Cloudian has launched a HyperScale AI Data Platform (AIDP), with its HyperStore S3-compatible object storage being the datastore for AI models and agents running on Nvidia GPU hardware and software.
HyperScale AIDP is built on the Nvidia RTX PRO 6000 Blackwell Server Edition GPU in line with Nvidia’s AI Data Platform reference design. HyperStore supports the S3 storage protocol over remote direct memory access (RDMA) to deliver faster object storage I/O performance to the Blackwell GPUs and so speed up AI response generation. Cloudian says HyperScale AIDP makes its customers’ unstructured data available to GenAI large language models (LLMs) and agents so that they can serve their staff and customers better.
Cloudian CTO Neil Stobart stated: “Rather than forcing enterprises to build complex, separate infrastructure for AI – which requires skills most don’t possess – we’ve engineered a platform that automatically transforms existing data into actionable intelligence at the storage layer.” As he puts it: “The HyperScale AI Data Platform represents a fundamental shift in how enterprises approach AI readiness.”
Cloudian’s thinking here is that, up until now, “AI implementations require enterprises to build complex, dedicated file system structures and separate vector databases to achieve optimal performance.” This means its exabyte-scale HyperStore and HyperScale products can enable “enterprises to deploy AI applications directly on their native S3-compatible data sources without requiring additional file system layers or separate vector database infrastructure.”
Vector databases hold the mathematical transformations of source unstructured data items that are used in semantic search by LLMs, and HyperStore has its own built-in vector database. The Cloudian HyperScale AI Data Platform automatically ingests, embeds, and indexes multimodal unstructured content, making it instantly searchable through vector search interfaces and immediately available for retrieval-augmented generation (RAG) workflows.
The company says that with its HyperScale AIDP, a customer’s AI agents “can access, search, and analyze information in near real-time,” with pretty much instant access to all the organizational knowledge – the reports, manuals, presentations, and multimedia content – stored in HyperStore. Incoming “data is instantly classified, enriched with real-time metadata, and vectorized as it’s stored, eliminating manual data preparation processes and complex infrastructure provisioning.”
The S3 object storage RDMA facility provides “direct data paths between storage, system memory, and GPU memory for ultra-low latency and up to 35 GBps per node (reads), scalable to TBs per second.”
Cloudian asserts that “organizations can now build AI applications that securely access enterprise data with reduced overhead and faster time to deployment, all while maintaining full control over their most valuable asset – their data.”
A Cloudian blog, Unlocking Enterprise Knowledge with Cloudian HyperScale AIDP, says: “Most people are familiar with chatbots that give generic responses without benefit of your enterprise knowledge. Or search engines that return endless lists of links to sort through. This system is fundamentally different because it applies agents, making decisions and reasoning through problems without needing step-by-step instructions from humans.”
The system “employs Llama-3.2-3B-Instruct, Meta’s instruction-tuned transformer model optimized for dialogue and reasoning tasks, featuring a 128K token context window. A diagram shows the request-response data flow:
Its four-GPU compute infrastructure has allocates each GPU specific role:
- GPU-1 – Llama-3.2-3B-Instruct LLM inference operations
- GPU-2 – Vector database operations and similarity search
- GPU-3 – Nvidia software reranking processes and relevancy scoring
- GPU-4 – Shared resources for embedding generation and auxiliary functions
This scheme provides optimized throughput without resource contention in the pipeline. An internal Cloudian HyperScale AIDP demo ingested a 1,000-page Cloudian admin guide in around five minutes, making it available for use by LLMs and agents. Existing Cloudian customers can acquire a HyperScale AIDP system, built around their HyperStore, and make the unstructured data stored within it available for AI LLM/agent RAG inferencing without the data leaving their HyperStore deployment. You can view a HyperScale AIDP demo here.
