KIOXIA America, Inc. announced the launch of its CM10 Series solid-state drives on August 3, 2026, delivering up to approximately 92% higher sequential read performance compared to the previous generation. The new drives represent the company's first PCIe 6.0 enterprise SSDs and include direct-cold-plate liquid cooling capability, designed to handle demanding AI workloads such as AI inference and Key Value cache. Built with KIOXIA's latest BiCS FLASH generation 10 TLC memory, the drives are designed to support the NVIDIA CMX context memory storage solution while offering significant improvements in performance, power efficiency, and cooling flexibility.
The CM10 Series delivers up to approximately 85% higher random read performance alongside the sequential read gains, helping to speed up data-intensive AI inference and enterprise applications while boosting overall system efficiency. The drives are currently being sampled to select customers. In a separate announcement, KIOXIA also introduced UFS 5.0 embedded flash memory solutions built on the latest JEDEC UFS 5.0 standard, leveraging the MIPI M-PHY v6.0 physical layer and UniPro v3.0 protocol with HS-Gear6 operation. These devices support theoretical interface speeds reaching up to 46.6 Gb/s per lane and approximately 10.8 GB/s of effective dual-lane read/write performance.
According to Neville Ichhaporia, senior vice president and general manager of the SSD business unit at KIOXIA America, Inc., "As AI infrastructure evolves, storage must deliver not only higher performance but also greater efficiency and deployment flexibility." The company said the CM10 Series reflects its continued commitment to advancing flash storage that helps customers build more capable, scalable, and efficient AI infrastructure. KIOXIA stated that its UFS 5.0 solutions help enable faster application loading, greater responsiveness, and unprecedented user experience across on-device AI applications when compared to UFS 4.1.
The report explains that demand for high-performance Key Value cache storage is growing rapidly as AI models continue to scale toward trillions of parameters and agentic AI context windows expand to millions of tokens. The NVIDIA CMX solution extends effective GPU memory with a shared, pod-level context tier optimized for ephemeral KV cache, the vendor said. By leveraging ultra-fast flash storage, CMX makes enterprise SSDs an increasingly important component of AI infrastructure, positioning the CM10 Series to meet these evolving requirements with the performance, capacity, and endurance needed for large-scale AI deployments. For mobile and edge devices, the analysis notes that as AI workloads continue moving onto devices, storage performance is becoming increasingly important—large language models, multimodal AI, advanced imaging, and real-time inference require significantly faster movement of data between storage and compute.
KIOXIA continues to advance its flash memory portfolio to meet the growing demand for higher capacity, greater performance, and improved efficiency in both enterprise AI infrastructure and AI-enabled mobile and edge applications. The direct-cold-plate liquid cooling capability in the CM10 Series enables more efficient cooling deployments for next-generation AI infrastructure, addressing the thermal challenges that come with higher-performance storage. The company's dual announcements signal a strategy to serve AI workloads across the spectrum, from massive data center deployments to on-device edge computing.