The data center storage landscape is undergoing its most significant architectural shift in a decade, driven by the insatiable demands of artificial intelligence workloads. As AI moves from experimental model training to大规模 inference and real-time agentic operations, the underlying storage infrastructure must evolve to eliminate bottlenecks and deliver unprecedented performance. At this week’s Future of Memory and Storage (FMS) 2025 event in Santa Clara, California, KIOXIA is responding to this challenge with a comprehensive showcase of next-generation PCIe 6.0 NVMe SSDs, positioning itself at the forefront of the AI-driven storage revolution. The company, recognized as the inventor of flash memory, is leveraging its latest BiCS FLASH technology to deliver products that promise not just incremental improvements but a fundamental leap in what enterprise storage can achieve for AI infrastructure.
KIOXIA Showcases Next-Gen PCIe 6.0 NVMe SSDs for AI Infrastructure at FMS 2025
KIOXIA’s presence at FMS 2025, running from August 4 through 6, is built around three core pillars that the company identifies as defining modern AI workloads: performance, scalability, and efficiency. These pillars are not abstract marketing concepts but represent tangible engineering challenges that organizations face when deploying AI from the enterprise data center to the cloud and high-performance computing (HPC) environments. By delivering storage solutions that address these specific areas, KIOXIA aims to help companies remove critical performance bottlenecks, improve overall operational efficiency, and scale their AI deployments more effectively. The strategy is clear: storage can no longer be a passive repository; it must be an active, high-performance component of the AI compute fabric.
The KIOXIA GP1 Series: 10 Million IOPS for GPU-Direct Workloads
Perhaps the most eye-catching announcement from KIOXIA at FMS 2025 is the new KIOXIA GP1 Series PCIe 6.0 NVMe SSDs. These drives are engineered for one specific and demanding purpose: feeding data to GPUs at the speed they require. The headline specification is a staggering 10 million random read input/output operations per second (IOPS) in a 512-byte random read configuration. This level of performance is optimized for GPU direct access, a critical capability for AI workloads where data must be moved directly to and from GPU memory with minimal CPU intervention. The GP1 Series effectively acts as a high-speed data cache for GPUs, eliminating the storage latency that often stalls complex model training and inference tasks. For enterprises deploying large-scale AI clusters, the ability to achieve 10 million IOPS from a single storage device represents a paradigm shift in how data pipelines are architected.
What performance improvements does the CM10 Series offer versus the CM9 Series for sequential and random reads?
The KIOXIA CM10 Series PCIe 6.0 enterprise SSDs deliver a 92% improvement in sequential read performance and approximately 85% higher random read performance compared to the previous-generation CM9 Series. This dramatic generational leap is made possible by the combination of the PCIe 6.0 interface and the new BiCS FLASH generation 10 TLC memory, making the CM10 Series the first SSDs to use this advanced flash technology with 332 stacked layers.
The KIOXIA CM10 Series: A Generational Leap with BiCS FLASH Gen 10
Alongside the GP1 Series, KIOXIA is showcasing its new KIOXIA CM10 Series PCIe 6.0 enterprise SSDs, which are purpose-built for modern AI workloads. The performance gains over the previous generation CM9 are substantial and quantifiable. The CM10 Series offers 92% higher sequential read performance and around 85% higher random read performance. These are not merely theoretical improvements; they translate directly into faster data loading for AI models, reduced training times, and more responsive inference pipelines. The CM10 Series achieves this by being the first SSDs to utilize the latest BiCS FLASH generation 10 TLC flash memory technology, which features a staggering 332 stacked layers. This advanced 3D NAND architecture, combined with CMOS directly Bonded to Array (CBA) technology, allows KIOXIA to deliver higher density and performance while maintaining power efficiency. For data center operators, the CM10 series represents a clear upgrade path for modernizing storage infrastructure to handle the most demanding AI and HPC applications.
Which KIOXIA product families showcased at the event are specifically promoted for GPU-direct access or near-GPU workloads?
The KIOXIA GP1 Series is specifically promoted for GPU-direct access and near-GPU workloads, delivering 10 million random read IOPS. Additionally, the KIOXIA CM10 Series is showcased for high-performance enterprise AI workloads, and the Context Memory Caching demonstration featuring the CM9 Series focuses on maximizing AI inferencing. The NX1 Series with liquid cooling is also targeted at dense AI environments where thermal management is critical near powerful GPUs.
NX1 Series: Direct Liquid Cooling for AI Density
As AI clusters become denser and more power-hungry, thermal management has become a critical design constraint. KIOXIA addresses this challenge with the NX1 Series SSDs, which support direct liquid cooling. These drives are engineered for data centers that are transitioning to liquid-cooled architectures to handle the immense heat generated by high-performance GPUs and accelerators. The NX1 Series allows storage to be placed in closer proximity to compute resources without fear of thermal throttling, enabling more compact and efficient server designs. This is a practical solution for operators looking to maximize their compute density per rack while maintaining optimal performance and reliability. The KIOXIA NX1 Series is displayed at the Liquid Cooling SSDs for AI exhibition at booth #307, demonstrating a clear commitment to infrastructure-level innovation.
What demos or sessions relate to scaling AI inference and vector databases at the Kioxia booth?
KIOXIA is hosting several demonstrations at its booth specifically focused on AI inference and vector databases. These include a Context Memory Caching demonstration using KIOXIA CM9 Series SSDs to maximize AI inferencing, and an AI RAG (Retrieval-Augmented Generation) demonstration titled “Improving Scalability for Vector Databases” featuring KIOXIA AiSAQ open-source software. These sessions show how storage directly impacts the performance and accuracy of AI applications that rely on rapid data retrieval.
Keynote and Panel: Redefining Data Center Storage with NVIDIA
Beyond the hardware demonstrations, KIOXIA’s thought leadership is on full display at FMS 2025. The company’s keynote presentation, titled “Redefining Data Center Storage for the AI Era with Flash Memory and SSDs,” is scheduled for Tuesday, August 4 at 11:10 am. This session will explore AI’s transformation from model training to inference and how this shift has dramatically altered the requirements for both memory and storage. In a significant show of industry collaboration, KIOXIA will also host a special session with NVIDIA titled “Balancing Memory and Storage to Operate AI Factories and Agents at Scale.” This panel and Q&A format will feature executives from NVIDIA and other industry leaders, offering a rare glimpse into the strategic thinking required to build and operate large-scale AI factories. The presence of this session underscores the critical role that storage vendors now play in the AI ecosystem, no longer relegated to the periphery but central to the compute architecture.
What flash memory technology and layer count does the CM10 Series use?
The KIOXIA CM10 Series is the first SSD to use the latest BiCS FLASH generation 10 TLC flash memory technology, which features 332 stacked layers. This advanced 3D NAND architecture leverages CMOS directly Bonded to Array (CBA) technology to achieve enhanced lateral shrink, enabling higher density and performance within the same physical footprint.
Booth #307: A Deep Dive into KIOXIA’s AI Storage Ecosystem
KIOXIA has constructed a massive two-level booth (#307) at FMS 2025, organized into specific exhibition zones that provide a comprehensive overview of its AI storage ecosystem. The structure of the booth itself tells the story of a company transforming from a component provider to a solution architect. Each zone targets a specific pain point in AI infrastructure. In addition to the previously mentioned Liquid Cooling SSDs and PCIe 6.0 Enterprise SSDs, the booth features demonstrations on Context Memory Caching for Maximizing AI Inferencing using the CM9 Series, which directly addresses the need for low-latency data access during inference. There is a dedicated zone for AI RAG (Retrieval-Augmented Generation) that highlights KIOXIA’s AiSAQ open-source software, designed to improve scalability for vector databases. This is a strategic move, as vector databases are a foundational technology for many modern AI applications, and their performance is heavily dependent on storage speed. Other zones showcase High Capacity QLC Storage for Fast Retrieval of Large Datasets, featuring a massive 245.76 TB KIOXIA LC9 Series NVMe SSD, the new UFS 5.0 for AI-ready high-performance memory solutions at the edge, and a CXL Memory Module with KIOXIA XL-FLASH that expands the memory hierarchy. Finally, a dedicated area presents the underlying technology: BiCS FLASH generation 10 Flash Memory, emphasizing the advanced scaling and CBA technology with 332 layers, and a High Capacity Package with low-latency QLC Flash memory featuring a 32-die stack BGA package designed for AI applications.
The breadth of the demonstrations at booth #307 makes it clear that KIOXIA is not merely refreshing its product line but is systematically re-engineering its entire portfolio to serve the AI-centric data center. The inclusion of open-source software, new form factors like UFS 5.0, and CXL-based memory expansion indicates a holistic strategy that spans from the core data center to the intelligent edge.
The transition to PCIe 6.0 is not just about faster interfaces; it is about enabling a new class of AI applications that were previously bottlenecked by storage. With products like the GP1 Series delivering 10 million IOPS and the CM10 Series offering a near-doubling of read performance over its predecessor, KIOXIA is providing the foundational hardware needed to build the next generation of AI infrastructure. As the FMS 2025 event unfolds, the conversations surrounding booth #307 and the collaborative panel with NVIDIA will likely set the agenda for how the storage industry supports the AI revolution for the remainder of the decade. The company’s focus on performance, scalability, and efficiency is a direct acknowledgement that in the age of AI, storage is no longer a safe, silent partner but a critical, high-speed engine that must keep pace with the world’s most demanding computation.
/i.s3.glbimg.com/v1/AUTH_08fbf48bc0524877943fe86e43087e7a/internal_photos/bs/2026/q/p/xkrnfASUWa5SSkay0x6A/marvels-wolverine-20260907193401.jpg "Marvel’s Wolverine Best Techniques to Unlock First")
