet combinations to improve overall computing power density. However, the traditional DTCO design methodology for single chips can no longer adapt to the development needs of multi-chiplet stacking systems; 2.5D/3D integration will face a series of "no man's land" challenges such as complex interconnect design, thermal stress management, multi-physics field coupling, and stacking yield testing and verification.
Dr. Zhao Yi emphasized that 3D stacking is not simple chip splicing; it must achieve the coupling of all links including architecture, layout and routing, simulation, testing, and process, and requires establishing a new STCO design paradigm for multi-chiplet systems. Guixin Technology has created a complete 3Sheng Integration EDA toolchain, coupled with Chips Z, a dedicated AI Agent for heterogeneous integration, fully covering the entire design process of system-level architecture exploration, 2.5D/3D physical layout and routing, multi-physics field simulation analysis, multi-chip DFT test fault tolerance, and multi-die joint verification sign-off, solving thorny problems such as stacked chip interconnection, thermal warpage, and multi-chip test yield.
This Edge Computing and Storage Application Industry Forum fully demonstrates that the mature landing of industrial edge is by no means the victory of a single link. Computing power chips, storage hardware, industrial real-time software, and advanced packaging must iterate collaboratively to truly match the comprehensive demands of industrial sites for real-time performance, high reliability, customizability, and sustainable iteration.