Korean
At ISC 2026, DDN unveiled an expanded lineup of its AI and HPC data platforms, rolling out new hardware, software and integrated capabilities to enhance GPU utilization, accelerate AI inference and simplify large-scale AI infrastructure deployment.
The new launches cover the AI400X3M storage appliance, the official general availability of DDN’s NVIDIA Dynamo-integrated distributed KV Cache acceleration technology, as well as upgraded security, observability and infrastructure management features for its full AI data platform stack.
These optimizations resolve core bottlenecks throughout the entire AI data pipeline, including data preprocessing, model training, inference, retrieval-augmented generation (RAG), logical reasoning and agentic AI workloads. DDN focuses on boosting overall infrastructure efficiency by maximizing GPU utilization while reducing power draw and total deployment costs.
DDN CEO and Co-Founder Alex Bouzari noted that modern production AI infrastructure cannot rely solely on compute power. He emphasized that efficient data orchestration, robust security and comprehensive operational visibility have become indispensable for enterprise AI deployment. DDN’s new updates are engineered to improve GPU efficiency, cut inference costs and reduce model time-to-first-token latency.
AI400X3M: Elevating AI and HPC Performance Density
Leading DDN’s new product portfolio is the AI400X3M, a next-generation turnkey storage appliance built on the mature, high-performance EXAScaler parallel file system.
AI400X3M Full Specifications
Performance
-
Sequential Read Performance: Up to 190 GB/s
-
Sequential Write Performance: Up to 110 GB/s
-
Random Read IOPS: 8M
System Features
-
Platform: Turnkey EXAScaler shared parallel file system appliance for AI, equipped with active-active storage controllers
-
Host Ports per Appliance: 4× XDR/400GbE OSFP or 8× NDR200/200GbE QSFP112
-
Drive Support: 24× 2.5-inch dual-port hot-swappable PCIe G5 NVMe drives
-
Capacity Options: 120/250/500TB, 1PB/2PB usable storage
-
Standard Firmware Features: LUN mapping and masking, intelligent write striping, read QoS control, data integrity check and correction, multi-mode interface (CLI, GUI, Python API), outbound syslog with RELP support, email/SNMP state change alerts
Safety & Compliance
-
Safety Certifications: IEC/EN/UL/CSA 62368-1, GB4943
-
EMC Certifications: EN 55032 Class A, EN 55035, EN 61000-3-2/3-3, FCC Part 15 Class A, VCCI Class A, ICES-003 Class A, GB/T9254 Class A, BSMI Class A
-
Environmental Compliance: RoHS, REACH, WEEE, ERP Lot9X
Physical Attributes
-
Form Factor: 2U rack-mountable
-
Dimensions: 3.43-inch (87mm) height, 17.56-inch (446mm) width, 33.7-inch (856mm) depth (without bezel); 35.2-inch (895mm) overall depth
-
Rail Range: 26.75–34.0 inches (680–860mm)
-
Weight: 98 lbs (44kg) empty; 108 lbs (49kg) maximum loaded weight
Power & Cooling
-
Input Voltage: 200–240V, 50/60 Hz
-
Power Supply: 2× hot-swappable redundant PSUs with IEC60320-C20 inlet
Operating Environment
-
Temperature Range: 5°C to 30°C (41°F to 86°F)
-
Relative Humidity: 8%–80% (non-condensing)
-
Altitude: 3,117 ft (950m) @ 30°C; 10,000 ft (3,048m) @ 23°C
The new platform delivers up to 35% higher read performance than the previous generation with a maximum throughput of 190GB/s. It supports single-rack scaling up to 30PB and adopts a hybrid flash-HDD storage architecture to balance performance, capacity and overall cost. Tailored for AI training, inference, checkpointing and HPC simulation workloads, the appliance effectively eliminates storage bottlenecks that restrict GPU efficiency. General availability is scheduled for the end of Q3 2026.
Distributed KV Cache Acceleration Now Generally Available
Following its technology preview at NVIDIA GTC 2026, DDN’s distributed KV Cache acceleration architecture with native NVIDIA Dynamo integration has reached full GA status, compatible with both DDN Infinia object storage and EXAScaler parallel file system.
Instead of limiting KV cache storage to local GPU memory, this distributed architecture offloads KV cache data to DDN’s high-performance storage platform, eliminating critical memory bottlenecks during large-scale inference.
Compatible with NVIDIA Dynamo, vLLM and other mainstream inference frameworks, the solution enables shared distributed KV cache pooling across multiple inference nodes. It delivers up to 55× faster KV cache loading speed for LLM workloads, significantly improving GPU utilization, accelerating token generation and cutting overall inference deployment costs. It is highly suitable for long-context inference, RAG, reasoning models and complex agentic AI workflows with frequent large-context access.
Unified End-to-End AI Data Pipeline
DDN has launched a unified AI data infrastructure strategy, combining Infinia object storage for inference workloads and EXAScaler parallel storage for model training and checkpointing.
Infinia provides low-latency metadata processing and high concurrent performance optimized for inference scenarios, while EXAScaler delivers ultra-high throughput for intensive model training. The integrated architecture eliminates data silos and sustains high GPU utilization throughout the full AI lifecycle.
Upgraded Security and Intelligent Infrastructure Features
DDN has rolled out multiple enterprise-grade enhancements, including bare-metal multi-tenancy, KMIP-based encryption and key management, VictoriaLogs-powered operational observability, and expanded multi-tenant API support.
On the storage optimization side, new intelligent file pinning and NAND-accelerated Hot Pools features keep hot frequently accessed data on high-speed flash storage and automatically tier cold data to low-cost HDDs, effectively improving storage efficiency and reducing flash capacity investment.
Continuous Expansion of Cloud AI Deployment
DDN continues to expand its cloud AI infrastructure ecosystem, including co-developed Managed Lustre solutions with Google Cloud and a production deployment at Salesforce based on Google Cloud Managed Lustre powered by EXAScaler.
The Salesforce deployment achieves 1.5× faster model training speed, 75% lower I/O latency and 42% reduced training costs, fully validating the value of DDN’s storage optimization for large-scale AI workloads.
With this series of updates, DDN is expanding its capabilities beyond traditional parallel storage to cover AI data orchestration, inference acceleration and full-stack infrastructure management, empowering large-scale GPU cluster deployments for enterprise, cloud, research and sovereign AI scenarios.
Beijing Qianxing Jietong Technology Co., Ltd.
Sandy Yang/Global Strategy Director
WhatsApp / WeChat: +86 13426366826
Email: yangyd@qianxingdata.com
Website: www.qianxingdata.com/www.storagesserver.com
Business Focus:
ICT Product Distribution/System Integration & Services/Infrastructure Solutions
With 20+ years of IT distribution experience, we partner with leading global brands to deliver reliable products and professional services.
“Using Technology to Build an Intelligent World”Your Trusted ICT Product Service Provider!
Sandy Yang/Global Strategy Director
WhatsApp / WeChat: +86 13426366826
Email: yangyd@qianxingdata.com
Website: www.qianxingdata.com/www.storagesserver.com
Business Focus:
ICT Product Distribution/System Integration & Services/Infrastructure Solutions
With 20+ years of IT distribution experience, we partner with leading global brands to deliver reliable products and professional services.
“Using Technology to Build an Intelligent World”Your Trusted ICT Product Service Provider!