Dell AI Data Platform Introduces Only 4-in-1 Storage for AI

• The Dell AI Data Platform storage engines provide purpose-built access to file (PowerScale), object (ObjectScale) and parallel file (Lightning File System), with each engine optimized for a distinct role in the AI data lifecycle. Block (PowerFlex) support targeted for 1H CY2027¹.
• For infrastructure providers with rapidly changing tenant requirements, Dell Exascale Storage runs these engines as software-defined personalities on a shared Dell PowerEdge foundation, starting with PowerEdge R7725xd.
• A single Term TLA with substitution lets operators reallocate licensed capacity across personalities as tenant mix changes, without starting a new procurement cycle.
• Exascale is a software-first architecture, not a replacement for Dell’s appliance families, which remain the standard choice for most enterprise deployments.

Why storage is the foundation of the Dell AI Data Platform

Picture a neocloud that wins a large AI training contract. They stand up a dense parallel file environment, optimized for throughput. Six months later, that same customer shifts toward inference and long-term retention — and suddenly needs high-capacity object storage at scale. Under a traditional model, that shift means a new hardware decision, a new procurement cycle, and a new silo sitting alongside the last one.

This is the operating reality for cloud service providers and neoclouds. They are not building against a single, predictable workload. They are building for tenants whose AI projects move from experimentation to training to inference — and whose storage demands move with them. The cost of getting that wrong is not inefficiency. It is stranded capacity, renegotiated contracts and infrastructure that can’t keep pace with the business it supports.

The problem is not that the right storage doesn’t exist. It’s that AI at scale needs more than one kind, and the industry has made you buy each one separately.

The Dell AI Data Platform is built to turn siloed data into AI-ready fuel across the lifecycle, from ingest to inference — spanning Storage Engines, Data Engines, Data Orchestration, and Cyber Resilience. Its foundation is storage, and its Storage Engines give AI clouds a dedicated, best-of-breed engine for each role:

• PowerScale — the engine for active file-based AI data. Built on OneFS, with NCP-certified performance, GPUDirect and RDMA support for data preparation, high-throughput pipelines, and inference.
• ObjectScale — the engine for persistent object data at enterprise scale. S3-native, from petabytes to exabytes under a unified global namespace, with S3 over RDMA for high-performance object access.
• Lightning File System — the engine for extreme-performance parallel access. Optimized for the most demanding AI training and inference environments, with up to 2x greater throughput per rack unit than competing parallel file systems².

Each engine is the right answer for its role. The question Dell Exascale answers is what happens when a tenant needs a different one six months later.

Dell Exascale Storage: one foundation, more than one answer for AI clouds

Dell Exascale Storage lets operators run PowerScale, ObjectScale and Lightning File System as software-defined personalities on the same qualified PowerEdge hardware, with PowerFlex block joining them in 1H CY2027. The neocloud in the scenario above doesn’t buy a second system when its tenant moves from training to inference. It changes what the system is.

That is the whole idea, and it has two halves.

The first is architectural. Every other approach to this problem asks you to pick a lane: a parallel file system vendor, an object vendor, a file vendor, a block vendor. Each is excellent inside its lane and has nothing to offer outside it. When your tenant’s workload leaves that lane, so does the value of the infrastructure you bought. Exascale is the alternative to picking — file, object, and parallel file today, block on the roadmap, each a first-class engine rather than one engine pretending to be four.

The second is commercial, and it’s the part that’s hard to copy: a single Term TLA with substitution means the licensed capacity itself moves between personalities. A vendor with one personality has nothing to substitute into. No other storage provider brings four storage types to a common software-defined foundation this way — and none pairs it with a license that lets capacity follow the workload.

Behind both halves is Dell — PowerEdge qualification, supply-chain scale and global services on a software-defined storage layer. Software-defined storage on standard servers is not new. Software-defined storage that Dell stands behind, deploys, and supports at CSP scale is.

Exascale is a software-first architecture, not a replacement for Dell’s appliance families. For most enterprise deployments, where an integrated hardware-software-support experience is the priority, the appliances remain the standard choice. Exascale is for operators whose workload mix is decided by someone else.

A common hardware foundation

This model starts with a common hardware foundation. Dell Exascale Storage runs on qualified, storage-optimized Dell PowerEdge servers, beginning with PowerEdge R7725xd, so each storage personality benefits from the latest generation of server, networking, power, and cooling infrastructure — along with Dell’s supply-chain scale.

Common foundation does not mean shared abstraction. Each personality provisions directly to bare metal. There is no virtualization layer between the software and the hardware, and no lowest-common-denominator storage service that every personality has to fit inside. OneFS is still OneFS. ObjectScale is still ObjectScale. What changes is the hardware they run on, not what they are.

That matters because the flexibility is worthless if it costs performance. Running the Lightning File System personality, Exascale delivers up to 6 TB/s of read performance per rack across sequential and random workloads3. Network connectivity scales to 800 GbE per node with support for NVIDIA CX-8 and CX-9 SuperNICs, helping bandwidth scale with node count as accelerated compute grows.

Capacity that follows demand: how substitution works

Software is licensed through a single Term Transformational License Agreement — one contract covering capacity across every supported personality, rather than a separate license per product.

Substitution is how that capacity moves. An operator contracts for a total capacity entitlement and allocates it across personalities as tenant demand dictates. When the mix changes, they reallocate against the agreement they already hold rather than opening a new one.

The practical effect: a neocloud can start weighted toward parallel file for a training tenant, shift toward object as that tenant moves to inference and retention, and add file capacity as another tenant’s data prep pipeline scales — all inside the same entitlement. No new procurement cycle, no second silo.

Storage that moves with the business

Go back to the neocloud from the beginning. Their tenant’s training contract winds down, inference and retention ramp up, and the parallel file capacity they bought six months ago is pointed at the wrong problem. Under a traditional model, that is a new hardware decision, a new procurement cycle, and a second silo. Under Exascale, it is a substitution against a license they already hold, on hardware they already own.

For CSPs and neoclouds, the measure of a storage platform is not raw performance or capacity. It is whether the infrastructure stays productive as the services running on it evolve. Every new hardware silo adds operational complexity, procurement overhead, and stranded capacity risk. Exascale takes that compounding cost off the table by making storage personality a software decision instead of a hardware one.

That is the difference between capacity that is fixed at the moment of purchase and capacity that moves with the business.

Dell Exascale is available now. Learn more here, or reach out to your Dell account team to get started.

FREQUENTLY ASKED QUESTIONS

Does Dell Exascale Storage replace Dell’s storage appliances?

No. Exascale is a software-first architecture, not a new appliance family. PowerScale, ObjectScale and Dell’s other appliance families remain the standard choice for most enterprise deployments, where a more prescriptive, integrated hardware-software-support experience is the priority. Exascale is built for environments that need to run file, object, block and parallel file as flexible services on shared infrastructure.

Which storage personalities run on Exascale, and when?

Exascale supports PowerScale (file), ObjectScale (object), and Lightning File System (parallel file). PowerFlex (block) is targeted for 1H CY2027.

How is capacity rebalanced across storage types?

Through a single Term TLA with substitution. Customers contract for software capacity across supported products and can shift how that licensed capacity is allocated over time without renegotiating the agreement.

How does Exascale relate to the Dell AI Data Platform?

Exascale defines a common way to run Dell’s core AI storage software on high-performance PowerEdge infrastructure. Where customers adopt the Dell AI Data Platform, Exascale serves as the storage foundation underneath the platform’s Storage Engines layer. It can also be deployed independently as a storage architecture.

¹ Based on publicly available documentation from leading enterprise storage vendors as of May 2026. Comparison refers to distinct file, object, block, and parallel-file engines on one reusable hardware platform, excluding single-engine multi-protocol designs. Block availability in 1H CY2027.
² Based on Dell preliminary testing comparing random and sequential throughput per rack unit, May 2025. Actual performance may vary.
³ Based on internal analysis of sequential and random read I/O for Lightning File System, February 2026. Actual results may vary.