Skip to main content

    Artificial Intelligence

    Dell AI Data Platform Makes Data Agent-Ready at SC26

    At SC26, Dell expands the Dell AI Data Platform to turn siloed enterprise data into governed context for AI agents on customer-controlled infrastructure.

    By George Papadopoulos
    Topics in this article:Artificial Intelligence

    Key takeaways:

    8 min read

    At SC26, Dell is expanding the Dell AI Data Platform to help enterprises turn siloed structured and unstructured data into governed context for production AI agents — on customer-controlled infrastructure, without forcing data movement.

    Here’s what’s new:

    • Data orchestration: New Agentic Data Gateway, hybrid pipelines and storage and data intelligence connect agents to trusted enterprise data.
    • AI storage: Dell will be expanding the Exascale portfolio of qualifying common hardware with the world's first direct liquid-cooled storage¹ and a high-density 1U platform, both expected in 1H 2027. Dell is also announcing that software-defined PowerScale on Dell Exascale has achieved NVIDIA-Certified Storage validation at the Foundation level.
    • Partner integrations: LangChain, OpenAI Codex, and Unstructured.io bring familiar agent, development and document-processing tools into the platform.

    Nearly every organization we talk to asks the same question: how do we move beyond AI pilots and effectively take advantage of agents to accelerate or replace our business processes? Usually, the answer has less to do with models or GPUs than with data. Agents need access to enterprise data, and they need to turn it into knowledge they can act on. Data without context is noise, and speed without context is a liability.

    That is why context is becoming as important as compute. At SC26 in Chicago, Dell is expanding the Dell AI Data Platform, the data foundation of the Dell AI Factory with NVIDIA, from an unstructured data pipeline engine into an open, agent-ready data platform. It turns siloed enterprise data into trusted, governed context that any application or agent can use, delivered on your own infrastructure and kept fully sovereign, allowing you to achieve the outcomes you need.

    The announcements cover three areas: Data Orchestration, Storage Engines and partnerships.

    Data Orchestration: from data pipelines to agent-ready context

    Data Orchestration in the Dell AI Data Platform adds four connected capabilities: an Agentic Data Gateway that gives external agents secure access to approved models, data and MCP tools across the Dell AI Data Platform, hybrid orchestration that runs pipelines wherever data and compute live, Storage Intelligence and Data Intelligence that help you know and trust your data and open access across structured and unstructured sources in a single workflow. With Data Orchestration, context is built once from governed data and then reused, so agents stop rebuilding the same understanding on every request. In Dell internal testing, this cut total token consumption by up to 76% and raised answer accuracy from 60% to 94%.²

    One governed front door for agents

    Agents are multiplying faster than enterprises can connect, secure and audit them. When every agent hand-codes its own access to semantic search, federated SQL and data pipelines and manages its own API keys, the result is credential sprawl, redundant retrieval and no record of what an agent saw or did. Agentic Data Gateway removes that sprawl. It is built into the Data Orchestration Engine, the layer where apps and pipelines already run, and it gives every agent one front door to AI Data Platform’s models, tools, knowledge and data engines. Agents built inside the Data Orchestration Engine and external agents built in Claude Code, Cursor and Codex connect the same way. Each call runs with the caller’s identity, data scope and budget and is traced end to end.

    Agentic Data Gateway is a customer-controlled Model Context Protocol (MCP) server and AI agent gateway. It exposes OpenAI-compatible and MCP endpoints, so existing agents and clients connect without custom integration. It routes requests to internal and external LLMs and MCP services, and it keeps an MCP registry for all the data services inside AI Data Platform such as SQL, Search, Storage Intelligence and others so agents can discover what is available. External MCP servers can also be registered to help agents securely interact with other services such as systems-of-record, alerting tools, web search and others. Credentials are managed centrally, so agents no longer need to carry their own API keys. Retrieval, authorization, quotas, token accounting and observability are handled in one place, giving teams an auditable path to thousands of concurrent agents. Agentic Data Gateway runs entirely inside the customer's environment and does not store prompts or long-lived secrets.

    Pipelines that run where data and compute live

    Data has gravity, especially unstructured data, and moving it creates delay, cost and another silo. Hybrid data orchestration lets the Data Orchestration Engine deploy and run pipeline containers on many kinds of customer-provided compute: Kubernetes clusters in different data centers, Dell AI Factory infrastructure including NVIDIA and AMD environments, non-Dell infrastructure and public cloud endpoints. Dell deploys, scales, observes and tears down those containers automatically. This helps customers reuse the GPU capacity they already own at their sites.

    Know and trust your data

    Many AI projects stall at step zero: no one knows where useful data lives. Storage Intelligence scans storage arrays to build a metadata repository of the files they hold. Teams can then see what they have and curate the relevant files into datasets for their pipelines. Storage Intelligence gives storage administrators, IT managers and security teams a unified view of data ownership, usage and cost spanning both Dell and non-Dell storage. They can act on these insights directly to optimize capacity, enforce compliance and prepare data for AI.

    When an agent acts unexpectedly, it’s critical to have the ability to answer what context it saw and where that context came from. The new Data Intelligence capability provides end-to-end lineage, data quality and data observability across all components in AI Data Platform. It also adds metadata management, search across tables, datasets and vector indexes and a business glossary. Together, these lay the foundation for a semantic layer that is essential for agents to understand business processes within an organization.

    The real power comes from using the two together. Storage Intelligence shows what data exists, who owns it and which files are ready for AI. Data Intelligence follows that data into pipelines and agent answers with lineage, quality and observability. Teams can find the right data, prepare it with confidence and trace any agent’s answer back to the files and steps behind it.

    Open and unified: structured and unstructured data together in the same workflow

    A new structured-data browser and pipeline brings tables directly into datasets. Structured data and unstructured content such as images, video, PDFs, JSON and document chunks can now be prepared together in the same pipeline. New bring-your-own vector database integrations send vectors either to Dell's Data Search Engine or to external databases such as pgvector. A decoupled architecture scales storage and data engines independently, including up to 12x faster vector indexing.³ No rip-and-replace required.

    Dell Exascale: the best storage for AI pipelines

    AI is changing storage from a passive capacity layer into an active part of the intelligence pipeline. As organizations move from isolated model training to agentic systems that continuously retrieve, reason over and act on enterprise data, storage is becoming central to how quickly, reliably and responsibly AI can work with information.

    That means storage decisions can no longer be based on capacity alone. Organizations need a flexible foundation that can keep pace with growing data volumes, changing AI workloads and rising expectations for performance, security and governance. The strongest architectures leave room to evolve to production-scale AI, without locking every workload into the same design.

    This shift matters most at the scale of the world's largest AI deployments, where data volumes, GPU clusters and performance demands grow together. At the storage layer of the Dell AI Data Platform, Exascale provides an adaptable, software-defined foundation for the entire AI pipeline, with file, object, parallel file and block storage on one 4-in-1 platform.

    At SC26, Dell is announcing that it will expand the Exascale portfolio of qualifying common hardware with new options based on the latest PowerEdge platforms. Customers will get more choice in how they align storage to their data center priorities, from bringing storage into liquid-cooled, rack-scale AI infrastructure to fitting more drives into a compact footprint. Both options are expected in 1H 2027:

    Liquid-cooled form factor for AI racks. This will be the world's first direct liquid-cooled storage¹ delivering distinct file, object, block and parallel file storage software on reusable common hardware, with 32 drives in a 2OU form factor built for Open Rack standards. Storage will be able to share the rack with liquid-cooled compute and networking. That helps data centers handle rising rack density and plan liquid cooling across the whole infrastructure rather than working around it.

    Higher-density 1U form factor. A 20-drive configuration that will deliver 67% higher drive density per rack unit.⁴ Customers get more storage in the same rack footprint and a compact way to scale storage alongside dense compute.

    Dell is also announcing that software-defined PowerScale on Dell Exascale has achieved NVIDIA-Certified Storage validation at the Foundation level. This confirms it is ready for AI-accelerated data access in large GPU environments and gives customers a validated starting point as they scale.

    Partnerships: meeting teams where they already work

    The Dell AI Data Platform is open by design. Three new technical partnerships bring tools that many teams already use to governed data that stays on their own infrastructure.

    LangChain

    Dell is collaborating with LangChain to bring LangSmith’s tracing and evaluation tools into Data Orchestration. Together, they give enterprises the visibility and control needed to run AI agents in production with confidence. With LangSmith integration, Dell AI Data Platform users can enable full traces for pipelines, from model calls to tool calls to outputs, so that the teams can debug failures and test changes before they ship. The LangSmith integration will be available through the Dell AI Data Platform Marketplace, Dell's hub for discovering and deploying applications, models and pipelines through Data Orchestration.

    OpenAI Codex

    Dell and OpenAI bring secure Codex development environments to Dell infrastructure using Red Hat OpenShift Dev Spaces and Dell AI Data Platform plugins. Codex can securely access governed search, analytics and orchestration capabilities to help teams investigate data issues and build, test and improve data pipelines. Code execution and enterprise data processing remain on customer-controlled infrastructure, while only the context required for inference is sent securely to OpenAI-hosted models under the customer’s applicable data policies and agreement.

    Unstructured.io

    A new File Transform application powered by Unstructured.io brings enterprise-grade document parsing, chunking and embedding into the Data Orchestration Engine. Options range from fast processing to Vision Language Model extraction. Ready-to-use RAG Preprocessing Pipeline templates take teams from raw documents to searchable knowledge bases, and Document Chat gives answers with sources cited. Both install from the Data Orchestration Engine Marketplace in minutes.

    Context is infrastructure

    Customers who build governed context are already seeing results. CTBC Bank, Taiwan's largest private bank, runs agentic AI across more than 100 data sources on the Dell AI Data Platform. Fraud detection time dropped from weeks to 0.03 seconds with no costly data movement, and the bank prevented more than $60 million in fraud losses in 2025.

    Compute gives you tokens. Context makes them count. With these announcements, the Dell AI Data Platform helps enterprises build that context where their data already lives, on infrastructure built for agents, and open to the models and frameworks they choose, with Dell Services available to help organizations design, carry out and operate it. Visit us at SC26 in Chicago, November 15–20, to see it in action.

    Learn more at Dell AI Data Platform.


    FREQUENTLY ASKED QUESTIONS

    What new Data Orchestration capabilities are coming to the Dell AI Data Platform?

    Dell is expanding the Data Orchestration Engine into an open, agent-ready data platform. New capabilities include Agentic Data Gateway, hybrid orchestration, Storage Intelligence, Data Intelligence and unified support for structured and unstructured data. Together, they help organizations connect agents to governed data, run pipelines wherever data and compute reside, understand what data they have and prepare multiple data types for AI workflows.

    How does Dell help customers keep control of their data and existing infrastructure?

    Agentic Data Gateway is designed to run inside the customer’s environment without storing prompts or long-lived secrets. Hybrid orchestration can run pipelines on customer-selected infrastructure, including Kubernetes clusters, Dell AI Factory systems, non-Dell infrastructure and public cloud endpoints. Customers can also continue using existing open formats, vector databases and tools, reducing the need for a rip-and-replace approach.

    What do the new LangChain, OpenAI Codex and Unstructured.io partnerships enable?

    The partnerships connect familiar developer and data tools to governed enterprise data on customer-controlled infrastructure. LangChain and LangGraph agents can connect to LangSmith for visibility into reasoning, tool calls, model calls and resource usage. OpenAI Codex plugins connect to Dell’s Data Analytics, Data Search and Data Orchestration Engines. Unstructured.io adds document parsing, chunking, embedding, RAG preprocessing templates and Document Chat with cited sources.

    How does storage differ for AI inference versus AI training workloads?

    Training is sequential and throughput-bound — you need massive sustained bandwidth to keep thousands of GPUs fed with data continuously. Fast checkpoint writes matter enormously: if a job fails, you can resume from a recent checkpoint instead of starting over. Dell storage engines are purpose-built for this, with NVIDIA GPUDirect Storage, NFSoRDMA and pNFS delivering parallel multi-stream throughput at scale.

    Inference is latency-bound and concurrent — you're serving many simultaneous requests and time-to-first-token (TTFT) is the critical SLA. The biggest lever is KV Cache offload: as part of the Dell AI Data Platform, NVIDIA CMX enables KV cache offload to shared Dell AI Storage — delivering up to 19× faster TTFT vs. standard vLLM (Read more) and letting you serve more concurrent sessions on the same GPU hardware. S3 over RDMA adds 230% higher throughput and 80% lower latency on object-heavy inference pipelines.

    Why is Dell Exascale the best storage for large, rack-scale environments?

    Dell AI Storage, with Exascale at its storage layer, provides an adaptable, software-defined foundation for dense AI infrastructure. It is the only 4-in-1 storage built for AI, HPC and demanding workloads, and Dell has announced that it is expanding the Exascale portfolio to include additional common hardware options, i.e., a higher-density 1U platform with 20 drives and 67% higher drive density per rack unit, along with a 2OU, 32-drive liquid-cooled platform designed for Open Rack standards. These options help storage keep pace with dense compute, share rack space with networking and support liquid-cooled, rack-scale AI environments, with expected availability in 1H 2027.

    How do NVIDIA certifications strengthen the Dell AI Data Platform?

    The Dell AI Data Platform brings NVIDIA-validated storage into a unified foundation for AI workloads. Dell Exascale for software-defined file storage is now NVIDIA-Certified at the Foundation level, simplifying deployment for AI-accelerated data access in large-GPU environments. Across the broader Dell storage portfolio, Dell PowerScale F710 is certified at the NVIDIA Cloud Partner and NVIDIA-Certified Enterprise Storage levels and is validated for NVIDIA DGX SuperPOD and DGX BasePOD. Together, these give customers a stronger path from AI projects to production-scale deployments.

    How does Dell storage help keep GPUs productive?

    Dell AI Storage helps keep GPUs productive by supporting parallel reads, small-file access, mixed sequential and random workloads, high concurrency, GPU-direct access, RDMA, intelligent data placement and caching. These capabilities are designed to reduce bottlenecks between data and compute.


    ¹ Based on Dell analysis, November 2026, comparing enterprise storage systems offered with storage software and announced for commercial availability. 'Direct-liquid-cooled' means fanless cold-plate liquid cooling of all heat-producing parts. 'For rack-scale AI' means built to use the same liquid-cooled rack design and coolant supply as AI systems that run a full rack of accelerators as one. Applies to the liquid-cooled 2OU option, expected early 2027. The platform brings together distinct file, object, block and parallel-file storage personalities on a common reusable hardware architecture, with block planned for the first half of 2027.

    ² Based on Dell internal testing, September 2026, using a synthetic toll-plaza operations corpus and the benchmark configuration described in the AIDP Tokenomics benchmark study (20 questions, 6 workload types. NVIDIA Nemotron-3-Super-120B-A12B[2] via NVIDIA [3]NIM). NVIDIA NIM, part of NVIDIA AI Enterprise, is a set of easy-to-use microservices designed for secure, reliable deployment of high-performance AI model inferencing across clouds, data centers, and workstations.[4] Token counts are from the LLM API usage field. Actual results will vary based on workload, data characteristics, model selection and system configuration.

    ³ Based on Elastic’s published benchmark of Elasticsearch with NVIDIA cuVS, December 2025. Testing used a single AWS g6.4xlarge instance with locally attached NVMe storage, 2.6 million vectors with 1,536 dimensions, and recall levels of up to 95%. Compares GPU-accelerated vector indexing with CPU-only indexing on the same hardware. Actual results may vary.

    ⁴ Based on 1U form factor with up to 20 drives, compared to previous generation 2U qualified common hardware with 24 total drives per node. Expected availability early CY2027. Nov, 2026.


    George Papadopoulos

    About the Author: George Papadopoulos

    With over 20 years in the storage industry, I lead innovation and deliver market-leading solutions as Senior Vice President of Unstructured, Secondary Storage & Portfolio Engineering at Dell Technologies. Overseeing a $5B portfolio, I drive flagship products like PowerScale, ObjectScale, and PowerProtect, while scaling engineering organizations and transforming development practices. Passionate about aligning technology with business outcomes, I specialize in storage modernization, agile execution, and fostering cross-functional collaboration. If your organization is tackling unstructured data challenges or rethinking storage strategies, I can help you achieve impactful results.