The most powerful data storage platform ever, built for AI.

A monumental shift in performance, scalability, simplicity, and future-ready adaptability. FlashBlade//EXA™ is the data storage platform to power high-performance computing (HPC) and AI’s next frontier.

Read the Announcement

The new benchmark for AI and HPC storage. Nothing else comes close.

10+ TB/s read performance.*

No more starving, idle GPUs. Write speeds scale up to 50% of reads performance.*

^*In a single namespace. Based on Pure Storage performance testing with controlled hardware environment.

3.4TB/s per rack.

Highest performance density, ever. Optimize power and cooling costs for energy-hungry GPUs.

>20X more files per namespace.

Handles the most diverse, multi-modal AI workloads with ease.

Power exabyte-scale workloads.

Supports powerful AI factories with unparalleled efficiency.

Support 1000s to 10s of thousands of GPUs.

Provides seamless scalability for AI, at any level.

50% faster deployment.

Deploy in half the time, with minimum complexity.

AI demands instant, frictionless, unlimited access to data. FlashBlade//EXA delivers.

Read the Solution Brief

Outdated storage is stifling AI innovation. Set it free.

Traditional high-performance storage was designed for predictable HPC workloads, requiring manual optimization on an ongoing basis, for performance scaling on parallel file systems. But AI has changed the game with modern, multimodal AI workloads—processing text, images, video, and more—concurrently across tens of thousands of high-powered GPUs.

Metadata overload hinders performance

Existing systems can’t keep up with metadata bottlenecks, causing inefficiencies and costly delays.

AI pipelines stall

Traditional storage lacks parallelism, requiring ongoing optimizations to feed massive GPU clusters, at full speed.

Scaling is a nightmare

AI and HPC infrastructure needs are dynamic, requiring seamless growth of data and metadata, but outdated storage systems are inherently complex and inflexible.

It’s time for a data storage platform built to accelerate AI and HPC innovation.

Read Our CEO's Perspective

A pioneer in data storage platforms for AI.

Innovating in AI since 2018.

Pure Storage FlashBlade//S™ is the trusted foundation for enterprise AI. With NVIDIA GPUDirect® Storage support, it eliminates bottlenecks, delivering direct-to-GPU data access at unmatched speeds. Backed by NVIDIA certifications—including NVIDIA DGX BasePOD™, NVIDIA DGX SuperPOD™, and more—FlashBlade//S is proven to accelerate AI at scale. Building on the success of FlashBlade//S, we’re pushing the boundaries further by optimizing metadata performance and scaling massively to meet the demands of large scale AI and HPC.

NVIDIA DGX SuperPOD™ with Pure Storage FlashBlade//S

NVIDIA DGX SuperPOD™ with Pure Storage FlashBlade//S

From multimodal data management to high-concurrency workloads, FlashBlade//EXA powers the future of AI-driven innovation.

The cutting-edge architecture built for extreme AI-scale.

FlashBlade//EXA architecture features two independently scalable components—the metadata core and data nodes—that seamlessly integrate with existing customer network infrastructure. This disaggregation ensures that the metadata core can deliver scalable, low-latency metadata access, and data nodes can enable extreme throughput without compromises.

FlashBlade//EXA Architecture

FlashBlade//EXA architecture features two independently scalable components - the Metadata Core and Data Nodes - that seamlessly integrate with existing customer network infrastructure. This disaggregation ensures that the metadata core can deliver scalable, low-latency metadata access and data nodes can enable extreme throughput without compromises.

FlashBlade//EXA Architecture

FlashBlade//EXA architecture features two independently scalable components - the Metadata Core and Data Nodes - that seamlessly integrate with existing customer network infrastructure. This disaggregation ensures that the metadata core can deliver scalable, low-latency metadata access and data nodes can enable extreme throughput without compromises.

Diagram of the innovative FlashBlade//EXA architecture

Actions

min. read

Get all the power—without limits.

FlashBlade//EXA builds on proven innovation, tackling the challenges of extreme scale, end-to-end AI workflows— delivering unmatched, multi-dimensional performance, scalability, metadata management, and simplicity.

FlashBlade//EXA bezel with golden splash effect

Maximize AI and HPC performance.

Eliminate GPU idle time and accelerate data throughput with massive, low-latency metadata operations and cost-efficient data scaling. Seamlessly expand storage to achieve the highest levels of aggregate read performance in a single namespace. Train, tune, and infer the most powerful AI models without delays—ensuring workflows stay efficient, and results arrive faster.

Simplify scaling and management.

Scale data and metadata effortlessly, reducing complexity and manual tuning—even at exabyte scale. With a disaggregated data and metadata architecture without limitations, FlashBlade//EXA keeps storage responsive to the ever-growing demands of AI.

Future-proof your infrastructure.

Stay ahead of the AI revolution with a highly configurable architecture that evolves with next-generation AI and HPC workloads. Seamlessly integrate with leading AI ecosystems like NVIDIA while supporting diverse, multimodal data sets and AI models in the most demanding environments —including high-frequency trading, drug discovery, and advanced driver assistance systems (ADAS).

Dive deeper into the innovative architecture.

AI and HPC shouldn’t be held back by storage. FlashBlade//EXA is built from the ground up to remove the limitations of conventional architectures.

Read the White Paper

FlashBlade//EXA Metadata Core Specifications

Scalability	1-10 Chassis, 10 Blades per Chassis
Capacity	1-4 DFMs per Blade, 37.5 TB DFM
Connectivity with 2 XFMs	16 x 400 GbE Uplinks
Physical	Metadata Chassis:  Dimensions (per chassis): 5U Power: 2600 W (nominal at full configuration) Pair of XFMs  Dimensions (per XFM): 1U Power: 310 W (nominal at full configuration)

Slide

FlashBlade//EXA Data Nodes Specifications

Scalability	Unlimited Scalability
Minimum CPU, DRAM Requirements	32 Cores, 192 GB DRAM
Capacity per Node	Number of NVMe Drives: 12-16 PCIe Gen4+ Drive Capacity: 3.8-61.44 TB (PCIe Gen5 drives recommended for best performance)
Connectivity	For Best Performance 2 x 400 Gb Ethernet NICs
Physical	Minimum Dimensions: 1U Drive Form Factor: Determined by Data Node Power: Determined by Data Node

Slide

Leadership perspectives. The bold vision.

“AI has disrupted storage, exposing legacy systems’ inability to handle massive parallelism. Modern storage must continuously feed GPUs and support demanding AI workloads—but high performance often brings complexity. With FlashBlade//EXA, Pure Storage applies a decade of expertise to optimize metadata while delivering the overall performance and simplifying management.”

Matt Kimball

VP & Principal Analyst, Moor Insights & Strategy

“At Penguin Solutions, we are experts at helping our customers deploy AI infrastructure at scale. We know the complexity involved, and that legacy storage systems can undermine high-performing AI infrastructure. FlashBlade//EXA is purpose-built for the demands of modern AI and HPC workloads. It provides the bandwidth, manageability at scale, and configurable, disaggregated architecture needed to power today’s large data centers. We see it as an ideal solution for intensive AI environments and HPC applications.”

Pete Manca

President, Penguin Solutions

Elegant gold splash effect

AI demands are limitless. FlashBlade//EXA removes limits.

You’ve seen the numbers. You’ve read the breakthroughs. Now it’s time to unleash your AI potential with the data storage platform that enables what comes next.

Book a Meeting Chat with an Expert

Your Browser Is No Longer Supported!

Older browsers often represent security risks. In order to deliver the best possible experience when using our site, please update to any of these latest browsers.

Metadata

FlashBlade//EXA metadata nodes provide highly available, scalable, low-latency metadata access. They eliminate bottlenecks and optimize performance with the Purity//FB metadata core, proven with thousands of deployments over eight years, built on a massively distributed transactional database and key value store technology.

Data

FlashBlade//EXA data nodes are off-the-shelf (OTS) servers optimized for seamless integration with FlashBlade//EXA metadata nodes. With compute-to-data node integration using RDMA, they enable non-blocking, high-speed data retrieval, delivering multi-terabyte-per-second performance.

Network

FlashBlade//EXA seamlessly integrates with existing customer network infrastructure, allowing for use of hardware compatible with NVIDIA Spectrum™-X networking. It utilizes industry-standard Layer3/BGP routing to manage traffic between metadata, data, and compute clusters.