Kioxia launches liquid-cooled SSDs for AI servers, with PCIe 6.0 CM10 delivering 92% faster reads

3 Sources

Share

Kioxia has unveiled two new enterprise SSD lines designed to tackle thermal challenges in AI infrastructure. The NX1 Series introduces direct liquid cooling to PCIe 5.0 drives, while the CM10 Series debuts PCIe 6.0 technology with performance gains up to 92%. Both leverage advanced BiCS Flash memory and support cold-plate cooling for GPU-equipped AI servers where traditional airflow is no longer sufficient.

News article

Kioxia SSD Lines Address Cooling Needs in AI Infrastructure

Kioxia has introduced two enterprise storage solutions that address mounting thermal challenges in hyperscale data centers and GPU-equipped AI servers. The NX1 Series marks the company's first PCIe 5.0 NVMe SSDs officially designed for direct liquid cooling, while the CM10 Series debuts as Kioxia's first PCIe 6.0 enterprise SSDs. Both product lines reflect the storage industry's response to AI workloads that generate unprecedented heat loads, particularly in systems where GPUs, CPUs, and networking components consume most available airflow capacity

1

2

.

NX1 Series Brings Cold Plates to E1.S Form Factor

The NX1 Series succeeds Kioxia's XD8 line and introduces a new in-house controller alongside eighth-generation BiCS Flash TLC NAND using CMOS directly Bonded to Array architecture. Sequential write performance improves by up to 38% compared to the previous generation, with random write performance gaining up to 20%

3

. The E1.S form factor comes in two variants: a 9.5mm version that supports both conventional airflow and cold-plate connection, and a thicker 15mm air-cooled option. Capacities range from 1.92TB to 15.36TB for one-drive-write-per-day models

1

.

The NX1 Series complies with the OCP Datacenter NVMe SSD Specification 2.6 and supports Flexible Data Placement, which allows software to organize data in ways that reduce internal write amplification and garbage collection. This capability potentially improves endurance and consistency in high-performance AI servers. Optional self-encrypting models provide partial TCG Opal compliance

1

. Neville Ichhaporia, senior vice president at Kioxia America, noted that "E1.S form factor SSDs are becoming increasingly important for AI and hyperscale system designs, keeping high-speed flash storage close to the CPU and GPU"

3

.

CM10 Series Leaps to PCIe 6.0 With Massive Performance Gains

The CM10 Series represents a more aggressive technology jump, leveraging PCIe 6.0 to deliver 92% higher sequential read performance and approximately 85% higher random read performance compared to the PCIe 5.0-based CM9 Series

2

. These next-generation SSDs for AI incorporate BiCS Flash generation 10 TLC memory technology, achieving 332 stacked layers with the highest bit density and smallest chip size for 1Tb products. The CM10 Series supports NVIDIA CMX context memory storage, expanding GPU memory on NVIDIA Rubin GPUs for AI inference—a critical capability as AI systems handle millions of tokens and require high-performance Key Value cache storage

2

.

Available in E3.S, E1.S, and 2.5-inch form factors, the CM10 Series offers capacities from 1.60TB up to 61.44TB with read-intensive (1 DWPD) and mixed-use (3 DWPD) options. The E3.S and E1.S 9.5mm variants support cold-plate liquid cooling alongside traditional air cooling. The drives comply with NVMe 2.1 and OCP Datacenter NVMe SSD Specification 2.7, while including comprehensive security features

2

.

Why Direct Liquid Cooling Matters for AI Infrastructure

Extending liquid cooling loops to storage addresses a fundamental shift in data center design. As GPUs and accelerators dominate thermal budgets in AI infrastructure, relying solely on high-speed fans becomes impractical. Direct liquid cooling through cold plates enables denser system configurations while reducing dependence on airflow, which matters for organizations building large-scale AI deployments where space and power efficiency directly impact operational costs

1

.

Both product lines are currently sampling to select hyperscale customers, with demonstrations planned at the Future of Memory and Storage conference on August 4-6 in Santa Clara, California

2

3

. Technologies developed for these enterprise drives may eventually influence workstation and high-performance consumer SSDs as thermal management becomes critical across computing segments

1

. Watch for broader adoption of liquid-cooled storage as AI systems scale and thermal density continues to climb.

Today's Top Stories

© 2026 TheOutpost.AI All rights reserved