FMW 2026 AI Storage Chip Map: Full-Stack Industry Landscape Across System, Device, and Media Layers

Overview

The "AI存储芯图" (AI Storage Chip Map) was released at the FMW 2026 Global Flash Memory Summit. It is not a simple logo collection but a complete AI storage industry ecosystem map, organized into three hierarchical layers: system layer, device layer, and media & chip layer, systematically covering the full industry chain from AI data generation to final storage media.

Layer 1: System Layer — AI Data Flow & Storage Systems

This layer focuses on three core AI workload stages — data sources, model training, and model inference — plus supporting data preprocessing and AI memory libraries.

AI Data Sources

Major cloud vendors (Huawei Cloud, Alibaba Cloud, Tencent Cloud, AWS, Google Cloud, Azure, Oracle), data platforms (Snowflake, Databricks, Cloudera), and databases (MongoDB, Oracle, IBM) represent the massive raw data origins for AI training and inference.

Data Preprocessing

Tools for data labeling, cleaning, and feature engineering include Scale AI, Labelbox, Databricks, Alibaba Cloud, and Volcano Engine — a critical pipeline stage.

AI Model Training

Vendors providing training compute and storage support: Huawei, H3C, Lenovo, Inspur, HPE, IBM, Tencent Cloud, Alibaba Cloud, delivering high-performance storage clusters for large-scale training.

AI Model Inference

Split into AI memory libraries (vector databases: Pinecone, Milvus, Weaviate, Chroma, Qdrant) and inference storage systems (Huawei, Lenovo, Inspur, Tencent Cloud, Alibaba Cloud, AWS, Google Cloud). Inference demands low-latency, high-concurrency storage.

Layer 2: Device Layer — Hardware Infrastructure

This layer forms the hardware foundation that determines the system layer's performance ceiling.

SSD Controllers

Global mainstream controller vendors: Microchip, Marvell, SMI, Phison, Maxio, DERA, Memblaze, DapuStor, ScaleFlux. These "brains" of SSDs directly dictate read/write performance, power, and reliability.

Memory & Flash Modules

SSD Modules: Samsung, Micron, Kioxia, SK hynix, Western Digital, Solidigm, YMTC, Huawei, Memblaze, DapuStor, ZhiXin — covering enterprise, data-center, and consumer SSDs.

DIMM Modules: Samsung, SK hynix, Micron, Kingston, ADATA, Jiangbolong — traditional DDR memory modules.

Computational Storage

Vendors: NGD, ScaleFlux, NYRIAD, DapuStor, Netint, Pliops. These push compute (compression, encryption, database acceleration) into the storage device, reducing data movement and boosting AI efficiency.

CXL Memory & Interconnect

Key players: Samsung, Microsoft, Intel, AMD, Marvell, Rambus, Astera Labs, Microchip, Montage, Montage Technology. CXL is the key interconnect for memory pooling and tiered memory in future AI servers.

Interface & Array Controllers

Broadcom, Microchip, Astera Labs, JMicron, Marvell, SAGE, Grad Technology — providing PCIe switches, RAID controllers.

Flash Testing

Calypso, VIAVI, Keysight, Teradyne, Advantest, SanBlaze — ensuring chip and module quality and reliability.

Layer 3: Media & Chip Layer — Physical Foundation

This layer determines capacity, bandwidth, power, and cost.

SRAM / DRAM

Intel, Samsung, Sony, Renesas, Micron, Nanya, Winbond, ISSI, Honeywell. DRAM serves as main memory for AI compute; SRAM for cache.

HBM (High Bandwidth Memory)

SK hynix, Samsung, Micron. HBM is the core companion memory for AI accelerators (e.g., NVIDIA GPUs), directly capping AI compute ceiling.

HBF (High Bandwidth Flash)

Samsung, Kioxia, SK hynix, SanDisk. An emerging medium highlighted at FMS 2026, aiming to extend HBM's capacity tier with flash.

NAND Flash

Samsung, SK hynix, Kioxia, Micron, YMTC, Solidigm, Western Digital. Core SSD medium and primary long-term storage for AI data.

Nor Flash

Winbond, Micron, Macronix, GigaDevice, ISSI — used for embedded code storage.

In-Memory Computing Chips

Mythic, MXC, Zhicun, Hengshuo, Yizhu, Houmo — perform compute inside memory cells, breaking the von Neumann "memory wall", a key direction for AI inference.

Emerging Memory Media

FeRAM: Fujitsu, Infineon, TI, LAPIS, Cypress.

MRAM: Everspin, Samsung, Sony, Toshiba, Honeywell, Intel, 2X Memory, Lomare, GlobalFoundries, TSMC.

RRAM: Crossbar, Fujitsu, SanDisk, Intel, UMC, SMIC, Xinyuan.

PCM: Micron, ST, Nanya.

Core Value & Trend Interpretation

Complete AI Storage Industry Loop

The map shows the full chain: AI data generation (system) → storage & processing (device) → physical media & chips (media), reflecting that AI storage competition is now full-stack collaboration, not single-point.

Chinese Vendors' Comprehensive Rise

Chinese companies appear across all layers: system (Huawei, Alibaba, Tencent, Inspur), device (Memblaze, DapuStor, Maxio, DERA, Jiangbolong, Bwei), media (YMTC, GigaDevice, Zhicun), indicating China's significant position in the AI storage ecosystem.

Technology Hotspots Tilting Toward AI

HBM, HBF, CXL, in-memory computing, computational storage are singled out — marking current investment and R&D priorities.

Open Ecosystem & Standard Collaboration

Open interconnect standards like CXL and UCIe are prominently noted, signaling AI storage moving toward open, disaggregated, and pooled architectures to address the memory wall from scaling models.

This "AI Storage Chip Map" serves as a battle map for the AI era, clearly showing AI redefining storage boundaries — from system to media, hardware to software — where storage is no longer passive support but a key variable determining AI performance.