Pure Storage recently announced the launch of its new FlashBlade//EXA, a high-performance storage platform designed for AI and HPC workloads. FlashBlade//EXA extends the company’s Purity operating environment and DirectFlash technology to provide extreme performance, scalability, and metadata management that addresses the increasing demands of AI-driven applications.

The platform is based on a new disaggregated architecture that allows independent scaling of data and metadata, optimizing flexibility while ensuring predictable performance for large-scale AI and HPC environments.

FlashBlade//EXA

Pure Storage’s FlashBlade//EXA supports AI workloads by integrating several key technologies:

• DirectFlash Technology: FlashBlade//EXA utilizes Pure’s DirectFlash technology to manage NAND flash memory natively, eliminating the abstraction layers associated with traditional SSDs and improving performance, reliability, and cost-efficiency.
• Purity Operating System: The system employs a key-value store architecture, enabling high-concurrency access with minimal contention, which is critical for AI-driven data processing.
• Disaggregated Storage Architecture: The platform enables independent scaling of compute and storage resources, allowing users to select third-party qualified data nodes while maintaining seamless interoperability with the FlashBlade//EXA infrastructure.
• High-Performance Metadata Management: AI workloads involve large-scale metadata operations, requiring low-latency access and high IOPS. FlashBlade//EXA supports high-speed metadata processing, preventing bottlenecks in training and inference workflows.
• Massively Parallel Processing: FlashBlade//EXA optimizes read/write performance by using parallel data access techniques, reducing bottlenecks when dealing with high concurrency workloads.
• Namespace and File System Scaling: The system supports an exabyte-scale file system with trillions of files within a single namespace, reducing complexity and improving manageability.
• End-to-End NVMe Fabric: FlashBlade//EXA employs an all-NVMe architecture to minimize latency and maximize throughput across all storage tiers.
• Intelligent Load Balancing: The platform dynamically distributes workloads across storage nodes to ensure optimal performance and resource utilization.
• Predictive Caching and Tiering: FlashBlade//EXA incorporates an intelligent caching mechanism that anticipates workload demands, ensuring rapid access to frequently used data and reducing I/O latency.
• Adaptive Data Flow Optimization: The system dynamically adjusts data flow patterns based on workload intensity, preventing congestion and optimizing data access speeds.

Performance

Pure Storage released several key performance specifications that show the new platform positioned as one of the highest-performing storage platforms in the industry:

• Read and Write Throughput: Delivers over 10 TB/s of sustained read performance, with write speeds at up to 50% of read speeds.
• Performance Density: Achieves 3.4 TB/s per rack.
• Scalability: Supports trillions of files and exabytes of capacity within a single namespace, reducing complexity for large-scale AI operations.
• Interoperability: The system integrates with industry-standard networking, including NVIDIA ConnectX NICs, Spectrum switches, and LinkX cables, optimizing AI workloads in GPU-powered environments.

Note that these are preliminary performance numbers that have not been independently validated. The product isn’t expected to ship until mid-2025.

Analysis

FlashBlade//EXA is a departure from Pure’s more traditional approach to storage, signally a strategic focus on AI and HPC workloads.

The increasing scale of foundational AI models and their computational requirements have outpaced traditional storage solutions, creating a market opportunity for platforms that can provide high-performance, scalable, and flexible storage architectures.

While some of that market has been satisfied by offerings from startups like VAST Data and WEKA, Pure is the first mainstream storage vendor to deliver a disaggregated solution that leverages its best-of-breed enterprise storage stack but adapted for the specific needs of large-scale AI and HPC.

FlashBlade//EXA is a market expander for Pure Storage, strengthening its presence in AI and HPC storage. The platform supports large-scale AI model training and inference, making it attractive for AI-native enterprises, hyperscalers, and specialized GPU cloud providers. At the same time, its disaggregated architecture allows enterprises to integrate FlashBlade//EXA with both on-premises and cloud-based storage solutions, ensuring seamless scalability across hybrid environments.

Overall, FlashBlade//EXA addresses critical storage challenges in AI and HPC, delivering high throughput, scalability, and metadata optimization. The platform’s disaggregated architecture aligns with evolving AI infrastructure trends, allowing organizations to scale storage independently from compute resources.

Pure Storage’s continued investment in AI-driven storage solutions positions it as a key player in the AI and HPC storage market. It has the potential to capture enterprise AI workloads seeking alternatives to traditional storage models.

Pure’s FlashBlade//EXA arrives just as enterprises begin to build the infrastructure required for the next phases of the AI revolution. Pure is right on time with this solution, beating its more traditional competitors to the punch. It’s the right product at the right time.

Competitive Outlook & Advice to IT Buyers

These sections are only available to NAND Research clients. Please reach out to info@nand-research.com to learn more.