Innodisk Expands AI Memory Portfolio and Submits Data Center SSDs for Sampling

Table of Contents

• You might want to know


• Main Topic


• Key Insights Table


• Afterwards...

You might want to know

• How is Innodisk adapting its enterprise SSD portfolio to different storage requirements in AI data centers?


• How could advanced server memory and CXL expansion modules help cloud providers use computing resources more efficiently?

Main Topic

Industrial memory and storage company Innodisk International (宜鼎國際, 5289-TW) is expanding its presence in the data center market with enterprise solid-state drives (SSDs) and memory modules for AI servers. Its products are aimed at cloud service providers and emerging cloud operators, often referred to as Neocloud, whose workloads include AI model training, inference, and high-performance computing. Innodisk says its data center SSDs, including the 3TS9-P, 4TS2-P, and 5TS-P, have all entered the formal sampling stage.

The company’s expansion reflects a change in the role of storage in AI infrastructure. As AI data center construction continues to gain momentum, SSDs are no longer simply devices for retaining files. Storage performance, reliability, and manageability can also affect how effectively AI servers handle data-intensive workloads. Different parts of a server environment need different storage characteristics, so Innodisk is positioning its portfolio to address both server boot and operating system functions, as well as local storage for workload data.

A central feature of Innodisk’s strategy is to offer storage products for distinct data center roles while pairing them with memory solutions designed for demanding AI workloads. The company’s enterprise SSD lineup spans multiple interfaces and form factors, and it offers firmware customization according to customer system requirements. This combination is intended to give data center operators options that can be matched to their system designs rather than relying on a single, general-purpose drive configuration.

Enterprise SSDs for boot and operating system duties

For server boot and operating system drives, Innodisk offers the PCIe Gen5 E1.S 5TS-P, the Gen4 4TS2-P, and the SATA III 3TS9-P SSD series. These products emphasize reliability and data integrity, both important for systems expected to operate continuously and support business-critical services. The company can also customize firmware to fit customer system requirements, allowing drive behavior to be tailored to a particular server platform or deployment.

The boot-drive products include enterprise power-loss protection (PLP) and end-to-end data path protection (ETEP). PLP is designed to protect data and drive operations when power is interrupted, while ETEP addresses the integrity of data as it moves through the storage path. Such features are relevant in data center environments, where unexpected interruptions or data errors can affect system availability and the reliability of applications.

Among the listed boot-drive options, the E1.S 5TS-P also supports event logging and self-monitoring. These functions can help operators observe drive conditions and review events during extended operation. In a large server fleet, monitoring and diagnostic capabilities can support maintenance planning and troubleshooting, complementing the drive’s focus on reliability.

Local storage for AI data and high-throughput access

Local storage serves a different purpose from a boot drive. It can hold important data used directly by AI systems, including model weights, vector databases, and inference datasets. These workloads can involve frequent access to large volumes of information, making performance and data placement important considerations. Innodisk’s PCIe Gen5 5TS-P series is designed for this local-storage role and reaches a maximum read speed of 14GB/s and a maximum write speed of 10GB/s.

The 5TS-P is available in U.2 and EDSFF E1.S form factors. Offering both formats gives customers flexibility when selecting hardware for their system designs, while also addressing factors such as high-speed data transfer, heat management, and maintenance. The choice between form factors can be influenced by server architecture and deployment needs, so availability in both formats is intended to accommodate a range of data center configurations.

Both the U.2 and EDSFF E1.S products have received OCP Inspired™ certification and support NVMe-MI out-of-band management. These capabilities are intended to assist enterprises and cloud operators with large-scale deployment and administration. Out-of-band management can provide a way to monitor and manage storage devices separately from the primary data path, which is useful when operating and maintaining a substantial fleet of servers.

Expanding memory options for AI servers

Alongside storage, Innodisk is broadening its AI server memory portfolio. It has introduced DDR5 12800 MRDIMM and 8000 RDIMM series. The company describes these products as addressing inference workloads that require high bandwidth and large-capacity memory, including long-context applications and Agentic AI. As AI applications work with longer inputs and more complex sequences of actions, memory capacity and bandwidth can become important elements of system design.

Innodisk also offers CXL expansion modules. According to the company, a single module can provide 128GB of capacity and 32GB/s of bandwidth without requiring additional DIMM slots. The modules support memory pooling across multiple compute nodes. This approach can make memory resources available across a wider system rather than limiting each resource to one node, potentially improving the utilization of memory that might otherwise remain idle.

For cloud providers, the stated value of memory pooling is the possibility of improving resource allocation within existing hardware architectures. Instead of relying solely on adding memory through conventional DIMM slots, operators can use CXL-based expansion to support a different approach to capacity and sharing. Its practical impact will depend on system compatibility, workload behavior, and deployment design, but the technology represents one way to address the growing memory needs of AI infrastructure.

Product availability and development roadmap

Innodisk reports that the data center SSDs 3TS9-P, 4TS2-P, and 5TS-P have all been formally submitted for sampling. The company can provide customized services according to customer system requirements. Sampling allows potential customers and system partners to assess products in the context of their own platforms and deployment plans.

For AI memory, Innodisk currently offers products in its CXL series. It plans to introduce the DDR5 MRDIMM and 8000 RDIMM series in subsequent stages, continuing to expand its data center portfolio. This sequence combines currently available CXL products with future memory offerings, while the SSD sampling program advances the company’s enterprise storage lineup.

Together, these developments show how the company is addressing several connected needs in AI data centers: reliable boot storage, high-throughput local storage, higher-bandwidth server memory, and more flexible memory allocation. The products are designed for cloud service providers and Neocloud operators, but their suitability will ultimately depend on customer evaluation, system compatibility, and the demands of particular workloads.

Key Insights Table

Afterwards...

As AI data centers continue to evolve, operators and technology providers will need to assess how storage, memory, and compute resources work together. Future exploration may focus on improving the efficiency of data movement between SSDs, memory, and processors, as well as evaluating how different storage tiers affect training and inference performance. The balance between capacity, speed, reliability, and power use will remain important as systems scale.

Memory pooling and CXL-based expansion also merit continued evaluation. Further work could examine how these systems perform across different applications, how they can be managed securely at scale, and how they affect the utilization of existing hardware. In parallel, advances in drive monitoring, firmware customization, and out-of-band management may help operators maintain reliability across large deployments.

The broader opportunity is to develop data center architectures that coordinate storage and memory more flexibly, while preserving reliability and making efficient use of deployed resources. Continued testing across real-world AI workloads, along with standards-based compatibility and operational evidence, will help clarify which approaches deliver the greatest practical benefit to cloud providers and other data-intensive organizations.