SHENZHEN, China, Sept. 26, 2026 /PRNewswire/ — On September 23, the 5th GMIF2026 Innovation Summit successfully concluded at the Renaissance Shenzhen Bay Hotel. Co-hosted by the Shenzhen Memory Industry Association (SMIA) and School of Integrated Circuits at Peking University, the summit, themed “The Future Built on AI Memory and Storage,” has brought together industry forces from IDMs, controllers, and memory solution providers to OSATs, equipment and material suppliers, server and AI infrastructure vendors, automotive electronics companies, AI applications developers, academic institutions, and investors. In-depth discussions explored the evolution of memory and storage technologies, industry trends, and application innovations in the AI era.

As AI training and inference workloads continue to scale, the boundaries of memory and storage technology and its applications keep expanding — from storage media and controllers to system architecture and AI applications themselves. GMIF 2026 approached the industry through the lens of the token economy, spotlighting inference efficiency, enterprise-grade storage, cloud-edge-device coordination, industrial capital, and the shifting global competitive landscape.

Five Years of GMIF: A Platform Built Around the Memory & Storage Industry

The summit officially kicked off with an opening remarks from Rixin Sun, President of the Shenzhen Memory Industry Association (SMIA). He looked back on GMIF’s growth since its founding in 2019. Now in its fifth consecutive year, the summit has steadily expanded in scale, industry reach, and influence, with its agenda consistently tracking the pulse of the storage sector.

President Sun said GMIF will continue to pursue a more specialized and differentiated approach going forward — staying close to frontier industry trends and supply-chain coordination, while connecting storage companies with adjacent players in AI compute and end-user applications to drive technical exchange, supply-demand matching, and ecosystem collaboration.

As AI Inference Accelerates, Memory & Storage’s Value Climbs

Daniel Yen, Executive Director at Morgan Stanley, opened the technical discussion with a look at how generative and agentic AI are reshaping infrastructure demands. As model weight loading, KV cache management, and data read/write scheduling grow more complex, Daniel noted that AI capital expenditure keeps rising — and memory and storage’s share of that investment, and its strategic value, are rising with it. He pointed to the “memory wall” as a defining challenge now driving fresh innovation in storage architecture, advanced packaging, and related technologies.

Yimao Cai, Dean of School of Integrated Circuits at Peking University, addressed the topic from the angle of AI inference architecture, discussing the growing role of high-bandwidth storage and multi-media integration. As large model inference drives up demand for capacity, bandwidth, and cost efficiency, Prof. Cai suggested that high-bandwidth flash (HBF) and heterogeneous multi-media storage architectures stand to play a larger role — combining HBM, NAND, and RRAM to strike a better balance between performance, capacity, and cost for AI inference.

Global IDMs Race to Meet the Demands of Agentic AI

Kevin Yoon, CVP & CTO of Samsung Memory China at Samsung Electronics, discussed memory architecture in the age of agentic AI. As agentic AI drives token generation and KV cache volumes to new heights, AI systems are placing greater demands on capacity, bandwidth, and energy efficiency. He outlined Samsung’s progress on Z-NAND, PCIe Gen6 SSDs, and ultra-high-capacity data center SSDs, and discussed how tiering across different storage media can more efficiently support KV cache and model weight storage.

Maya Zhang, Senior Director of Product Marketing at Sandisk, focused on data storage needs in the age of AI inference. As multimodal models, long-context processing, and increasingly sophisticated agents continue to grow KV cache demands, she noted, NAND flash is becoming ever more central to AI infrastructure. She added that more flexible data tiering and reuse across SSDs can improve efficiency for different AI workloads, and that high-density technologies like QLC are set to see broader adoption in AI use cases.

Benny Ni, GAR Sales VP at Solidigm, addressed enterprise SSDs’ role in AI infrastructure amid growing data volumes and operational demands. As model size, token counts, and inference complexity all continue to climb, he said, memory offloading and data tiering are becoming core components of AI system architecture — with high-capacity QLC SSDs paired with high-performance storage offering a more efficient, cost-effective data foundation for inference.

Cloud, Edge, and Device: AI Opens New Ground for Memory & Storage

John Xavier Lionel, Head of Global Storage Business at Arm, spoke on system-level coordination in AI inference, noting that inference spans compute, memory, storage, and data movement — all of which require holistic architectural optimization. As small models, quantization, and heterogeneous NPUs continue to advance, he said, AI deployment will increasingly span cloud, edge, and device, with local storage taking on a larger role in hosting model weights, knowledge bases, and application data.

Stanley Huang, AVP at Silicon Motion, focused on storage requirements in multi-agent, concurrent-use scenarios, where differing tasks demand tailored QoS, latency, and resource allocation. He described how Silicon Motion’s controller and resource-scheduling technologies improve storage efficiency under complex workloads, spanning enterprise SSDs, server storage, mobile UFS, autonomous driving, and robotics applications.

Sam Sun, Chairman at BIWIN, discussed how AI is simultaneously driving storage demand across data centers, edge, and endpoint devices — each with distinct requirements. In data centers, he noted, AI training and inference are pushing up demand for enterprise SSDs and server memory; at the edge and endpoint, applications like AI/AR glasses, AI PCs, smart vehicles, and industrial equipment are pushing storage toward smaller form factors, lower power consumption, higher reliability, and tighter system integration. Sam Sun said BIWIN continues to leverage its integrated solutions and manufacturing capability across enterprise, embedded, PC and mobile, industrial, and automotive product lines — with innovations like Mini SSDs, ultra-compact embedded storage, and wide-temperature industrial SSDs opening new ground in the AI era.

From Compute Infrastructure to Token Production, Deeper System-Level Integration

Tao Zhou, General Manager of the Server Division at Lenovo ISG China, discussed the concept of the “token factory.” As enterprise AI moves from proof-of-concept to large-scale deployment, he said, improving the efficiency of AI infrastructure and sustaining stable token output have become critical industry priorities. He also described how top-level design, data governance, compute optimization, and security management — combined with pooled training/inference resources and hardware-software coordination — are helping enterprise AI infrastructure evolve from simple compute buildout into systematic operations.

Fan Zhang, Chief Computing Architect at NEXWISE, discussed integrated management and scheduling across cloud, compute, and storage resources from an operations standpoint, showing how coordination across compute, storage, and software platforms can improve overall infrastructure efficiency.

Wei Xiong, CTO of Infplane, focused on storage tiering in large model inference, explaining how hot/cold data tiering and intelligent scheduling can shift more inference workload onto SSDs — reducing memory footprint and improving token output efficiency.

Across servers, AI computing centers, and storage systems, deeper coordination among compute, memory, storage, networking, and software scheduling is emerging as a defining trend in AI infrastructure.

From Core Technology to Real-World Application, AI Memory & Storage Ecosystem is Converging Fast

As AI continues to move into automotive, robotics, and enterprise applications, the connection between the storage industry and AI use cases keeps deepening.

Junjia Chen, AI Product Director at SYNCORE, discussed the application of agent architecture in automotive scenarios, focused on smart cockpits and vehicle-wide intelligence.

Lusha Chen, General Manager for APAC at Dify, introduced a “workflow plus agent” model for enterprise AI, connecting large models, corporate knowledge bases, business systems, and end-user applications to embed AI more deeply into enterprise workflows.

Gongjie Liu, Regional Director for Central & Southern China & General Manager of Branch Office at Paratera, discussed multi-model access, unified management, and enterprise AI services built around a MaaS platform.

Zhen Li, VP of Genstoraige, spoke to the concept of “storage-powered compute,” covering data hosting, high-speed interconnects, and intelligent scheduling within AI systems.

Yunjie Ye, Chairman of Numbers Law, presented the company’s work on next-generation mechanical storage architectures for large-capacity data storage.

Ming Zhao, General Manager of OKN Technology, addressed testing requirements for the AI era, describing how test equipment is evolving to support higher speeds, more complex scenarios, and real-world workload simulation for PCIe 6.0 SSDs, memory, and increasingly demanding AI applications — laying the groundwork for the development and large-scale deployment of next-generation storage products.

The Future Built on AI Memory & Storage

From AI inference to agentic AI, from emerging storage media to enterprise SSDs, from data centers to edge and endpoint devices, and from controllers, packaging, and testing equipment to servers, AI computing platforms, and applications, GMIF 2026 highlighted the technological innovation and industry transformation taking place across the memory and storage ecosystem in response to AI.

The massive volumes of data that AI generates, retrieves, and moves are steadily elevating memory and storage’s role within the broader computing stack. Demand for high-capacity, high-bandwidth, and high-reliability storage in data centers continues to climb, while emerging endpoints — AI PCs, smart vehicles, AI/AR glasses, robotics — keep opening new application space. Together, cloud, edge, and device are shaping a richer and more complex set of storage requirements for the AI era.

Meanwhile, the technology itself keeps moving — HBM, HBF, NAND flash, enterprise SSDs, and a handful of newer media all advancing in parallel, with controllers, packaging, testing, and software scheduling evolving alongside them into tighter, more coordinated systems.

Now in its fifth successful year, GMIF has established itself as a leading platform for exchange across the global memory and storage industry. Looking ahead, GMIF will continue to focus on technological innovation, industry trends, and supply-chain collaboration — bringing together key players from across the global industry, pursuing an increasingly specialized and differentiated event model, and fostering deeper exchange and cooperation across the storage value chain.

Media Contact:

Carina Gu
wenjing.gu@gmif.com.cn 

View original content to download multimedia:https://www.prnewswire.com/news-releases/ai-memory–storage-the-5th-gmif2026-innovation-summit-successfully-concludes-in-shenzhen-302890873.html

SOURCE Shenzhen Memory Industry Association (SMIA)