Written by: Jacob Zhao, IOSG

Recently, "the first domestically produced storage stock" Changxin Storage officially landed on the Growth Enterprise Market, exploding the scene with an astonishing surge of 500%. Although the storage sector is still affected by the recent round of corrections, AI storage is still being crazily re-evaluated by capital in the current wave of technological narratives. Meanwhile, decentralized storage in the Web3 field has fallen into a long-term silence and loss. Why is it that although both are named "storage," their market performance diverges so drastically? The fundamental answer lies in the thorough differentiation of underlying value functions.

The re-evaluation of storage in the AI era is fundamentally a carnival regarding "hot data efficiency", serving the ultimate maximization of computing power utilization and commercial monetization; while decentralized storage insists on the value proposition of "cold data credibility", defending data fairness, censorship resistance, and the long-term memory of human civilization. The former is an efficiency system for hot data, while the latter is a credibility system for cold data. The current capital market undoubtedly firmly stands on the "efficiency" side, but human civilization ultimately still needs an unalterable memory base. The long-term value of credible cold storage has never disappeared; it has merely been dormant in the dark side of the cycle, waiting to be repriced by the times.

Why storage has become the focus of the AI industry chain again

In the traditional IT era, storage was a "capacity business." Enterprises' CIOs focused on unit capacity costs, hard drive reliability, disaster recovery plans, archiving strategies, and equipment renewal cycles of up to 3-5 years. Storage was seen as an accessory that followed server procurement.

This round of storage boom is not a traditional cycle recovery, but a repricing of AI's ability to move data. In the era of large models, storage logic has qualitatively changed from "capacity priority" to "efficiency supremacy," rigorously tackling extreme metrics like GPU feeding rates, checkpoint writes, and extremely low latency in RAG. This indicates that the value of storage is leaping from "the final resting place of data" to "the high-speed channel for data to enter computation."

The evolution of resource bottlenecks in AI infrastructure is essentially a battle to supplement the "barrel effect." The true utilization rate of computing power is not a linear addition of single assets but a strict multiplicative effect: true computing power utilization = GPU × HBM × DRAM × SSD × network × file system; any weak link will lead to a collapse in overall computing power utilization. In the AI era, storage has for the first time transitioned from a "cost center" to an "efficiency engine." This is the fundamental logic behind the repricing of storage.

The panoramic view of AI storage architecture: from HBM bandwidth organs to data lake foundations

AI storage is not merely a stack of single hardware but a tightly coupled, layered scheduling complex system. In this system, industrial value and capital focus highly converge on HBM, enterprise-grade SSDs, SSD controllers, NVMe/CXL protocols, and high-performance storage systems. To clearly analyze the flow of value, we will categorize the AI storage architecture into four core levels from top to bottom:

• Compute proximity memory layer (bandwidth core): with HBM as the absolute main force, supplemented by DRAM and CXL memory pooling technology. This layer directly fits GPU/CPU packaging or buses, aiming to break the "memory wall," the first checkpoint determining whether computational power can be fully released.
• High-speed persistent storage layer (IO hub): the core logic is enterprise-grade SSD = NAND chips + SSD controllers + NVMe/PCIe data pathways. This layer undertakes high-frequency checkpoint writes, massive training set loading, and RAG hot data caching, being the most explicit persistent storage increment in AI data centers.
• Low-cost large capacity storage layer (capacity base): composed of HDDs, cold storage, and data lake archiving systems. In the face of exponentially expanding multimodal raw data, historical logs, and compliance backups, this layer still provides an irreplaceable TCO (total cost of ownership) advantage.
• AI storage systems and data software (scheduling brain): including high-performance parallel file systems, distributed object storage, vector databases, and RAG data governance layer. What AI truly consumes is not bare hardware, but the usability of data that has been efficiently organized, indexed, and authorized by the software stack.
• As an ecological extension, decentralized storage does not directly engage in the millisecond-level races for AI hot data, instead anchoring itself in public data set certification, AI training data provenance and long-term cold memory archiving, establishing its unique ecological niche as a "trustworthy cold layer."

HBM: The "bandwidth organ" closest to computing power in the AI storage chain

High Bandwidth Memory (HBM) is not traditional storage but a high bandwidth memory layer near the GPU. Its core mission is not to store data, but to continuously "feed" data to computing power at extremely high bandwidth. HBM is the closest link to computing power in the AI storage chain, with the highest certainty, directly deciding whether the GPU can be "fed," making it the current most critical supply chain bottleneck.

The core architecture of HBM is "3D DRAM stacking + 2.5D advanced packaging": through TSV vertical stacking and CoWoS heterogeneous integration, it achieves extreme compression of storage-computation distances and a qualitative leap in bandwidth. Its industrial barrier is not just DRAM design but a systemic engineering of DRAM process, TSV, ultra-thin stacking, packaging, heat dissipation, testing, and client certification. A quality defect in any link may lead to the scrapping of the entire HBM stack.

Currently, only SK Hynix, Samsung, Micron can stably mass-produce, building a triple moat of top DRAM process, packaging capability, and NVIDIA/AMD client certification.

DRAM and CXL: System memory base and memory pooling engine

HBM addresses GPU proximity's extreme bandwidth; DRAM solidifies the server's system memory base; CXL aims to break physical boundaries and reconstruct how data center memory resources are organized.

• DRAM: mainly carries CPU-side cache, data preprocessing, temporary storage of intermediate states, and system operation, serving as the most basic layer of system memory in servers. The global DRAM market is highly concentrated among the three giants SK Hynix, Samsung, and Micron; Changxin Storage (CXMT) is a core variable in China’s DRAM domestic substitution.
• CXL (Compute Express Link): is a new generation of cache coherence interconnect protocol designed for data centers, aiming to break the limitations of traditional DIMM slots, local memory capacity, and server memory resource islands, promoting memory architecture towards expansion, pooling, and sharing evolution. CXL is currently still in the early stages of transitioning from platform support to large-scale deployment, with high long-term architectural value; core companies include Astera Labs and Lanqi Technology.

Enterprise-grade SSD: Data hub built by NAND, controllers, and NVMe

Enterprise-grade SSDs are the core high-throughput persistent increment of AI data centers, continuously "feeding" data to GPUs with extremely high throughput, extremely low latency, and stable QoS, covering the entire lifecycle of training data loading, checkpoint writing, RAG retrieval, inference caching, and log backflow.

In the AI storage architecture, SSDs are not isolated hardware but a highly coupled system that can be distilled into an industry formula: enterprise-grade SSD = NAND chips + SSD controllers + NVMe/PCIe data pathways. The three layers represent independent links in the supply chain:

• NAND chips (raw material layer): determine storage density and unit cost, while controllers manage performance release and lifespan. Representative companies include Samsung, SK Hynix (Solidigm), Micron, Kioxia, Western Digital, Yangtze Memory Technologies.
• SSD controllers (performance empowerment layer): determine performance release, data error correction, QoS stability, and wear leveling. Representative companies include Phison, Silicon Motion, Marvell, Maxio.
• NVMe/PCIe (data pathway layer): determine the efficiency of data transmission from storage to computation. Integrating GPUDirect Storage technology reduces CPU memory bounce buffer and CPU involvement, significantly alleviating I/O bottlenecks. Representative companies include Broadcom, Marvell, Astera Labs.

HDD / Cold Storage / Archiving: Low-cost base for AI data lakes

AI will not eliminate HDDs. With the multimodal large models’ demand for video and image data, as well as the exponential growth of enterprises' compliance logs and historical data sets, the demand for low-cost cold data storage is surging. In the AI storage architecture, SSDs and HDDs collaborate hierarchically based on business value: SSDs handle hot data and high throughput, while HDDs are responsible for low cost and long-term preservation. Representative companies include Seagate, Western Digital, Toshiba.

AI storage software stack: Scheduling hub for data usability

What AI truly consumes is never bare disks but "data services" that have been meticulously organized by the software stack. This architecture transforms underlying hardware into knowledge assets that can be directly called by AI, specifically divided into four layers:

• High-performance storage systems (supply system): with concurrent throughput and low latency as the core, solving GPU clusters' "data hunger" issue through parallel file systems, ensuring rapid transfer of training and inference. Representative companies include VAST Data, WEKA, Pure Storage.
• Object Storage (raw data lake): with management centered around objects, keys, and metadata, carrying massive amounts of unstructured data. It does not pursue extreme low latency but builds a capacity base with low cost and cloud-native characteristics. Representative companies include AWS S3.
• Vector Databases (semantic indexing layer): responsible for storing, indexing, and retrieving vectors generated by embedding models, helping AI accurately locate relevant information from vast knowledge. Representative companies include Pinecone, Milvus.
• RAG Data Layer (knowledge invocation layer): going beyond single retrieval, covering data slicing, cleaning, permission control, and citation provenance, ensuring that enterprise data can be securely, accurately, and traceably called by large models. Representative companies include Databricks.

From AI hot storage to decentralized cold memory: Efficiency maximization vs Credibility maximization

AI storage is an extremely efficiency-driven system, focused on maximizing computing output. HBM bandwidth determines if the GPU can be fed, SSD throughput determines data set and checkpoint read/write efficiency, and low latency is related to real-time experiences in RAG and inference. These metrics ultimately converge into GPU utilization rates and unit token costs, directly determining the commercial profitability or loss of AI applications. The ultimate goal of AI storage is not preservation, but acceleration, serving productivity.

On the other hand, the value function of decentralized storage is fundamentally different. It questions whether the data will still exist in ten years, whether it has been tampered with, and if it can withstand single-point censorship. Through cryptographic proof and distributed networks, it builds an open-access and permanently preserved public data foundation. Its ultimate goal is to defend the absolute truth and sovereign independence of data, serving fairness, censorship resistance, and civilizational memory.

AI storage is "hot storage" providing fuel for future productivity, while decentralized storage is "cold memory" preserving irremovable historical records for human civilization. The former serves efficiency, pursuing ultimate speed; the latter serves credibility, defending silent memory. The former dictates how fast models run, while the latter determines whether memories will be deleted. Currently, the market relentlessly rewards efficiency, placing AI storage at the forefront, while decentralized storage seems to be experiencing valuation collapse and a narrative drained of life.

The vision and reality of decentralized storage

There are many decentralized storage projects, but according to industry mindset and ecological accumulation, the core representatives remain Filecoin and Arweave. Although both belong to "decentralized storage," their underlying architectural philosophies are almost entirely different paths - the former approaches AWS's elasticity through market contracts, while the latter approaches the eternity of libraries through one-time social contracts.

• Filecoin: constructs the most complete verifiable economic system through PoRep and PoSt. It should not continue to clash with consumer-grade cloud storage against AWS but shift towards AI data provenance, public data set hosting, and compliance archiving to provide verifiable chains for model auditing and copyright proof. The necessary path is to encapsulate into an S3 compatible API supporting fiat currency payments, upgrading from a "cheap storage market" to a "verifiable computing infrastructure."
• Arweave: with the narrative of "one-time payment, permanent storage," pushes miners to save and quickly access as much, especially scarce historical data as possible through Blockweave and SPoRA mechanisms. Its best position is as a human public memory foundation - saving human rights records, war crime evidence, cultural classics, archiving legal and financial history, providing permanently accessible long-term memory for AI agents. Arweave's value lies not in speed but in its capacity to bear civilizational memory across cycles.

The dilemma faced by decentralized storage projects like Filecoin and Arweave is not due to incorrect value propositions but due to long-term mismatches between productization, retrieval experience, real demand, and token incentives. This reveals a huge gap between geek ideals and mainstream commercial applications:

• Supply-demand incentive mismatch: Early networks like Filecoin expanded rapidly through tokens but did not build a strong enough demand side, leading to huge capacity but insufficient utilization and payment conversion. Rewarding "I can store" rather than "I need to store."
• Lack of enterprise-level service capabilities: AWS's barrier is not the hard drive but the "data operating system" made up of APIs, SLAs, permission management, compliance audits, and technical support. Enterprises buy "peace of mind" rather than dealing with experimental infrastructures that require handling keys and node selection themselves.
• Retrieval experience shortcomings: "Putting data in" does not equal "stably, with low latency retrieving it out." Decentralized nodes, complex topologies, and the lack of unified SLAs make it difficult to support AI hot data workflows, making it more suitable for credible cold archiving and data provenance.
• Insufficient privacy compliance: Enterprise private data cannot simply be written into public permanent networks; the conflict between the right to delete and permanent immutability is inherent. Decentralized storage is more suited for public data and long-term archives rather than indiscriminately accommodating core private data.
• Token economics amplifying cycles: Bull market financialization obscures insufficient demand, while bear market miners' ROI declines expose commercial shortcomings. Tokens can cold start supply but cannot automatically create demand and sustainable revenue.

Many other decentralized storage projects focus on specific ecosystems or niche tracks: Storj/Sia's cross-cycle industry mindset and Web3 narrative influence are weaker than Filecoin/Arweave; BNB Greenfield/Walrus is tied to specific public chain ecosystems like BNB or SUI; Celestia/EigenDA belong to the data availability (DA) layer, serving Rollup transaction confirmations rather than long-term archiving; 0G and other AI/DA hybrid narrative projects attempt to integrate storage, data availability, computing, and AI agent settlement into a set of AI-native modular infrastructure, but their real demand, developer adoption, and commercialization loops remain to be validated.

Future opportunities for decentralized storage: the long-term pendulum of efficiency and credibility

During the explosive period of technological dividends, capital crazily chases efficiency, with assets like GPUs and HBM being assigned extremely high premiums, causing decentralized storage advocating "credibility and fairness" to be marginalized. However, the pendulum of history will not stay forever on the efficiency side. Unreasonable bans and content deletions by super platforms, AI copyright litigation exploding necessitating proof of data provenance, geopolitical conflicts triggering data sovereignty disputes, data monopoly leading to the disappearance of public archives, and regulatory pressures for compliance with model training data—events like these could brew a repricing of "trustworthy storage," and the future opportunities for decentralized storage may still manifest unique value in the following directions:

• AI data Provenance: building "data lineage proof" with cryptographic verification to confront regulatory and audit pressures.
• Public datasets and civilizational archives: anchoring censored files and cultural heritage to construct irreplaceable, indelible memories.
• Credible archiving and compliant certification: achieving credible self-proof through Hash certification to provide high-level digital notarization.
• Integration of ZK/TEE/DID technologies: resolving privacy tensions, upgrading from a single "storage protocol" to "trustworthy data infrastructure."
• Invisibility of product routes: providing S3 compatible APIs and fiat billing, allowing users to directly purchase "trustworthy archiving" services.

AI storage and decentralized storage serve different purposes: one pursues extreme efficiency, providing fuel for our march toward the future; the other defends silent memory, preserving our right to look back into the past. The current market unreservedly rewards efficiency, making decentralized storage appear silent or even in collapse; however, as the AI era further amplifies data monopoly, copyright disputes, and the fragility of historical memory, decentralized storage may encounter a revaluation of value in its role as a "trustworthy cold layer." Memories that cannot be easily erased by platforms, companies, or any single authority may transform from romantic idealism and marginal belief to a necessary infrastructure.

Disclaimer: This article has utilized AI tools such as Claude Opus 4.8, ChatGPT-5.5, Qwen 3.7 to assist in the creation process, and the author has made efforts to proofread and ensure the information is true and accurate, but oversights are inevitable; your understanding is appreciated. It should be particularly noted that the content of this article is solely for information integration and academic/research exchanges, does not constitute any investment advice, and should not be regarded as recommendations for the buying or selling of any tokens.