The Fourth GMIF2025 Innovation Summit (Global Memory Innovation Forum) was successfully concluded recently in Shenzhen. Under the theme "AI Applications, Innovation Empowered", the GMIF2025 of the year brought together leading representatives across the global memory industry chain to engage in in-depth discussions on compute-storage convergence, AI deployment, and ecosystem collaboration, exploring pathways for innovation and development in the AI-driven memory era.
The Fourth GMIF2025 Innovation Summit was successfully concluded recently in Shenzhen. Credit:GMIF
In addition to a lineup of insightful keynote sessions, the summit offered diversified formats such as exhibitions, annual awards, and global live broadcast of memory brands, serving as an integrated platform featuring technical exchanges, innovation showcases, brand promotion, and industry collaboration, offering a panoramic view of emerging trends and practical applications in the memory field.
At the very beginning, Rixin Sun, President of Shenzhen Memory Industry Association (SMIA), signified that with the rapid proliferation of AI technologies on the edge side in 2025, the memory and storage industry are stepping into the spotlight. The industry growth mode is shifting into a new structural phase characterized by high performance, high efficiency, and system collaboration. Since its inception, GMIF has upheld the vision of "connecting the memory ecosystem and fostering industry win-win collaboration," aiming to unite global industry forces, drive cross-sector cooperation, and accelerate the integration of China’s memory sector into the global landscape.
Credit:GMIF
Market Shift: AI Drives Memory Industry into a New Growth Cycle
The global memory market is entering a new phase of structural growth driven by AI technologies. According to Xiaomin Han, General Manager of Consulting Services from JWinsights, the ongoing expansion of use cases in smartphones, data centers, and NEVs is unlocking a fresh round of growth for the storage industry. Challenges in advanced chip supply within the domestic industry chain have not diminished the strong investment momentum of internet companies, and it's predicted to achieve large-scale commercialization by 2027, which will propel a new value upgrade of storage products.
Meanwhile, the scale and structure of data are undergoing profound changes. Maya Zhang, Product Marketing Director from Sandisk, shared that the global total data volume will reach 200ZB by 2025, among which 80% of the total is unstructured data. The increasing demand for real-time data is propelling the memory architecture evolving toward compute-storage convergence. Sandisk has achieved a 33% power reduction through technical innovation and plans to advance capacity from 400TB to 800TB.
The popularization of AI applications has driven a multi-level leap in storage demands. Wallace C. Kou, CEO of Silicon Motion Technology Corporation, noted that the shift toward compute-storage convergence is accelerating the leap in storage configurations from TB to PB/EB levels. Individual servers now require over 200TB, and use cases are expanding quickly from cloud environments to edge terminals like AI PCs and intelligent vehicles. Silicon Motion is enhancing its collaboration across the industrial chain and participating in interface standardization to develop next-generation high-performance storage solutions.
The market is poised for substantial expansion. Zining Wu, Founder and Chairman of InnoGrit Corporation, stated that global data volume is projected to hit 1,000 ZB by 2030, with AI-related storage expected to represent about two-thirds of the total. This surge in demand is set to grow the enterprise SSD market from USD 23.4 billion in 2024 to USD 49 billion in 2028. To support rising GPU computing requirements, the industry is targeting a jump in AI SSD IOPS from the millions to the hundreds of millions by 2027, marking a transformative memory performance breakthrough.
Technological Breakthroughs: Architectural Innovation Accelerates Industry Upgrades
The rapid advancement of AI technology poses unprecedented challenges to storage systems, prompting industry players to seek breakthroughs at the architectural level.
Kevin Yoon, VP and CTO of Memory Division from Samsung Electronics, highlighted that AI agent characterized by autonomous decision-making capability is evolving, so traditional storage architectures face challenges in terms of bandwidth, power consumption, and latency. To address this, Samsung is mass-producing 24Gb GDDR7, expanding its CXL 3.1 ecosystem, and planning PCIe Gen6 SSD launches, introducing a new category of "storage-class memory (SCM)" aimed at reducing AI inference latency.
Credit:GMIF
Edge computing is emerging as a new innovation focus. Matt Bromage, Head of Global Storage Business from Arm, projected that 70% of AI inference workloads will soon be processed at the edge, promoting the growth of "ambient intelligence." Arm is advancing technologies such as near-data computing and Chiplet integration, leveraging customized designs to achieve performance improvements of up to 1,000 times.
Benny Ni, GAR Sales VP from Solidigm, stressed that AI development is increasingly focused on real-world applications. Solidigm has launched 122GB QLC high-capacity products and liquid-cooled SSDs to meet the dual requirements for performance and energy efficiency of edge computing.
On the ecosystem side, Haibing Xie, Director, Application Design in Center, Intel China, introduced an open ecosystem supporting over 900 large AI models, which helps promote AI PC adoption through hardware-software synergy. Xiaobing Wang, Director of Customer Projects from MediaTek, pointed out that the explosive demands for edge AI computing bring thermal and bandwidth challenges, and MediaTek is optimizing system performance through advanced packaging, new process technologies, and architectural innovations.
Prof. Guangyu Sun from Peking University's School of Integrated Circuits stressed that it's necessary to balance storage costs and computing efficiency through path optimization and dynamic resource management. With the practical deployment of advanced packaging, collaborative innovation between storage and compute systems becomes increasingly critical, and attention shall be especially shifted into heterogeneous integration, in order to achieve large-scale commercial applications.
Industry Collaboration: Building a Secure and Self-Reliant Memory Ecosystem
Amid AI-driven transformation, domestic enterprises are accelerating ecosystem development through technological innovation and collaboration.
Credit:GMIF
BIWIN is making concrete efforts to land its Integrated Solution and Manufacturing (ISM) strategy, specifically covering independent controller design, solution R&D and advanced packaging and testing capabilities. Han He, CEO of BIWIN, claimed that the company is committed to transitioning from a traditional module manufacturer to an AI solution provider. Supported by its ODM services, customized solutions and compute-storage integration capability, BIWIN will inject new energy into the memory industry.
Maxio Technology is taking active steps to scale up production of new products while facing new requirements for performance, intelligence and energy efficiency enhancement. Guoyang Li, General Manager of Maxio Technology, underlined that Maxio remains steadfast in its vision as "One Chip, One Ecosystem, One Future", strengthening industrial chain cooperation to inject new impetus into the development of the storage industry
In underlying technology area, Montage Technology continues to pioneer CXL-based solutions. Shawn Xiao, Director of Ecosystem Partnerships and Field Applications, shared that its memory expansion solution is able to support 95-99% local DRAM throughput with 24%+ TCO savings in data center applications. In AI inference scenario, its single CPU performance can compete with dual-CPU configuration, effectively breaking the limits of "memory wall."
Ming Zhao, General Manager of OKN Technology, expressed that as AI memory raises its demands for reliability, the company offers wide-temperature testing solutions ranging from -70℃ to 150℃, capable of enhancing testing efficiency by 80%. At the same time, OKN is making forward-looking arrangements for the testing of high-speed interfaces such as PCIe 6.0. Its "equipment +service" business model underpins OKN to support the R&D and mass production demands of the memory industry.
Tailored for AI semiconductor market demands, Lucas Lu, Director of R&D Department from GreaTech Substrates, showed that the company is capable of processing ultra-thin substrates with a thickness of 70 μm and line width/spacing of 10/15 μm. It can mass-produce 110×110 mm, 26-layer high-performance FCBGA substrates to meet differentiated demands on both the compute and storage ends. With six smart manufacturing bases in place, the company provides end-to-end substrate solutions for the AI era.
Applications Deployment: Accelerating Adoption Across Vertical Industries
As AI technologies evolve, the application of memory solutions across industries is the crucial key to promoting industry revolutions.
Credit:GMIF
In automotive manufacturing sector, Xuewen Chen, Chief AI Scientist from GAC Group, indicated that AI is expanding far beyond its single application in autonomous driving, while strengthening its presence in reshaping the whole industry value chain from R&D and manufacturing to terminal products. By improving R&D efficiency, enhancing manufacturing flexibility, and enabling autonomous decision-making in vehicles, AI is driving the automotive industry toward a fully intelligent transformation.
In terms of the IT sector, Cheng Zhu, Product Director at China Greatwall, said that the company has established a full-spectrum product lineup compatible with domestic technology routes including Phytium, Hygon, and Kunpeng, offering end-to-end solutions from general servers to AI servers. With a four-layer security architecture in place, China Greatwall is deeply involved in key national IT projects, showcasing solid execution in government and mission-critical sectors.
Regarding artificial intelligence applications, Shang Shang, General Manager of the Information Security from iFLYTEK, introduced that the Spark Large Model is the first nationwide open model in China trained on a domestically developed compute-storage platform. Through technical optimization, it delivers performance comparable to trillion-parameter foreign models with just 70 billion parameters. Notably, innovation efforts have increased the utilization rate of Huawei's Ascend 910B accelerator cards from below 30% to 95%, significantly boosting training efficiency on domestic platforms.
These successful cases showcased a clear transition in AI technologies from research pilots to broad commercial rollout, providing solid support for digital transformation across industries. AI storage and memory solutions will become central to a growing range of scenarios in the acceleration of intelligent transformation as technology matures and ecosystem perfects.
Solutions Showcase: Industry Leaders Demonstrate Collective Strength
Beyond keynote sessions, GMIF2025 featured product exhibitions, global live broadcast of memory brands, and the annual awards ceremony, fostering diversified collaboration within the semiconductor memory ecosystem.
Over 20 leading enterprises showcased cutting-edge technologies. Highlights included Sandisk's high-performance consumer products based on 218-layer BiCS8 NAND, Solidigm's enterprise QLC SSDs with liquid-cooled solutions engineered for AI and cloud computing, Kioxia's all-scenario products such as ultra-large 245.76TB enterprise SSD based on BiCS FLASH, and Intel's Core Ultra and Xeon 6 processors with integrated NPU capabilities.
Silicon Motion presented its SM2504XT controller for AI PCs; InnoGrit unveiled PCIe 5.0/Gen6 SSDs and RISC-V-based controllers; BIWIN displayed its full range of embedded and industrial storage products under its Integrated Solution and Manufacturing strategy; and Maxio Technology demonstrated comprehensive controller solutions for all scenarios.
Exhibitors such as NAURA, Montage Technology, TOPDISK, OKN Technology, KingSpec, Hemei Jingyi, Silergy, and others displayed core equipment, testing systems, and chip modules, highlighting the collective strength of the global memory supply chain.
The exhibition received strong industry acclaim for providing both technological insight and business collaboration opportunities.
The GMIF2025 Annual Awards were announced concurrently at the summit. 37 sub-awards have been established in five major categories, namely Industry Contribution Award, Technological Innovation Award, Terminal Application Award, Market Service Award, and Intelligent Equipment Award. Leading companies at home and abroad, including YMTC, CXMT, Kioxia, Samsung Semiconductor, Sandisk, Silicon Motion, Solidigm, Maxio, Intel, MediaTek, BIWIN Storage, InnoGrit, Arm, China Greatwall, Montage, iFLYTEK, GreaTech Substrates, Loongson, LKAUTO, Heyan Technology, TOPDISK, KingSpec, POWEV, Microview, Maxwell Technologies, Silergy, Skyverse, AMIES Technology, Hemei Jingyi, Tytantest, Attach Point Intelligent Equipment, OKN Technology, Quanxing Technology, Greater Bay Advanced Technology, XING YUN, Yanxin Microelectronics, and UNIMOS, received honors, showcasing innovative vitality in the global memory industry.
The "Bay Area Night" session following the award ceremony provided a relaxing networking platform for attendees to engage in meaningful exchanges. The event also featured a global live broadcast of memory brands, during which BIWIN Storage, Silicon Motion, InnoGrit Corporation, POWEV, KingSpec, Quanxing, Silergy, and OKN Technology shared forward-looking insights into storage technologies and AI application prospects.
For more information about GMIF, please visit.
Article edited by Joseph Tsai