At the Pure Storage Accelerate Event in Las Vegas the company announced next generation solid state storage products with capacities up to 300TB and its Enterprise Data Cloud a storage and data management system that is intended to simplify data management. These products should provide primary storage for heavy data workloads including AI training and inference.
Pure introduced new FlashArrays, a new FlashBlade and new Direct Flash Modules. The FlashArray//XL R5, shown below, has raw storage capacities up to 1.9PB and can run up to 200% more transactions per rack unit.
The FlashArray//ST is geared for high performance, providing more than 10 M input/output operations per second. The company also introduced a FlashBlade//S R2. The FlashArray’s have native S3 implementation and support block, file and object storage.
Pure has doubled the storage density of the QLC Direct Flash Modules to 300TB. One of these DFMs is shown below. These are custom solid state drive modules designed by Pure. The higher capacity in one drive results in higher energy efficiency and smaller rack space for a given storage capacity.
John Cosgrove, founder of Pure told me that their DFMs have 1/10th the error rate of other SSDs. This is because they understand and can control the sources of errors better. For instance, half of SSD failures are due to communication bugs with PCIe, but they implement solutions that avoid these communication failures.
The company also understands which NAND flash die in the DFMs are more susceptible to failure and takes steps to ensure that any failures don’t impact overall performance. As a consequence, even with such large NAND flash capacity the DFMs avoid major drive rebuilds that may be required for other SSDs. This provides reliable data access.
Pure’s Enterprise Data Cloud provides unified storage access to data on-premise, in the cloud or hosted as shown below.
The Data Cloud is built up from layers of hardware and software. The evergreen architecture includes Pure Storage Flash Systems, their Blades and Arrays and SLAs that the company says provide financial and operational flexibility by offering a hardware purchase with a consumption-based subscription with hardware replaced over time.
The virtualized cloud of storage can manage 100’s of storage systems like they are one system. The control plane offers intelligent automation and orchestration of workflows, real-time observability as well as predictive analytics and insights. The later includes anomaly detection and and cyber assessments. A variety of service can be run on the Data Cloud.
Pure has partnered with Rubrik to enhance cloud security. Pure Fusion provides work load automation. CrowdStrike provides LogScale integration to enable immutable resilient storage. PureProtect VMware to VMware on-demand recovery is supported as is AI Copilot for easy interaction of the Data Cloud.
Patrick Smith, Field CTO for EMEA led a customer panel where they discussed how with foundational models available through intensive AI training these models can be fine-tuned and customized for particular applications that deliver an acceptable degree of accuracy for a particular application. Such fine tuning of existing models rather than creating new models requires much less effort and energy to accomplish. DeepSeek’s introduction in January 2025 was an example of such an approach. Nvidia and others have referred to the increasing use of existing foundational models for AI inference solutions as the Post Training World.
This post training world requires less intensive computing but it is heavily dependent upon real time data management and solutions. This is an area where companies like Pure Storage hope to provide primary storage to support existing model fine tuning and inference to solve real world solutions in a variety of industries.
Pure Storage introduced new hardware including 300TB DFM, FlashArrays and a new FlashBlade along with its Enterprise Data Cloud for managing and protecting on-premise, cloud and hosted data to enable a post-AI training world.
