White Line | In addition to SK Hynix, which other tracks and companies in the storage sector are worth paying attention to?

Source | WhiteLine

Compiled by | WuShuo Blockchain

Seeking direction, before changes arrive

WhiteLine is produced by the WuShuo team, moving from Crypto to the broader capital market, focusing on trend changes in the AI era.

In this episode, host Minta tells about a track being repriced by AI: storage. In the past, storage was often seen as a typical cyclical industry, with the market mainly focusing on consumer electronics demand, inventory cycles, and product prices; but after the rapid growth of AI training and inference demand, storage is being recast into the framework of data centers and computing infrastructure.

This episode first deconstructs the three-layer industry map of AI storage, and further discusses the positions of companies such as Micron, SanDisk, Kioxia, Penguin, and Adeia in different links, to help everyone understand: AI storage is not a single HBM logic, but an industry network composed of product upgrades, system solutions, and upstream IP.

The core conclusion is that storage has not completely escaped the cycle, but the core variables within the cycle have changed. What is truly worth paying attention to is not "who is the next SK hynix", but whether AI continues to bring data center demand, whether the product positions of enterprise SSDs and high-capacity NAND continue to move up, and whether price, supply, and cost advantages can still support profits. Storage in the AI era is moving from a pure price cycle to a more complex infrastructure cycle.

The following is a summary of this episode's video:

1. AI Reshapes the Storage Cycle: From HBM to System and Infrastructure Upgrades

Over the past few months, discussion of the storage sector has heated up. During the price increase phase, the market attributed it to new demand from AI inference; when sentiment cooled, the cyclical nature was repeatedly emphasized. The core question is: Is storage still a purely cyclical industry?

The cycle has not disappeared, but the driving factors have changed. Traditional logic revolves around consumer electronics demand, inventory cycles, and product prices. In the current framework, AI training and inference incorporate storage into the infrastructure system.

AI storage forms a three-layer structure: HBM for high-speed data transfer to GPUs, represented by SK hynix and Micron; NAND, enterprise SSDs, and high-capacity Flash focus on low-cost, high-capacity storage, with SanDisk and Kioxia as core participants; CXL and other technologies are used for memory expansion, scheduling, and sharing, representing a systematic organization of memory resources.

The industry's focus has expanded from a single HBM to two main lines: capacity cost optimization and system architecture upgrades.

2. Micron's Earnings Release Signals: DRAM and NAND Both Tight

Micron's Q3 FY2026 revenue was $41.4 billion, with next-quarter guidance of $49 billion to $51 billion. DRAM revenue was $31.3 billion, accounting for about 76%, and NAND revenue was $9.9 billion, accounting for about 24%.

DRAM prices rose more than 60% YoY, NAND rose more than 80%, reflecting overall tight supply and demand in storage, with data center demand expansion and supply constraints jointly driving prices up.

On the NAND side, SanDisk's Q3 revenue was $5.95 billion, up 97% QoQ, and data center business grew 233%. The demand structure is shifting from consumer electronics to data centers, enterprise SSDs, and AI inference.

Market divergence remains: Optimistic expectations come from AI inference and enterprise demand as well as long-term contracts supporting cycle stability; cautious views hold that NAND's commodity nature remains strong, and the risk of price and profit pressure after supply release still exists.

3. NAND Cost, CXL Scheduling, and Adeia IP System

The core variables of storage are concentrated on supply, demand, and price. AI inference strengthens data center demand, raising the weight of enterprise SSDs and high-capacity NAND, but supply changes still dominate cycle fluctuations.

Kioxia can be seen as the Japanese version of SanDisk, focusing on NAND and SSDs, with core competitiveness in cost efficiency. Its advantages lie in deep participation in manufacturing, approximately 60% capacity share, cost per GB about 20% lower, and continuous improvement in process efficiency by about 20%.

CXL defines the memory resource scheduling standard, allowing CPU and GPU to access external memory, alleviating the memory imbalance issue in data centers. Penguin Solutions integrates CXL memory expansion cards with inference servers to optimize KV cache, improving memory capacity and GPU utilization.

Adeia focuses on the IP licensing system, holding more than 13,750 patents. Among them, Hybrid Bonding is used to improve chip interconnect density and efficiency, serving advanced packaging and storage evolution. Kioxia, Micron, and others have signed related licensing agreements.

4. Adeia: Technical Infrastructure for Packaging and Interconnect Layers

Traditional chip stacking relies on underfill and solder connections, which have issues of space occupation and signal loss. Hybrid Bonding directly bonds metal and dielectric materials, achieving shorter paths and higher interconnect density.

This technology covers HBM, NAND, and 3D stacking. It is a key foundational capability for storage moving from single-chip expansion to multi-layer stacking, affecting the connection efficiency and transmission loss between logic and storage.

The storage industry structure is gradually showing a layered characteristic: product layer, system layer, and IP/infrastructure layer coexist. Adeia is located at the bottom IP layer, participating in the entire industry upgrade through patent licensing.

The storage structure in the AI era is network-distributed, with value coming from product delivery, system scheduling, and underlying technology licensing. The cyclical nature of storage remains, but the core has shifted to an infrastructure cycle.