The More Impressive AI Agents Become, the More Valuable Storage Is! SK Hynix's NAND Business Plans $100 Billion US IPO

According to ZhiTong Finance APP, media sources citing insiders have reported that Solidigm, a subsidiary of SK Hynix Inc., one of the world’s largest storage chip giants, is considering an independent initial public offering (IPO) in the US stock market as early as next year, planning to list independently in the United States.

The expansion of agent applications represented by Muse and Astra is pushing the incremental demand for AI infrastructure from model computation to task execution, high-efficiency context management, and massive-scale data access. As a result, enterprise-grade solid-state drives (SSD) have become a major focus for storage investment. Against this backdrop, Solidigm, a subsidiary of SK Hynix, is considering an IPO in the US as early as 2027 and is in talks with potential advisors regarding listing matters.

According to insiders, the world’s second-largest DRAM/NAND storage chip manufacturer is negotiating with potential advisors about listing its NAND flash business subsidiary on the US stock market. As the related information is still not public, these sources requested anonymity.

Some insiders stated that the listing could value Solidigm at as much as a staggering $100 billion. They noted that discussions are ongoing and details such as the IPO timeline and valuation may change.

Some US media outlets first disclosed details about the bank discussions and proposed listing schedule, citing sources who requested not to be identified. A Solidigm spokesperson declined to comment.

Solidigm’s industry position aligns precisely with the trend of AI investment shifting from accelerators to full data center systems. Its core business involves NAND flash and enterprise-grade SSDs, with products entering scenarios such as cloud computing, servers, and data centers. On August 5th, CoreWeave announced a multi-year strategic agreement with Solidigm for prioritized enterprise SSD capacity supply, explicitly stating that storage has become a key bottleneck for AI platform capacity planning. This also means that large AI cloud service providers are incorporating storage supply into their long-term infrastructure plans to ensure synchronous expansion of computation, networking, and storage. For Solidigm, such partnerships help increase demand visibility and provide actual customer backing for the independent valuation of its enterprise storage business; the announcement did not disclose the contract value or specific procurement volume.

In the age of AI inference, the core value chain of enterprise SSDs lies in the comprehensive and seamless integration of NAND, controllers, firmware, and system validation—delivering stable data access capabilities for customers: When costly AI accelerators require continuous data supply, the value of storage is reflected in capacity provision, computational utilization, and the operating costs of the entire system.

In 2021, SK Hynix acquired Intel’s flash memory business and rebranded it as “Solidigm,” officially launching the company. According to its website, the company manufactures large-scale data NAND storage products for data centers, with some devices as compact as a deck of cards but with capacities reaching up to 122TB.

In August this year, the company announced an agreement with CoreWeave, a so-called “New Cloud” large-scale AI cloud enterprise, to supply enterprise SSD storage capacity to power CoreWeave’s leading AI cloud platform. Other customers, according to its website, include VAST Data, Dell Technologies, and Chinese internet giant Tencent Holdings.

It is understood that this California-based storage chip leader, headquartered in Rancho Cordova, operates 13 business locations worldwide, including in Mexico, Canada, and China, and has over 2,000 employees.

Storage chip components for AI data center server clusters remain the clearest supply bottleneck in the AI computing power industry chain. According to market research firm TrendForce, cumulative increases in server DRAM contract prices will reach about 270% by 2026, and enterprise SSD prices will rise about 235%; in 2027, HBM contract prices could still increase by 70%–140%. These figures reflect the dual impacts of AI computing capacity expansion and rising storage costs. TrendForce’s latest estimates show that DRAM and NAND combined will account for 47% of leading cloud service providers’ capex in 2026, rising to 68% in 2027, reflecting both increased volumes and higher prices.

In terms of stock prices, as of the close of US equities on September 25, 2026, and measured by year-end 2025 close (local currency, excluding dividends), US storage chip leader Micron (MU.US) has risen approximately 279.2% year-to-date; SK Hynix’s Korean-listed shares have gained about 186.0%.

Who is Solidigm under SK Hynix?

Solidigm is a US-based enterprise data storage company owned by SK Hynix, specializing in NAND flash-based SSDs and supporting storage technologies, focusing on data centers, cloud computing, and edge AI. It operates as an independent subsidiary headquartered in Rancho Cordova, California, and, according to its official website, has 13 business locations globally and over 2,000 employees.

Its business originates from Intel’s former NAND flash and SSD businesses. In 2020, SK Hynix announced its intention to acquire the business for a total consideration initially agreed at about $9 billion, completing the first phase in December 2021, and established Solidigm to take over product development, manufacturing, and sales of Intel’s SSD business; the second phase, covering remaining NAND technology and manufacturing operations, was completed on March 27, 2025. As a result, Solidigm inherited Intel’s long-accumulated enterprise storage technology, engineering teams, and customer relationships.

Specifically, it delivers complete enterprise-grade SSD products to clients, with SSD value based on the synergy of flash medium, controllers, firmware, and system design. NAND retains data after power-off; controllers and firmware manage read/write, error correction, wear leveling, and performance tuning; enterprise products must meet requirements for continuous operation, data integrity, write endurance, and consistent response times. Thus, Solidigm’s capabilities cover storage hardware, firmware, and supporting software, optimizing products for customers’ actual workloads.

Solidigm’s main business needs to be clearly distinguished from SK Hynix’s current core business—HBM. HBM is high-bandwidth DRAM mainly providing ultra-fast data access for accelerators like GPUs; Solidigm’s main products are part of the NAND-based storage system, used for large data capacity retention and partially serving as reusable inference cache within suitable software architectures. The SK Hynix group covers DRAM, HBM, NAND, and SSD businesses, and Solidigm represents a major enterprise-level flash storage platform under this group, but not all of the group’s NAND storage is classified as Solidigm.

The more advanced and capable AI agents become, the greater the need for large-scale expansion of storage chips

The fundamental change brought by Muse and Astra is that a single user instruction can initiate multi-stage, continuously running workflows. Meta revealed that Muse runs in a dedicated secure VM, able to execute tasks across applications; Astra has enhanced computer operations, programming, and complex professional work capabilities. A research or development task may sequentially trigger information retrieval, file reading, code execution, model inference, and result validation, generating intermediate outcomes that need to be preserved.

From an engineering perspective, GPUs and dedicated AI accelerators handle model computation, high-performance CPUs support browsers, VMs, tool execution, and scheduling; data and indexes for enterprise knowledge bases and retrieval-augmented generation (RAG), work files, and audit logs expand both memory and persistent storage demand. As agent adoption increases, both “computing capability” and “data processing capability” are expected to grow.

The second wave of storage demand comes from cache management required by longer context and more concurrent tasks. In mainstream Transformer inference architectures, input is pre-filled and output is generated step-by-step; key-value cache (KV Cache) retains reusable intermediary computation states. High-frequency data needed for active generation is stored in HBM, system DRAM serves as the buffer, and reusable cache can be tiered into SSDs and shared flash according to access frequency and latency requirements, later preloaded back into memory. Nvidia’s CMX architecture has specifically proposed adding a flash layer for inference context between GPU memory and conventional shared storage. The economic significance is increasing the volume of retainable, reusable context, reducing redundant computation and data wait times, and thus supporting more concurrent tasks. Enterprise-grade SSDs, therefore, gain new use-cases in the inference workflow.

Solidigm’s high-density products are clearly tailored for these requirements. Its D5-P5336 has a maximum capacity of 122.88TB, utilizing QLC technology and targeting data lakes, object storage, and other large-capacity, read-intensive workloads. For data center operators, higher single-drive capacity means fewer devices are needed for the same capacity, optimizing rack space, power, and cooling costs; workloads requiring higher write performance or stricter latency are covered by alternative SSD products and software configurations.

From an investment perspective, Solidigm’s strong growth prospects are driven by the expansion of AI data volumes, enterprise storage upgrades, and ongoing client pursuit of lower per-unit capacity costs and higher system efficiency. Improvements in model efficiency reduce the cost of completing tasks, attracting more workloads to large-scale AI inference systems; as user and task volumes expand faster than resource savings for each task, demand for computation, storage, networking, and power continues to grow in sync.