The Dell AI Data Platform is getting an Elastic-powered Data Search Engine and a Data Analytics Engine built with Starburst, along with a Data Analytics Engine Agentic Layer and MCP Server, plus Nvidia cuVS integration.
The AI Data Platform is a group of components in Dell’s AI Factory, providing the infrastructure part. It has four elements: Storage Engines meaning PowerScale and ObjectScale, a set of Data Engines for analytics, processing and search, Cyber Resilience, and Professional Services to help implement a deployment. This platform separates storage from compute (processing) and is integrated with Nvidia’s AI Data Platform reference design. It’s designed for AI workloads like training, fine-tuning, retrieval-augmented generation (RAG) and inferencing.
Dell’s Arthur Lewis, Infrastructure Solutions Group President, said: “The Dell AI Data Platform is purpose-built to simplify data complexity, unify pipelines and deliver AI-ready data at scale. From real-time diagnostics in healthcare to predictive maintenance in manufacturing, Dell Technologies and trusted collaborators like Nvidia, Elastic and Starburst are empowering industries to move from AI pilots to production faster and with reduced risk.”
The underlying PowerScale scale-out, clustered file + object system storage storage has Nvidia Blackwell GB200 and GB300 NVL72 integration and other software updates. Dell positions PowerScale as needing up to eight times fewer network switches than VAST Data and Pure Storage, using up to 72 percent less electricity, and needing up to 80 percent less rack space:
In particular, the PowerScale F710, which has achieved Nvidia Cloud Partner (NCP) certification for high performance storage, delivers 16k+ GPU-scale with up to 5x less rack space, 88 percent fewer network switches, and up to 72 percent lower power consumption compared to these two competitors.
There is an ongoing Dell Project Lightning to add parallelism for higher data transmission bandwidth to PowerScale. Dell told us: “The team has done a ton of work on PowerScale OneFS software to better parallelize IO – from locking, to protocol work to back-end parallelism. The team has also delivered an optimized client-side driver to push more throughput to clients capable of 400-plus Gbps. Finally, we’ve always had the concept of compute/network-only ‘accelerator’ nodes and as of our introduction of the PA110, these Performance Accelerators unlock a lot more streaming and metadata performance without adding more storage. Watch this space. There’s more to come (very) soon.”
Dell’s cloud-native S3 Object Scale storage is getting S3 over RDMA support for lower latency object data access, performance tuning for large and small object workloads, reduced IO for high-volume, metadata-intensive workloads, and all-flash configurations with high-capacity drives to uplift the capacity per node.
ObjectScale is available as an appliance or through a new software-defined option on Dell PowerEdge servers that is up to eight times faster than previous-generation all-flash object storage (ECS 3.8 on EXF900 for object read performance). The product also has deeper AWS S3 integration and bucket-level compression.
Data Engines
The Data Search Engine project was initially revealed by Dell in August. It has been developed in collaboration with Elastic for indexing and search, and integrates with MetadataIQ data discovery software to search billions of files on PowerScale and ObjectScale using granular metadata.
Dell says it’s designed for workloads such as retrieval-augmented generation (RAG) and enables “customers to interact with data as naturally as asking a question.” Developers can build RAG applications in tools like LangChain with the engine, ingesting only updated files to save compute time and keep vector databases current.
Vector search is helped by Nvidia cuVS integration, cuVS being a Github-downloadable library for vector search and clustering on a GPU. It takes advantage of parallel pipelines to cut semantic search time.
A Data Analytics Engine has been developed with Starburst, whose technology uses Trino open source distributed SQL to query and analyze distributed data sources. It enables data querying across spreadsheets, databases, cloud warehouses and lakehouses. A component Engine Agentic Layer transforms raw data in seconds, “using LLMs to automate documentation, glean insights, and embed AI into SQL workflows.”
It has >50 data source connectors and “unifies access to vector stores, enabling RAG and search tasks across Iceberg, Dell’s Data Search Engine, PostgreSQL + PGVector and more.”
There is MCP server support and an agent API with the MCP server for Data Analytics Engine, enabling multi-agent and AI application development.
Comment
Dell is building an AI data pipeline stack to support LLMs and agents. It’s taken the view that providing basic storage in this AI era, no matter how fast and capacious, is not enough in this Gen AI era. Nor is it enough to support external AI pipeline components. Dell wants to support the whole pipeline, from multi-source, ingest, selection, filtering, transformation, and search to analytics and agents. All the main enterprise storage suppliers are realizing and responding to this view, and Dell is not going to be found wanting.
There is more to come, with faster PowerScale storage just one example.
Availability
- PowerScale GB200 and GB300 NVL72 integration with NCP validation is available now.
- ObjectScale S3 over RDMA will be available in Tech Preview in December 2025, along with other SW updates.
- The first release of the Dell Data Analytics Engine Agentic Layer and the Data Analytics Engine MCP Server will be available in February 2026.
- The Data Search Engine and cuVS integration will be available in 1H 2026.
