Everpure claims its FlashBlade arrays can now deliver up to 20x faster Time to First Token (TTFT) for AI inference workloads, using a new Pure KVA (Key-Value Accelerator) feature that pre-stages AI context directly into GPU memory. The company also promises a median 2:1 additional data reduction from always-on DeepReduce compression on FlashBlade//E and FlashBlade//S200R2 systems. Both capabilities ship in October alongside a broader data intelligence portfolio that moves the storage vendor squarely into general-purpose data management.

The announcements, which expand on one made in June, position FlashBlade as both an AI hardware and software storage engine and the anchor of a governance layer that reaches beyond Everpure's own arrays into public clouds, SaaS applications and mainframes.

Prakash Darji, General Manager, Data & Digital Experience at Everpure, framed the problem in blunt terms. "Enterprise AI is hitting a wall not because the models are lacking, but because data is not ready for real-time, autonomous agents. We are eliminating that friction," he said. "By making enterprise data continuously governed, automated, and instantly accessible, we're giving organizations the foundation to move AI out of the lab and into production with the necessary confidence."

KV cache and compression

The KVA feature addresses one of the persistent bottlenecks in inference serving: getting context into GPU memory fast enough to keep accelerators busy. By pre-staging AI context, FlashBlade supports enterprise multi-tenancy with zero dataset relocation, eliminating GPU idle time, increasing token throughput and cutting response lag for real-time applications, according to the company.

DeepReduce takes a different angle on the same economics. It scans storage blocks continuously across FlashBlade systems to find sub-block data similarities that traditional deduplication misses, even on content that is already compressed. Everpure says customers should achieve a median 2:1 global data reduction on top of existing compression, with usable capacity expanding automatically without impacting write performance or requiring manual scheduling. The claimed payoff is a smaller hardware footprint and lower cross-cloud expenses.

The company also published a reference architecture built on open weight AI models, aiming to give enterprises more control over their data and more predictable AI costs by cutting API token usage billed by external providers.

Data Intelligence across the estate

The bigger strategic move is Everpure Data Intelligence, built on the formerly acquired 1touch technology. It discovers, classifies and contextualizes enterprise information stored not only on Everpure arrays but on other on-premises storage, public clouds, SaaS applications and third-party systems including mainframes. Connectivity runs through CIFS/SMB, NFS, OneDrive/SharePoint and Google Workspace. New capabilities give autonomous agents and administrators direct, secure access to live enterprise context without custom API work.

The software reports freshness, file age and concentration signals at the share and folder level. Infrastructure owners receive tiering and reclaim signals from the same scan pass that shows security teams where exposure sits. The tiering signals remain passive for now — there is no optimized, policy-driven data placement, though Everpure suggests it may come.

Classification identifies private and sensitive information such as PII and PHI, tracks lineage, and applies AI and ML to move beyond compliance tags into broad classification. Once classified, data can be prepared for AI by vectorization using Data Stream technology.

The most ambitious piece is a multi-dimensional semantic knowledge graph that maps raw data to its business meaning. "So an AI agent doesn't just locate a customer record; it understands that record's relationship to purchase orders, geography, and sales channels," Everpure says. "Feeding agents only relevant, contextualized data sharpens accuracy while shrinking context windows and token costs."

Unified Data Intelligence

The UDI software splits into a Unified Data Plane — which contains the KVA and DeepReduce features — and an Intelligent Control Plane with cross-fleet configuration management, a Configuration Advisor built on Portworx Observability, and autonomous workflows.

The Configuration Advisor, currently in beta, targets a real operational sore point. "Most estates carry hundreds of inconsistent per-volume configurations, set years apart by different people," an Everpure blog explains. "Pure1 Configuration Advisor (beta) reduces them to a small governed set of policies, which removes the steady manual load of maintaining protection, performance, and QoS by hand. It recommends and changes nothing on its own, so you see the impact of a change before you approve it." Admins can codify results through Everpure Fusion or their own AI tooling, or export them as an executive summary, JSON or Markdown.

Native MCP integration implements the open Model Context Protocol so AI agents and security tools can query live data catalogs in natural language and learn each dataset's sensitivity class. A privacy-first design shows who can access each file share and how stale it is without reading file content, letting administrators fix exposure and reclaim capacity before opening shares to AI agents. Pure1 workflows for Data Intelligence, a pre-built recipe in Pure1 Workflow Orchestrator, runs on schedule to discover sources, extract sensitivity classifications through MCP, iterate hosts and attached volumes, assign data classes and hand policy to Everpure Fusion for execution. Deployment runs through the existing Pure1 console with no professional services engagement or separate management servers.

Sovereignty pressure

Everpure backs the push with commissioned research: its Global Data Sovereignty Report 2026 surveyed 2,100 enterprise leaders across eight countries and found that 88 percent of large enterprise C-suite leaders believe a data sovereignty failure could cost them their jobs. The company argues sovereignty has become a defining business risk, spanning foreign-mandated data exfiltration to potential denial-of-service incidents, and that total visibility of the data estate is the prerequisite for meeting sovereignty regulations.

Competitive fallout

The move puts Everpure on a collision course with data management specialists. Datadobi, DataDynamics, Diskover and Komprise will find it harder to penetrate Everpure customer accounts from now on. CEO Charles Giancarlo has signaled more technology tuck-in acquisitions in the mold of 1touch — which invites speculation about targets such as Diskover and its ROT (redundant, obsolete and trivial) data identification technology.

With KVA, DeepReduce and the UDI stack all landing in October, the competitive dynamics in AI-adjacent data management should sharpen quickly, as incumbents in unstructured data governance face a storage incumbent with an installed base and a growing appetite for the layer above it.