Optimizing data for AI

getty

Everpure (formerly Pure Storage) had their Accelerate event this week in Las Vegas and made several announcements targeting AI workloads. Let’s go into some of their announcements.

The company announced its Everpure Data Stream to bring advanced AI capabilities to enterprise data across the storage hierarchy and make that data available for AI workloads. The company says that Data Stream reduces raw data preparation from months to minutes, enforces stream-level access controls to keep information inside a corporate network and provides a scale-out architecture to let storage and compute scale independently to match model requirements.

Everpure Data Stream extends upon the Nvidia AI data platform reference design. It replaces manual data ingestion and manipulation with a GPU accelerated path for using that data to enable real-time AI results. The company also says that it is developing next generation AI solutions with Nvidia STX, a modular foundation for AI-native storage using Nvidia Vera and the BlueField-4 storage processor.

Everpure said that their FlashBlades deliver ultra-low latency and leverage KV Cache Accelerators to optimize memory efficient during inference, scaling as needed with various FlashBlade products to expand to support a large AI factory. Portworx provides the container platform to deploy, manage and run AI workloads.

The company announced a further AI enhancement with its Data Intelligence (formerly 1touch.io) and updates to the Enterprise Data Cloud. These various features for improving AI workloads are shown in the image below.

By taking data out of application silos and focusing on the availability of data for AI workloads that data can be made self-describing and carry its own meaning and logic where ever it goes. Governance is embedded in the data layer to do lifecycle management and enforce privacy rules without using outside software. Apps and agents read from and contribute to the data pool, but do not own that data.

Data Intelligence has three core capabilities. Universal discovery offers visibility into structured and unstructured data regardless of storage format, including major databases such as SQL Server and Oracle. Automated governance scans the entire system to identify sensitive information and track its lineage, providing a complete map of the data to ensure secure compliance. The AI-ready context maps raw data directly to its real-world business definition to create a semantics knowledge graph to enable AI agents to instantly understand, query and safely act on that information.

Evergreen//One Overdrive will be available in Q3 2006 and will provide a temporary cloud-like performance boost for on-premises storage to seamlessly absorb traffic spikes up to 25% above baseline without requiring permanent subscription upgrades. By Q4 2026 the Intelligent Control Plane will allow workload rebalancing and mobility to optimize capacity distribution and guaranteeing steady application performance. This quarter Copilot Workflow Execution will allow storage administrators to use natural language to plan, validate and enabling secure end-to-end operations.

Also, this quarter Enhanced Cyber Anomaly Detection will spot coordinated, suspicious login patterns and behavior that individual storage arrays might miss. In Q4 2006 Fusion Compliance and Agentic Triage will automatically detect hardware or software configuration drift to enforce global corporate data governance and using agentic AI to suggest root causes for remediation.

At Everpure Accelerate the company announced its Data Stream to optimize and organize data for real-time AI workloads and its Data Intelligence for optimizing and securing data for AI workloads.