Weekly Storage & AI Data Platform Briefing

Week of May 17, 2026 — Period Covered: May 11–17, 2026

---

1. Executive Summary

This week's briefing identifies six genuinely new signals with strategic significance for NetApp's EMEA, LATAM, and Middle East operations. The dominant theme is competitive escalation: Dell Technologies World (May 18–21) drops its pre-event announcements revealing AI Factory 2.0, HPE launches a significant Alletra refresh with native file and agentic operations, Everpure completes its data intelligence acquisition, and Scality enters the AI-native storage category. Meanwhile, Iran's move to assert control over Strait of Hormuz undersea cables fundamentally escalates the Middle East connectivity crisis.

Signal 1 — Dell Technologies World: AI Factory 2.0 Launches with 256 Blackwell Ultra GPUs/Rack, PowerScale pNFS 6x Performance Gain, and KIOXIA 9.8 PB in 2U. Dell's May 18 keynote (Michael Dell + Jensen Huang) unveils AI Factory 2.0 featuring PowerEdge XE9780/XE9785 servers supporting up to 192 NVIDIA Blackwell Ultra GPUs with direct-to-chip liquid cooling, expandable to 256 GPUs per IR7000 rack — claiming 4x faster LLM training vs. XE9680. Storage announcements include PowerScale pNFS delivering 6x performance vs. NFSv3 for large-file AI workloads, and KIOXIA demonstrating 9.8 PB in a single 2U Dell PowerEdge R7725xd using 40× LC9 E3.L 245.76 TB NVMe SSDs. Dell Lightning File System (150 GB/s per rack, GA since April 2026) and Exascale Storage (6 TB/s per rack, H2 2026) complete the portfolio. [CROSS-VALIDATED] NetApp implication: This is the most significant competitive event of the quarter. Dell's "data-first AI" narrative directly targets the same positioning NetApp has claimed with AIDE. The PowerScale pNFS 6x claim against NFSv3 is carefully worded — ONTAP's NFS performance should be benchmarked specifically against PowerScale pNFS to validate our comparative position. Dell Lightning (150 GB/s/rack) competes directly with EF-Series/AIDE for tier-0 AI workloads, while the KIOXIA 9.8 PB/2U density demonstration raises the bar on capacity density. NetApp SEs need updated competitive materials by Monday.

Signal 2 — HPE Alletra X10000 Adds Native File + 23 PB Scale; B10000 Gets Agentic AI Operations (May 12). HPE announced a major GreenLake refresh: Alletra X10000 gains native file storage alongside existing object (single platform, dual protocol), scales to 16 nodes/23 PB raw capacity, adds RDMA for file, and carries a new 100% data availability guarantee. Alletra B10000 adds agentic AI support that autonomously detects, analyzes, and resolves storage issues — moving "beyond signature-based predictive analytics with semantic understanding, adaptive reasoning, and prescriptive intelligence." B10000 scales from 4 to 6 controller nodes (+50% performance) with dual-node fault tolerance. HPE Data Fabric v8.1 adds policy-based tiering with conversational AI interface. X10000 file GA in Q2 2026; 16-node scale-out and RDMA in Q3 2026. [CROSS-VALIDATED] NetApp implication: HPE's X10000 now directly competes with ONTAP's unified file+object proposition on a single platform. The 100% data availability guarantee is a bold marketing move — NetApp should evaluate whether to counter with its own availability commitment backed by ONTAP's proven HA architecture. The agentic AI operations in B10000 represents the next competitive frontier: self-managing storage. ONTAP's AIOps capabilities (via Console/BlueXP) need to be articulated as equal or superior. The combined file+object+RDMA positioning on X10000 is a direct shot at ONTAP's unified protocol story.

Signal 3 — Everpure Completes 1touch Acquisition (May 11): Data Intelligence Layer Now Integrated into Storage Platform. Everpure (NYSE: P) closed the 1touch deal, adding AI-powered data discovery, classification, and semantic enrichment directly into the Everpure Platform. Stock jumped 15% on the news. 1touch provides the ability to discover, classify, contextualize, and enrich data across all datasets and any environment — making enterprise data "inherently AI-ready at the source." Everpure reports Q1 FY2027 earnings May 27. [CROSS-VALIDATED] NetApp implication: Everpure is executing a strategic pivot from "storage vendor" to "data intelligence platform." The 1touch integration adds exactly the capabilities AIDE was designed to deliver — data discovery, classification, and AI-readiness. This is a direct competitive positioning challenge. NetApp must ensure AIDE's broad availability (targeted "early summer 2026") happens before Everpure can demonstrate integrated 1touch capabilities in production. The 15% stock jump indicates the market values this pivot — investors see data intelligence as the next storage battleground.

Signal 4 — Iran Asserts Control Over Strait of Hormuz Undersea Cables: IRGC Demands Permits and "Protection Fees" from Foreign Operators. IRGC-linked media (Tasnim) announced that Iran will take "full control" of seven critical undersea fiber-optic cables passing through the Strait of Hormuz, demanding foreign operators obtain Iranian permits and pay protection fees to maintain infrastructure. This escalates the cable crisis from passive disruption to active state intervention. Cable repair remains paralyzed, with both the Strait of Hormuz and Red Sea classified as no-go zones for repair vessels. CNN reports this represents a new geopolitical leverage tool for Iran over Gulf digital infrastructure. [CROSS-VALIDATED] NetApp implication: This escalation fundamentally changes the risk calculus for Gulf state AI investments. Stargate UAE, AWS HUMAIN (Saudi), and all Gulf data center operators now face a structural connectivity vulnerability that cannot be resolved through technology alone. For NetApp field teams: any Gulf customer conversation about AI infrastructure should now include a data sovereignty and replication architecture discussion. ONTAP SnapMirror and MetroCluster provide the technical answer for multi-site resilience across connectivity disruption scenarios. This is no longer a temporary cable outage — it's a permanent geopolitical risk that requires architectural design changes.

Signal 5 — Scality Launches ADI (Autonomous Data Infrastructure): AI-Powered, Policy-Driven Tiered Storage Platform (May 12). Scality ADI introduces an AI-powered operations engine ("Guardian") layered on proven object storage, spanning NVMe SSD, HDD, tape, and cloud within a single namespace with policy-driven lifecycle management. Agent workers handle predictive maintenance, workload alignment, and cyberthreat detection — with humans kept in the loop for decisions. Software-defined, disaggregated architecture targets multi-petabyte to exabyte scale for AI, cyber resilience, and sovereign control use cases. [SINGLE SOURCE] NetApp implication: Scality ADI is noteworthy because it positions autonomous, AI-driven data lifecycle management as a product category. While Scality is primarily an object storage player (not directly competing with ONTAP for primary workloads), the "autonomous infrastructure" narrative is one that all enterprise storage vendors will need to address. NetApp's FabricPool, storage tiering, and AIOps capabilities deliver similar outcomes but are not packaged or marketed as an "autonomous" platform. Watch for customer interest in the "self-managing storage" category.

Signal 6 — NetApp × Red Hat: OpenShift VM-Level Backup, DR, and GCNV Integration (May 12). NetApp announced new data management capabilities for Red Hat OpenShift including VM-level protection with incremental-forever backups and change block tracking, orchestrated disaster recovery for Kubernetes-based VMs (public preview), and GA of Google Cloud NetApp Volumes with Trident CSI for Red Hat OpenShift Dedicated on Google Cloud. [CROSS-VALIDATED] NetApp implication: This is a portfolio strengthening signal rather than a competitive threat. The GCNV + OpenShift Dedicated GA fills a gap in the Google Cloud containerized enterprise storage story. The DR for OpenShift VMs positions ONTAP as the data protection backbone for customers migrating from VMware to Red Hat OpenShift Virtualization — a migration path accelerating due to Broadcom's VMware pricing. Field teams should use this in every VMware-to-OpenShift conversation: "Your VMs move to OpenShift, your ONTAP-based data protection and DR move with them."

---

2. Competitive Radar

Dell Technologies — AI Factory 2.0 at Dell Technologies World (May 18–21)

What they announced: Dell's most significant annual event launches May 18 with a joint Michael Dell + Jensen Huang keynote. Pre-event announcements and analyst briefings (May 12–14) reveal:

• AI Factory 2.0: PowerEdge XE9780/XE9785 (air-cooled) and XE9780L/XE9785L (liquid-cooled) supporting up to 192 NVIDIA Blackwell Ultra GPUs, configurable to 256 GPUs per IR7000 rack. 4x faster LLM training vs. XE9680. Enclosed rear-door heat exchanger captures 100% of rack heat, saving up to 60% on cooling.
• PowerScale pNFS: 6x faster performance with large files in enterprise AI environments vs. NFSv3. Flexible File Layout enables parallel distribution across cluster nodes.
• Dell Lightning File System: 150 GB/s per rack, 2x throughput vs. competing parallel file systems, 20x vs. flash-only scale-out file. GA since April 2026.
• Dell Exascale Storage: 3-in-1 software-defined (PowerScale + ObjectScale + Lightning) with 800 GbE support, 6 TB/s per rack. H2 2026 availability.
• Data Intelligence: 12x faster vector indexing, 3x faster data processing, 19x faster time-to-token (Qwen3-32B testing). Data Orchestration Engine (from Dataloop acquisition) provides no-code/low-code AI data pipelines.
• KIOXIA Partnership: 9.8 PB in a single 2U PowerEdge R7725xd using 40× KIOXIA LC9 E3.L 245.76 TB NVMe SSDs — 8x less power than comparable capacity in conventional configurations.
• NVIDIA AI Enterprise now available directly through Dell.
• 4,000+ customers deploying AI Factory, early adopters reporting 2.6x ROI in first year.

NetApp's current answer: ONTAP unified storage (NFS/S3/iSCSI), AIDE for AI data management, EF-Series for high-performance block, NVIDIA STX/CMX partnership. NetApp does not have a "Lightning-equivalent" tier-0 parallel file system. AIDE remains in early access (broad availability "early summer 2026").

Gap or advantage:

Suggested talk track: "Dell announced impressive hardware at DTW — 256 Blackwell GPUs in a rack is extraordinary compute density. But the question isn't how fast you can train — it's whether your data infrastructure can keep up with that compute, manage the lifecycle of petabytes of AI data, and move it seamlessly between on-prem and cloud. Dell gives you fast hardware. ONTAP gives you a data fabric that spans from the GPU to the cloud, with snapshots, replication, compliance, and tiering built in. Dell Lightning is fast — but it's another silo. ONTAP is fast AND unified."

Sources: SiliconANGLE — AI Factory, SiliconANGLE — Data Intelligence, TweakTown — KIOXIA, KIOXIA 9.8 PB Press Release

---

HPE — Alletra X10000/B10000 Refresh + GreenLake Private Cloud (May 12)

What they announced: Comprehensive GreenLake refresh targeting AI data readiness:

• Alletra X10000: Native file storage added to existing object platform (single unified system). Scales to 16 compute + 16 storage nodes, 23 PB raw. 100% data availability guarantee. RDMA for file storage (Q3 2026). Positioned for AI training, inference, and KV cache workloads.
• Alletra B10000: Agentic AI support — autonomous detection, analysis, and resolution of storage issues. Scales from 4 to 6 controller nodes (50% more performance). Dual-node fault tolerance. Software release v10.6.0.
• Data Fabric v8.1: Policy-based data placement and movement (tiering), conversational AI assistant, automated reporting.
• Zerto: AI-powered data protection, integrated with Microsoft Defender for real-time threat visibility.

NetApp's current answer: ONTAP unified storage (file+object+block), Console/BlueXP for management with AIOps, SnapMirror for DR, FPolicy + SnapLock for compliance. ONTAP has delivered unified protocols for years; HPE is catching up on the X10000.

Gap or advantage: HPE's "agentic AI operations" on B10000 is a marketing-first move — the autonomous issue resolution is limited to storage-layer problems. ONTAP's broader ecosystem intelligence (via Console, ActiveIQ/Digital Advisor heritage) covers similar ground without the "agentic AI" branding. The 100% data availability guarantee is aggressive and will resonate with risk-averse buyers. NetApp should evaluate a comparable commitment backed by ONTAP MetroCluster and HA pair architecture.

Suggested talk track: "HPE just added file to X10000 — congratulations, they now have what ONTAP has delivered for over a decade. The difference: ONTAP's unified protocol stack is battle-tested at scale across thousands of enterprises, with 30+ years of data management innovation built in. Their 'agentic AI' for storage operations is interesting, but ask what happens when the issue crosses the storage boundary — does it understand your application, your replication topology, your cloud tiering policy? ONTAP's intelligence extends across the entire data lifecycle, not just one storage array."

Sources: HPE Press Release May 12, Blocks & Files — HPE Updates, HPCwire — HPE AI Inference

---

Everpure — 1touch Acquisition Complete (May 11); Pivot to Data Intelligence Platform

What they announced: Everpure closed the 1touch acquisition, integrating data discovery, classification, and semantic enrichment into the Everpure Platform. Stock surged 15%. The combined platform discovers, classifies, contextualizes, and enriches data across all datasets in any environment — from SaaS to edge — making enterprise data "AI-ready at the source." Q1 FY2027 earnings scheduled May 27.

NetApp's current answer: AIDE (AI Data Engine) provides data discovery, classification, and preparation for AI workloads. AIDE is in early access with lighthouse customers; broad availability targeted "early summer 2026."

Gap or advantage: Everpure's 1touch integration is closed and being marketed. AIDE is in early access. The capability overlap is significant — both aim to make unstructured data AI-ready through automated discovery and classification. The gap is timing: Everpure can now demonstrate an integrated platform; NetApp cannot yet do so at scale. However, AIDE's advantage is its integration with ONTAP's full data management stack (snapshots, replication, tiering, compliance), while 1touch on Everpure is limited to the Everpure storage footprint.

Suggested talk track: "Everpure acquired a data classification startup and called it a platform. That's one piece of the puzzle. AIDE doesn't just classify your data — it integrates classification with ONTAP's data lifecycle: classify, protect, tier, replicate, and comply — all in one platform. Classification without action is just a catalog. AIDE makes classification actionable across the entire data estate, on-prem and in cloud."

Sources: Everpure Press Release, Blocks & Files — Everpure 1touch, CoinCentral — Stock Jump

---

Scality — ADI (Autonomous Data Infrastructure) Launch (May 12)

What they announced: Scality ADI introduces AI-powered "Guardian" operations engine layered on proven Scality object storage. Platform spans NVMe SSD (TLC/QLC), HDD, tape, and cloud within a single namespace. Policy-driven lifecycle management with AI agent workers handling tiering, predictive maintenance, workload alignment, and cyberthreat detection. Positioned for sovereign control, cyber resilience, and AI data at multi-petabyte to exabyte scale.

NetApp's current answer: ONTAP FabricPool provides automated tiering; ONTAP AIOps via Console provides predictive intelligence; SnapLock/FPolicy deliver compliance. ONTAP does not position itself as "autonomous infrastructure" in marketing, though the underlying capabilities are comparable.

Gap or advantage: Scality competes primarily in the object storage and sovereign infrastructure space — not directly against ONTAP for primary enterprise NAS or SAN workloads. However, the "autonomous data infrastructure" category positioning is worth monitoring. If customers begin asking for "self-managing storage," NetApp needs a clear narrative.

Suggested talk track: Use only if Scality appears in a competitive situation: "Scality ADI is an interesting approach to object storage lifecycle management. For customers who need unified file, block, AND object with autonomous lifecycle intelligence, ONTAP delivers all three protocols with integrated tiering, snapshots, and AI-driven operations through Console — without requiring you to adopt a separate platform for each protocol."

Sources: Scality ADI Press Release, Blocks & Files — Scality ADI

---

3. NVIDIA AI Infrastructure

Dell Technologies World: AI Factory 2.0 with Blackwell Ultra

The May 18 keynote marks the production-scale availability announcement for NVIDIA Blackwell Ultra in Dell infrastructure. Key specs relevant to storage planning:

• 256 Blackwell Ultra GPUs per IR7000 rack — each GPU generating significant data I/O for training and inference
• PowerEdge XE9712 with GB300 NVL72 — 50x more AI reasoning inference output, 5x throughput improvement vs. Hopper
• Dell Pro Max GB300 Desktop — 20 petaFLOPS FP4, 748 GB coherent memory, trillion-parameter models at the desk
• NVIDIA AI Enterprise available directly through Dell — reducing friction for enterprise AI deployment
• Vera Rubin/Vera CPU support planned via new PowerEdge XE server (Dell confirmed)

Storage demand implications: A fully loaded IR7000 rack with 256 Blackwell Ultra GPUs will generate unprecedented storage I/O demand. Dell's answer is Lightning (tier-0, 150 GB/s/rack) + PowerScale (tier-1, pNFS) + ObjectScale (capacity tier). NVIDIA's DSX reference architecture requires certified storage at each tier. NetApp must ensure AIDE + EF-Series are validated against these compute density specifications.

NVIDIA Dynamo 1.0: Production KV Cache Management Reaches Storage Layer

NVIDIA Dynamo 1.0 (production since GTC March 2026) is gaining traction as the inference operating system for AI factories. Critical storage-relevant development this week: VAST Data published detailed integration documentation showing 35 GB/s throughput to a single H100 GPU via GPUDirect Storage, with 10x prefill improvement for large-context scenarios using NIXL + LMCache.

Dynamo's KV Block Manager (KVBM) now supports S3 and Azure Blob APIs — meaning any S3-compatible storage (including ONTAP S3) can serve as a KV cache offload tier. KVBM uses tiered caching: GPU memory → CPU memory → local SSD → remote storage (S3/blob). Events are emitted when KV blocks move between tiers, enabling cluster-wide routing optimization.

NetApp relevance: ONTAP's S3 compatibility positions it as a viable remote KV cache tier for Dynamo KVBM. However, VAST Data has published validated integration numbers (35 GB/s/GPU) while NetApp has not. The competitive gap: VAST is the reference storage partner for Dynamo KV cache offload, and they're publishing proof. NetApp engineering should validate ONTAP S3 as a Dynamo KVBM backend and publish comparable benchmarks.

[UPDATE] Vera Rubin, BlueField-4

No new timeline changes. Vera Rubin NVL72 production remains H2 2026. Dell confirmed Vera CPU support in new PowerEdge XE servers (revealed at DTW pre-briefs).

Sources: NVIDIA Dynamo Technical Blog, VAST Data + Dynamo Integration, Dell AI Factory 2.0 Details

---

4. NetApp Portfolio & Ecosystem

NetApp × Red Hat OpenShift Collaboration (May 12)

NetApp announced expanded data management for Red Hat OpenShift, directly addressing VMware-to-OpenShift migration momentum:

• Backup and Recovery for OpenShift: Now supports VM-level protection with automated workflows using incremental-forever backups and change block tracking. This is critical for customers running Red Hat OpenShift Virtualization as a VMware replacement.
• Disaster Recovery for OpenShift: Public preview — extends beyond backup to orchestrated DR for Kubernetes-based VMs. First DR service for containerized VM workloads backed by ONTAP.
• GCNV + Trident CSI for OpenShift Dedicated on Google Cloud: Now GA. Enterprises running Red Hat OpenShift Dedicated can consume Google Cloud NetApp Volumes with full Trident integration.

Strategic significance: This announcement positions NetApp as the de facto data services platform for VMware-to-OpenShift migrations. With Broadcom's VMware pricing driving migration conversations in EMEA, the timing is strategic. The GCNV + OpenShift Dedicated integration fills a gap in the Google Cloud containerized storage story and strengthens the three-cloud first-party services narrative (ANF + GCNV + FSx for ONTAP).

Network Storage Advisors: 2026 Strategic Landscape for Enterprise AI Storage

Published May 7–11 (gaining traction this week), this 73-page report benchmarks AI storage systems from DDN, Dell, Everpure, Hitachi Vantara, HPE, IBM, NetApp, VAST Data, and WEKA. It includes dashboards comparing performance, capacity, power, and space across NVIDIA-certified DGX BasePOD and SuperPOD configurations. The report categorizes vendor strategies into three types (portfolio, storage, workload) and three solution classes (configured, optimized, specialized).

NetApp relevance: This is the first major analyst report to benchmark all nine enterprise AI storage vendors head-to-head against NVIDIA reference architectures. Field teams should obtain a copy ($5,000) to understand NetApp's positioning relative to competitors across DGX configurations. If NetApp is positioned favorably, it's a third-party proof point for customer conversations.

[UPDATE] AIDE, Console, Q4 FY2026 Earnings

AIDE remains in early access; broad availability "early summer 2026." NetApp Q4 FY2026 earnings announcement scheduled May 28 (after market close). Consensus: revenue ~$1.87B (8% YoY growth), EPS $2.21–$2.31 (guidance), analysts expecting $1.88. Key watch items: AIDE customer traction, NAND cost impact on margins, cloud ARR momentum.

Sources: NetApp × Red Hat Press Release, Blocks & Files — Red Hat OpenShift, Network Storage Advisors Report

---

5. Enterprise Storage Vendor Landscape

KIOXIA — 9.8 PB in 2U Server; 245.76 TB NVMe SSDs at Dell Technologies World

KIOXIA announced a breakthrough in storage density at Dell Technologies World: 9.8 PB of flash in a single 2U Dell PowerEdge R7725xd server using 40× KIOXIA LC9 Series E3.L form factor 245.76 TB NVMe SSDs with PCIe 5.0. Key comparison: achieving 9.8 PB conventionally would require seven additional servers with 280 additional drives, consuming 8x the power. KIOXIA also showcased FIPS 140-3 validated drives for government and regulated workloads. [CROSS-VALIDATED]

NetApp relevance: The 245.76 TB per SSD represents the leading edge of enterprise flash density. When these drives become broadly available for storage system integration, they will dramatically change capacity-per-RU economics. NetApp engineering should be tracking KIOXIA LC9 Series qualification for AFF/ASA platforms. The density claim (9.8 PB in 2U) is server-based, not storage-system-based — but the underlying drive technology will eventually appear in storage arrays from all vendors.

Scality — ADI (Autonomous Data Infrastructure) Platform

Detailed in Section 2. Scality's entry into AI-native, policy-driven tiered storage with autonomous operations positions them in the sovereign cloud and exabyte-scale object market. Key differentiator: single namespace spanning NVMe, HDD, tape, and cloud with AI-driven lifecycle management.

[UPDATE] VAST Data, DDN, IBM, Hitachi Vantara, MinIO

VAST Data: No new product announcements this week. Published detailed Dynamo 1.0 integration documentation (covered in Section 3) demonstrating 35 GB/s KV cache offload per GPU. This integration documentation is a competitive proof point that VAST is the leading Dynamo storage partner.

DDN: No material announcements during May 11–17.

IBM: No material storage announcements during May 11–17.

Hitachi Vantara: Named a Leader in 2026 GigaOm Radar for Object Storage (VSP One Object). No new product announcements this week.

MinIO: No material announcements during May 11–17.

Sources: KIOXIA 9.8 PB Announcement, Scality ADI Launch

---

6. Hyperscaler & Cloud Storage

Google Cloud NetApp Volumes — OpenShift Dedicated GA + iSCSI Block Storage

Two significant GCNV developments this week:

1. GCNV + Trident CSI for Red Hat OpenShift Dedicated: GA on Google Cloud. Full Trident integration for containerized enterprise workloads. Enables customers to consume ONTAP-managed persistent storage in managed OpenShift environments on Google Cloud.

2. GCNV iSCSI Block Storage: Now GA for the Flex Unified service level — enables migration of high-performance SAN workloads with sub-millisecond latency, dynamic scaling up to 5 GiBps and 160,000 IOPS per volume. This fills a critical gap: GCNV can now serve block workloads, not just file.

3. GCNV External Replication (SnapMirror): GA for Standard, Premium, and Extreme service levels — bi-directional SnapMirror between on-prem ONTAP and GCNV.

NetApp relevance: These three updates significantly strengthen GCNV's competitive position against native Google Cloud storage services and against competitors (Dell, HPE) who lack equivalent first-party cloud services. The iSCSI GA is particularly important — it opens GCNV to SAN workloads that were previously served by Persistent Disk or other block services.

AWS — S3 Access Points for FSx for NetApp ONTAP

Amazon introduced S3 Access Points for FSx for ONTAP, enabling direct access to AWS AI/ML and analytics services including Amazon Athena, SageMaker, QuickSight, and OpenSearch. This eliminates the need for data copies when using FSx for ONTAP data with AWS AI services.

NetApp relevance: This is a significant integration win. FSx for ONTAP customers can now feed data directly to SageMaker and other AI services through S3 Access Points without ETL or data movement. The talk track: "Your enterprise data lives in FSx for ONTAP with all the ONTAP management capabilities — and AWS AI services consume it natively through S3 Access Points. No copies, no movement, no additional cost."

[UPDATE] Azure, OCI

Azure Strategic Storage Services Partner Program (May 1, covered last week) continues. No significant new ANF announcements during May 11–17. Azure region in Saudi Arabia remains on track for Q4 2026.

Sources: Google Cloud NetApp Volumes Release Notes, NetApp × Red Hat Press Release

---

7. AI Data Pipeline & Training/Inference Infrastructure

NVIDIA Dynamo 1.0 + Storage: The Inference Infrastructure Stack Crystallizes

The most important technical development for storage vendors in the inference era is the crystallization of the Dynamo-based inference stack. This week's VAST Data integration publications make explicit what was architecturally implied at GTC: storage is a required tier in the Dynamo KV cache hierarchy.

How the stack works:

1. GPU HBM — hot KV cache blocks for active inference
2. CPU DRAM — warm blocks, accessed via NIXL
3. Local NVMe SSD — cool blocks, GPUDirect Storage path
4. Remote storage (S3/blob) — cold blocks, offloaded via KVBM

VAST Data demonstrated 35 GB/s throughput to a single H100 GPU at tier 3/4, with 10x prefill improvement for large-context scenarios. KVBM now emits events when blocks move between tiers, enabling intelligent routing.

Critical point for NetApp: ONTAP S3 is architecturally compatible with KVBM tier 4 (remote storage via S3 API). However, competitive performance at tier 3 (local/near-storage requiring GPUDirect throughput) demands EF-Series or AIDE-class performance. NetApp needs:

1. Validated ONTAP S3 benchmarks as a Dynamo KVBM backend
2. EF-Series/AIDE positioning for tier-3 KV cache with GPUDirect Storage
3. A published reference architecture showing ONTAP in the Dynamo stack

vLLM v0.20.2 — Model Runner V2 Delivers 56% Throughput Gain on GB200

vLLM's May 2026 stable release (v0.20.2) introduces Model Runner V2 (MRV2) with up to 56% higher throughput on GB200 via GPU-native Triton kernels and async scheduling. FP8 inference is now default on H100 and Blackwell. Combined with Dynamo 1.0 orchestration, this represents the production inference stack converging.

NetApp relevance: Higher inference throughput means more tokens per GPU per second — which means faster KV cache turnover and higher storage I/O demand at the cache offload tiers. Every improvement in inference engine performance translates to increased storage bandwidth demand. The "storage is the AI bottleneck" narrative (from last week) is reinforced by these throughput improvements.

Sources: NVIDIA Dynamo KV Cache Blog, VAST Data — Dynamo + VAST, LMCache + Dynamo

---

8. NAND Flash & SSD Market

Q2 2026 Contract Prices: 70–75% QoQ Increase Flowing Through

TrendForce's Q2 2026 NAND contract price projections (70–75% QoQ) are now flowing through to actual enterprise SSD pricing as contracts signed in late Q1 take effect. Key market dynamics confirmed this week:

• Supply constraint persists: Meaningful capacity expansion unlikely until late 2027/2028. Cloud service providers securing supply via long-term agreements, further limiting spot availability.
• Enterprise SSD allocation priority: NAND capacity increasingly directed toward enterprise SSDs (AI demand), with consumer/client segments scaled back.
• Client SSD restocking: Buyers restocking preemptively out of concern that server demand could absorb all available capacity.
• No relief in sight: Analysts see no demand softening through H1 2027 minimum.

KIOXIA LC9 Series: 245.76 TB SSDs Signal Density Inflection

The KIOXIA LC9 E3.L 245.76 TB SSD demonstrated at Dell Technologies World represents the largest-capacity enterprise SSD publicly demonstrated in a production server configuration. While current pricing is not disclosed, the density (9.8 PB per 2U) at PCIe 5.0 speeds signals that flash density continues advancing even during the supply crisis.

Procurement guidance for EMEA, LATAM, Middle East customers: The Q2 contract price increases are now landed. Customers who haven't locked in pricing are paying 70–75% more per TB than Q1. The case for ONTAP data reduction (4:1+ at crisis pricing) has never been stronger. Additionally, the KIOXIA density breakthrough suggests that customers planning large AI deployments should consider future-proofing capacity planning — drives of this density will eventually drive down cost-per-PB, but not before 2027–2028.

Sources: TrendForce Q2 2026 Forecast, Tom's Hardware — DRAM/NAND Prices, KIOXIA LC9 at DTW

---

9. EMEA, LATAM & Middle East Focus

Middle East — Iran Asserts Control Over Strait of Hormuz Undersea Cables (ESCALATION)

What changed this week: IRGC-linked media (Tasnim, May 9) announced that Iran will take "full control" of seven critical undersea fiber-optic cables passing through the Strait of Hormuz. Foreign operators must now obtain Iranian permits and pay "protection fees" to maintain infrastructure on the seabed. CNN's May 17 reporting confirms this represents Iran viewing undersea cables as a new geopolitical leverage tool — analogous to its historic control over oil tanker transit through the Strait. [CROSS-VALIDATED]

Scale of exposure: 17 cables transit the Strait of Hormuz, carrying approximately 30% of global internet traffic. The cables connect Gulf state data centers (UAE, Saudi Arabia, Qatar, Bahrain) to global internet backbone routes. All Gulf sovereign AI investments (Stargate UAE, AWS HUMAIN, NEOM) depend on this connectivity.

Impact on Gulf AI investments:

• Stargate UAE (200 MW Phase 1, Q3 2026): Connectivity to global AI model providers (OpenAI) depends on undersea cables transiting Hormuz. If Iran restricts or degrades cable access, Stargate's utility as a global AI hub is diminished.
• AWS HUMAIN Saudi Arabia: $5B AI Zone depends on AWS backbone connectivity, which routes through regional undersea cable networks.
• Microsoft Azure Saudi Arabia (Q4 2026): Same connectivity dependency.

NetApp relevance (CRITICAL): This escalation makes data replication and multi-site architecture a non-negotiable requirement for every Gulf data center customer. The architectural response: active-active deployments across Gulf + Europe (or Gulf + Asia), with real-time SnapMirror replication ensuring data survivability if cable connectivity degrades or is severed. MetroCluster for campus-level resilience. Specific customer conversation: "If Iran restricts cable traffic through Hormuz, can your AI training continue? Can your inference services failover? If the answer is no, you need a multi-site data architecture. ONTAP SnapMirror and MetroCluster were designed for exactly this scenario."

Europe — EU CADA Tech Sovereignty Package (May 27) Now 10 Days Away

The European Commission's flagship Tech Sovereignty Package — containing the Cloud and AI Development Act (CADA) and Chips Act 2.0 — is confirmed for May 27 adoption. After two delays (from March to April, then to May), this appears final. Key expected provisions:

• Restriction of U.S. cloud providers for sensitive EU government data (healthcare, finance, judicial systems)
• EU-wide eligibility requirements for cloud service providers handling sovereign workloads
• Tripling EU data centre capacity target
• Single EU-wide cloud policy for public administration procurement
• Sovereign cloud contracts worth €180M over six years already awarded (Post Telecom/OVHcloud, STACKIT, Scaleway, Proximus)

NetApp relevance: CADA will create binding requirements for EU-hosted data infrastructure. If U.S. cloud providers are restricted for sovereign workloads, on-premises ONTAP and European cloud options (ANF in EU regions, GCNV in EU regions) become the required architecture. CADA may also create opportunities for NetApp's sovereign data management capabilities — SnapLock, data residency enforcement, encryption, and audit logging. Field teams should prepare for a wave of customer questions after May 27 about sovereign-compliant storage architectures.

Europe — Sovereign AI Infrastructure Buildout Continues

• Mistral AI (France): €830M institutional debt for 13,800 NVIDIA chips + data center near Paris (operations starting Q2 2026). Working with NVIDIA on 18,000 Grace Blackwell system cloud platform.
• Nebius (Finland): Building one of Europe's largest AI factories (announced March 2026, construction ongoing).
• EU Sovereign Cloud Contracts: €180M awarded across OVHcloud, STACKIT, Scaleway consortia.
• Forrester: European tech spend exceeds €1.5 trillion in 2026, driven by AI, cloud, and sovereignty mandates.

Middle East — Gulf Investments Continue Despite Connectivity Risk

Stargate UAE: 200 MW Phase 1 on track for Q3 2026 (unchanged from last week). 5,000+ workers deployed. Full campus scales to 1 GW.

Saudi Arabia Total AI Investment: Exceeds $20B as of early 2026. HUMAIN has secured 211 land plots, developing 100 MW sites in Riyadh and Dammam. Joint venture with STC/center3 targeting 1 GW capacity.

MENA Tech Spend: $169B in 2026 per Gartner.

Latin America — Power Grid Constraints and AI Infrastructure

No major new developments this week beyond continued coverage of power grid constraints. Brazil's transmission bottleneck and Mexico's renewable capacity shortfall persist. The Mexico AI-GDC sovereign platform (covered last week) continues planning phase.

Sources: Iran International — IRGC Cable Control, CNN — Iran Hormuz Cables, Business Today — $10 Trillion Chokepoint, EU Parliament — CADA, CNBC — EU Cloud Restrictions

---

10. Open Source & Research

NVIDIA Dynamo 1.0: Production-Grade Open-Source Inference OS

While Dynamo 1.0 launched at GTC (March 2026), this week's publications from VAST Data and industry blogs confirm broad production adoption. Key ecosystem integration points:

• Supported engines: SGLang, TensorRT-LLM, vLLM
• Cloud adoption: AWS, Azure, GCP, OCI, CoreWeave, Together AI, Nebius
• Enterprise adoption: ByteDance, Meituan, PayPal, Pinterest, Cursor, Perplexity
• Storage integration: S3/blob API support for KV cache offload (KVBM), GPUDirect Storage for local tier

Dynamo's 7x inference performance boost on Blackwell (industry benchmarks) makes it the de facto inference orchestration layer. Any storage vendor not validated against Dynamo risks being excluded from the AI inference stack.

vLLM v0.20.2 (May 2026)

Model Runner V2 (MRV2) delivers up to 56% higher throughput on GB200 via GPU-native Triton kernels and async scheduling. FP8 inference default on H100/Blackwell. Continuous batching now default. Server-sent events for real-time token streaming.

SGLang Updates

SGLang maintains RadixAttention advantage for agent workloads: 29% throughput edge over vLLM on H100 (16,200 vs 12,500 tok/s on Llama 3.1 8B), up to 6.4x on prefix-heavy workloads (RAG, multi-turn chat). SGLang Diffusion for video/image generation. Day-0 support for latest models (MiMo-V2-Flash, Nemotron 3 Nano, Mistral Large 3).

No Significant New Papers, Funding Rounds This Week

No material new arXiv storage/AI papers or storage infrastructure funding rounds identified during May 11–17.

Sources: NVIDIA Dynamo 1.0 Newsroom, VAST Data + Dynamo, vLLM Release Notes

---

11. Regulatory & Compliance

EU CADA — Tech Sovereignty Package Confirmed for May 27

The Cloud and AI Development Act (CADA) proposal — twice delayed — is confirmed for European Commission adoption on May 27, 2026. This is now 10 days away and represents the most significant regulatory event for cloud and storage infrastructure in the EU this year.

Expected provisions:

• Restriction on U.S. cloud platforms processing sensitive EU government data
• EU-wide eligibility requirements for cloud service providers (expected to include data residency, sovereign control, and potentially encryption/key management requirements)
• Public procurement rules favoring sovereign cloud offerings
• Target to triple EU data centre capacity
• Bundled with Chips Act 2.0

NetApp compliance positioning:

• Data residency: ONTAP + ANF/GCNV in EU regions enforce data residency without cross-border exposure
• Sovereign control: SnapLock WORM, FPolicy logging, encryption with customer-managed keys
• Procurement eligibility: If CADA requires sovereign-aligned providers, NetApp's EU-region cloud services (ANF, GCNV) and on-premises ONTAP are compliant by design
• Risk for NetApp: If CADA restricts "U.S.-origin technology" rather than just "U.S. cloud providers," it could impact FSx for ONTAP (AWS service) and potentially ONTAP itself (U.S.-headquartered vendor). Monitor the specific wording.

[UPDATE] EU AI Omnibus — Formal Adoption Pending

The May 7 provisional agreement on the AI Omnibus (covered last week) awaits formal adoption by both Parliament and Council. Both institutions intend to complete formal adoption before August 2, 2026. No changes to the deferred timelines (Annex III: December 2027; Annex I: August 2028).

Iran — Cable Control as Regulatory Risk for Gulf Infrastructure

Iran's assertion of control over Hormuz undersea cables (detailed in Section 9) creates a de facto regulatory risk for Gulf data infrastructure. Any organization operating data centers in the Gulf must now consider:

• Cable access continuity as a compliance risk (if service depends on uninterrupted connectivity)
• Data sovereignty implications if traffic must transit Iranian-controlled infrastructure
• Whether Iranian "protection fee" demands constitute sanctions risk for Western companies

NetApp capabilities addressing these risks:

• SnapMirror: Real-time replication to secondary sites outside Gulf/Hormuz dependency
• MetroCluster: Synchronous replication for zero-RPO within campus/metro distances
• FabricPool/Cloud Tiering: Data copies in cloud regions (EU, APAC) outside the Hormuz path
• Encryption: End-to-end encryption of data in transit and at rest — relevant if traffic transits adversary-monitored infrastructure

Sources: EU CADA Legislative Train, CADE — EC Delay, Kiteworks — May 27 Tech Sovereignty

---

12. Methodology & Confidence Markers

Research Passes Completed

Total unique queries executed: 31 WebSearch + 3 Perplexity = 34 queries

Confidence Summary

Deduplication Actions

The following signals from the May 10 briefing were intentionally excluded as "already covered" unless a material new development occurred this week:

• EU AI Omnibus provisional deal (May 7) — no change; awaiting formal adoption (covered May 10)
• NVIDIA $40B+ investment blitz (Corning, IREN) — no new developments (covered May 10)
• IREN × Nostrum Group Spain acquisition — no update (covered May 10)
• SanDisk Q3 FY2026 / WDC / memory supercycle — referenced as context for NAND pricing; no new earnings data (covered May 10)
• "Storage as AI bottleneck" narrative — now manifesting in vendor actions (Dynamo integration, HPE positioning); context referenced but not repeated as a signal (covered May 10)
• WEKA Augmented Memory Grid / NeuralMesh — no new announcements (covered May 10)
• Dell Technologies World preview — SUPERSEDED by actual pre-event announcements (material update, fully covered in this report)
• Everpure ticker change / Accelerate preview — SUPERSEDED by 1touch close (material update)
• Stargate UAE / AWS HUMAIN — status updates included under Hormuz escalation context (covered May 10 and prior)
• NetApp Q4 FY2026 earnings date — unchanged, May 28 (covered May 10)
• Strait of Hormuz cable crisis — MATERIAL ESCALATION (Iran asserting control); fully covered as new signal
• Azure Strategic Storage Services Partner Program — no update (covered May 10)
• TrendForce Q2 NAND 70-75% QoQ — referenced as context; now flowing through to contracts (covered May 3, May 10)

Key Information Gaps

• Dell Technologies World day-of announcements (May 18–21): The keynote content summarized here is from pre-event briefings and analyst reports. Additional announcements may emerge during the event itself — next week's briefing will capture any new developments.
• AIDE broad availability date: Still "early summer 2026" — no specific date. With Everpure's 1touch now closed and Dell's data intelligence capabilities shipping, the timing pressure on AIDE is acute.
• NetApp + Dynamo validation: No published benchmarks for ONTAP S3 as a Dynamo KVBM backend. VAST Data has published; NetApp has not.
• CADA exact wording: The May 27 proposal text is not yet available. The difference between "U.S. cloud providers" and "U.S.-origin technology" has significant implications for NetApp.
• Everpure Q1 FY2027 earnings (May 27): Will provide first financial data on the combined entity and may include 1touch integration details.
• HPE X10000 file performance benchmarks: HPE announced native file but has not published throughput benchmarks vs. ONTAP NFS — competitive comparison data is needed.
• Iran cable situation: Uncertain whether Iran will enforce permit/fee demands or whether this is posturing. Either way, it changes customer architecture conversations.

---

This briefing was generated on May 17, 2026. Next edition: May 24, 2026.

Research pipeline: 4-pass multi-source (WebSearch × 31 + Perplexity × 3). All claims tagged with confidence markers. Source URLs provided throughout.

Key upcoming events: Dell Technologies World (May 18–21, THIS WEEK), EU CADA Tech Sovereignty Package (May 27), Everpure Q1 FY2027 Earnings (May 27), NetApp FY2026 Q4 Earnings (May 28), Everpure Accelerate (June 16–18).