Strategic PartnersIssue 017 records
Everpure
First Everpure stream edition. The main development is Everpure's September 30, 2026 production-AI data management release (MCP access to Data Intelligence catalogs, metadata-only file exposure views, PureKVA on FlashBlade, always-on DeepReduce), now covered by independent trade press that scrutinises it: StorageReview flags that the up-to-20x time-to-first-token claim has no published test configuration, Blocks & Files notes tiering signals are passive, and HPCwire reports DeepReduce works only on Everpure hardware and some automation is beta. MLPerf Storage v3.0 provides the only audited evidence (FlashBlade//EXA top checkpoint and KV-cache results at large vendor-submitted scale). Everpure's October 20 AI factory blog discloses silent build failures useful to research computing teams. A July MeriTalk SLG interview frames AI-driven flash price pressure on SLED budgets. No new named SLED AI deployment with measured results was found; SLED applicability is interpretive.
- Evidence records
- 7
- Cross-source patterns
- 2
- Evidence classes
- 4 independent reporting3 vendor claim
- Outcomes
- 4 emerging2 mixed1 cautionary
- Source freshness
- 5 recent1 new this fortnight1 older, newly relevant
- Research completed
- 2026-10-30
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
Everpure's AI performance headlines need disclosed test conditions; audited results are narrower than the claims
The vendor's September 30 release claims up to 20x faster time to first token; StorageReview reports no published test configuration for that figure. The only audited evidence located, MLPerf Storage v3.0, shows strong checkpoint and KV-cache throughput for FlashBlade//EXA but at a large vendor-submitted configuration that StorageReview did not verify, with key competitors absent.
Operating questionWhat disclosed configuration, workload and scale will we require before an Everpure AI performance claim enters an SLED evaluation?
Supporting evidenceEverpureStorageReview (Harold Fritts)StorageReview (reporting on MLCommons MLPerf Storage v3.0)
Metadata-first classification is being positioned as the gate before agents touch data
Everpure's release, Blocks & Files and HPCwire all describe Data Intelligence classifying sensitivity and exposure (File Edition metadata-only, MCP queries) before data is opened to agents. Independent reporting adds limits: passive tiering, beta automation and DeepReduce restricted to Everpure hardware. No SLED deployment evidence yet.
Operating questionWhich shares would an agency need classified and remediated, and who approves MCP access, before any copilot or agent is connected?
Supporting evidenceEverpureBlocks & Files (Chris Mellor)HPCwire/AIwire (Alex Woodie)
Evidence ledger
Full record · every source keeps its link and limitations
Everpure Announces New Data Management Capabilities for Production AI at Scale
Everpure's primary announcement of five capabilities said to be available in October 2026: native MCP integration for Everpure Data Intelligence (agents query live data catalogs and see sensitivity class), turn-key deployment through Pure1, metadata-only 'Privacy-First File Intelligence' showing who can access each file share and how stale it is, PureKVA on FlashBlade claiming up to 20x faster time to first token, always-on DeepReduce compression, and an open-weight-model token optimization reference architecture. The release carries a disclaimer that performance metrics are informational and roadmap timing is at Everpure's discretion.
Why it matters, evidence and limitations
- Why it matters
- Limited but real: state agencies, universities and districts opening file shares to copilots or agents face the same pre-condition the release targets, knowing which shares hold sensitive data (student, health, tax, CJIS-adjacent records) and who can reach them. No SLED customer, deployment or price is cited.
- Evidence and measured results
- Vendor press release only. No customer deployments, test configurations, sample sizes or baselines are published for the 20x TTFT or compression claims; the release itself states results may differ materially by environment and dataset.
- Limitations and uncertainty
- Forward-looking statement; availability stated as October 2026 but not independently confirmed in this run. Performance figures are vendor-reported without methodology. Independent reporting (Blocks & Files, StorageReview, HPCwire) in this edition adds detail and scrutiny.
Everpure bulks up AI offer
Independent trade reporting adds detail and limits to the September 30 announcement: Data Intelligence (from the 1touch acquisition) scans CIFS/SMB, NFS, OneDrive/SharePoint and Google Workspace sources, identifies PII and PHI and tracks lineage; Everpure cites a median 2:1 DeepReduce reduction for FlashBlade//E and FlashBlade//S200R2 on top of existing compression; Pure1 Configuration Advisor (beta) recommends but changes nothing on its own. The editor notes tiering signals are passive with no policy-driven data placement, and that Everpure's sovereignty survey (2,100 leaders, eight countries) is vendor-commissioned.
Why it matters, evidence and limitations
- Why it matters
- The connector list (SharePoint, OneDrive, Google Workspace, SMB/NFS) matches where most state, local and campus unstructured data actually sits, which makes the classification claim directly testable in SLED estates. Sovereignty framing is enterprise/international and needs translation to U.S. state data-residency rules.
- Evidence and measured results
- Journalist summary of vendor briefing and blogs with editorial commentary; no independent testing. Reduction ratio is Everpure's figure for specific FlashBlade models.
- Limitations and uncertainty
- Reporting relies on vendor materials; the 'policy-drive' typo is in the original. Speculation about future acquisitions is the editor's opinion and is not used here.
Everpure Adds PureKVA to FlashBlade, Claims Up to 20x Faster Time to First Token, and Makes DeepReduce Always-On
StorageReview explicitly flags that Everpure's up-to-20x time-to-first-token figure for PureKVA has no published test configuration and should be treated as a ceiling until the setup is shared. It also notes DeepReduce itself dates from February 2026 (median about 2:1 on already-compressed data), that only the always-on mode is new, and that Everpure did not name the models or GPUs in its open-weight token optimization reference design.
Why it matters, evidence and limitations
- Why it matters
- Universities and state shared-services centres evaluating on-premises inference for chat assistants or RAG should not build business cases on headline multipliers; this source gives procurement officers a concrete reason to require disclosed test conditions.
- Evidence and measured results
- Independent reporter's analysis of vendor announcement; StorageReview did not test PureKVA. The caution is about disclosure, not a measured contradiction.
- Limitations and uncertainty
- Short news piece; does not evaluate the claims empirically. Absence of published configuration is as of September 30, 2026.
Everpure Enhances Data Intelligence to Bolster Customer AI Projects
AIwire reports, with Everpure executive interviews, that Data Intelligence now comes in a File Edition (metadata-only sensitive/not-sensitive classification) and a Context Edition (full classification, contextualization and knowledge graph); that DeepReduce works only on Everpure hardware, not third-party storage; that Evergreen//One SLAs double minimum guaranteed performance across three tiers; that Red Hat AI Factory reference architectures are coming (one at launch, two later); and that the Data Intelligence automation workflow is beta, with more planned for the second half of the year. The writer cautions that perfect data management is practically impossible and past big-data projects show the difficulty.
Why it matters, evidence and limitations
- Why it matters
- The File versus Context Edition split matters for budget-constrained agencies: metadata-only classification may be enough to triage redundant, obsolete and trivial data before AI use. Evergreen//One SLA changes are relevant to SLED buyers already using storage-as-a-service.
- Evidence and measured results
- Trade reporting built on executive interviews; no customer outcomes or tests. Some timing statements (beta, later availability) are vendor statements.
- Limitations and uncertainty
- Article dated October 7 describes announcements Everpure published September 30; event and publication dates differ. Quotes are vendor executives.
What Production AI Factories Need from the Data Plane
Everpure describes its Data Architecture for Red Hat AI Factory with NVIDIA: FlashBlade//S serving S3 and NFS, Portworx FlashBlade Direct Access for shared read-write-many volumes, Kueue admission, and a capstone fine-tune of Qwen3-32B with DeepSpeed ZeRO-3. It discloses two build failures: a checkpoint silently incomplete because rank 0 uploaded only local shards, and jobs that queued silently when admission quotas covered GPUs but not CPU and memory. It states FlashBlade//S earned NVIDIA-Certified Storage for Vera Rubin and that the inference reference architecture is still being completed.
Why it matters, evidence and limitations
- Why it matters
- Directly useful to university research computing and state AI centres building shared GPU clusters on OpenShift/Kubernetes: the silent-failure lessons are platform-agnostic operational checks. Governance points on lineage for model promotion map to public-sector accountability expectations.
- Evidence and measured results
- Vendor engineering blog describing internal builds; no customer deployment, sample or measured production outcome. Self-reported failures add credibility but are not independently verified. The blog calls MLPerf results 'independent evidence'; they are audited vendor submissions.
- Limitations and uncertainty
- Marketing context (GTC Berlin demos); inference and hybrid reference architectures are not complete. Not specific to government or education.
MLPerf Storage v3.0: 877 GiB/s Checkpoints, a Cloud First, and a Leaderboard Turned Over
In its first MLPerf Storage appearance, Everpure's FlashBlade//EXA posted the round's top checkpoint result (877.5 GiB/s write, 833.0 GiB/s read at 30 data nodes, scaling near-linearly from 327.6 GiB/s at 10 nodes) and top KV-cache read (1,623.4 GiB/s on 51 hosts), plus 85,736 tokens/sec (Llama 3.1 8B) and 33,403 tokens/sec (70B) storage-only. StorageReview stresses these are audited vendor submissions it did not verify, that v3.0 results are not comparable to v2.0, and that VAST Data, WEKA and DDN sat the round out, so the audited and deployment records point to different vendor lists.
Why it matters, evidence and limitations
- Why it matters
- For research universities and state HPC/AI centres, audited checkpoint and KV-cache results are better evaluation inputs than vendor multipliers, but the tested //EXA configuration (30 blades plus 30 data nodes, about 866 TB) is far larger than most SLED deployments.
- Evidence and measured results
- MLCommons-audited benchmark results relayed by an independent outlet; vendor-submitted configurations; no SLED deployment. Results show storage throughput under synthetic workloads, not end-to-end application outcomes.
- Limitations and uncertainty
- Published September 1, 2026, before this stream's first run; included because it is the only audited evidence located that tests Everpure's AI storage claims and is directly relevant to the September 30 PureKVA claims. Competitor storage vendors are referenced only for benchmark context.
How State and Local Leaders Can Plan Through AI-Driven Infrastructure Pressure
Everpure's SLED vice president argues AI-driven demand for memory and flash is a multi-year structural pressure on state, local and education buyers, who have appropriated budgets and slower procurement; cites Gartner (second-hand) that some prices rose as much as 300 percent; and recommends right-sizing, avoiding overprovisioning and consumption models. Examples: Arkansas Department of Transportation storage-as-a-service avoiding a second data center (more than $9 million saved, more than 40 hours a day of maintenance saved) and Mississippi Department of Revenue scaling backup capacity linearly with data growth.
Why it matters, evidence and limitations
- Why it matters
- Directly SLED: describes how AI-driven component shortages affect government and school budgets and procurement cycles, with two named state agency examples.
- Evidence and measured results
- Vendor executive interview in trade media; savings figures and the Gartner price figure are relayed by the vendor without methodology or primary documents. Examples are not independently verified here.
- Limitations and uncertainty
- Published July 24, 2026, before this stream's first run; included as the only SLED-specific Everpure AI-related source located, establishing a baseline for supply-pressure claims. Vendor-positioned; '300 percent' is attributed to Gartner without a link and the quoted text contains a grammatical error in the original.
How to read this edition
Source findings, measured results and limitations come from the cited publications. Patterns, operating questions, role takeaways and implementation considerations are Lighthouse Advisory interpretation, stated as questions to validate locally rather than guaranteed outcomes. Vendor and operator claims are labeled as claims. Full research method.
- Independent reporting
- Independent reporting with attributable sources but without a formal evaluation design.
- Vendor claim
- A supplier-provided assertion that has not been upgraded to independent evidence.