Innodisk Targets AI Servers with Customizable SSDs and CXL Memory

Innodisk announced on November 12, 2026 an expansion of its AI data-center portfolio, pairing enterprise SSDs with DDR5 and CXL memory for cloud service providers and neocloud operators handling AI training, inference and high-performance computing.

The 3TS9-P, 4TS2-P and 5TS-P SSD series are available for sampling and system-specific customization. Innodisk also offers its CXL series and plans to showcase the portfolio at SC26 in Chicago on November 16–19, 2026.

CC Wu, general manager of Innodisk’s Flash Division, said the company is extending its industrial embedded-computing background into the AI data center supply chain. Customized firmware, security features, system-compatibility work and long-term supply commitments are intended to accelerate deployment, reduce integration risk and maintain infrastructure stability.

Storage for Boot Drives and Local AI Data

The storage portfolio separates boot and operating-system drives from local storage for AI model weights, databases and inference datasets. The boot/OS range comprises the E1.S 5TS-P, 4TS2-P and 3TS9-P, with customizable firmware, power-loss protection and end-to-end data-path protection.

For local data, Innodisk rates the 5TS-P series at up to 14GB/s sequential reads and 10GB/s sequential writes. The E1.S model combines a PCIe Gen5 x4 interface with NVMe 2.0 and 3D eTLC flash, and comes in capacities from 1.92TB to 7.68TB. Its endurance rating is one drive write per day, with a five-year warranty.

The E1.S product page also lists event logs, self-monitoring, online firmware updates and out-of-band management for large-scale deployments. The drive carries OCP Accepted recognition for the OCP Datacenter NVMe SSD Specification v2.0. A 15mm asymmetrical heatsink and thermal throttling support sustained operation by limiting heat buildup and adjusting performance to prevent overheating.

Memory Targets Capacity and Expansion Limits

Innodisk positions the memory range for long-context and agentic AI inference, saying these workloads increase the need for memory capacity and bandwidth while requiring lower latency. The lineup combines DDR5 server memory with CXL expansion modules.

The DDR5 12800 Multiplexed Rank DIMM runs at 12,800 MT/s and comes in capacities from 32GB to 128GB. Multiplexed registering clock driver and data buffer components enable simultaneous access to two ranks. Innodisk reports a 60% bandwidth increase over DDR5 8000 MT/s RDIMM.

The DDR5 8000 RDIMM is listed at 8,000 MT/s in capacities from 16GB to 64GB. It uses registered ECC memory, while TVS diode and eFuse protection guard against power surges and overvoltage.

These DDR5 products were already part of Innodisk’s lineup. The company’s newsroom dates the MRDIMM introduction to July 30, 2026, and the DDR5 8000 RDIMM launch to September 29, 2026, so the November announcement incorporates earlier memory releases into the broader data-center portfolio.

The CXL Memory Module comes in 64GB, 96GB and 128GB versions, using an E3.S 2T form factor and a PCIe Gen5 x8 interface. Innodisk rates it at up to 32GB/s of additional bandwidth. Supporting CXL 1.1 and 2.0 through CXL.mem and CXL.io, it expands memory without additional DIMM slots and enables pooling across servers, allowing compute nodes to share memory resources.

The portfolio gives Innodisk a broader AI-server offer spanning storage throughput, memory capacity and drive management. Its near-term business significance rests on customer qualification of the sampling SSDs, with firmware customization, compatibility work and long-term supply support connecting the company’s industrial-memory background to its data-center ambitions.