How AI Data Platforms Are Shaping the Future of Enterprise Storage
NVIDIA's Jacob Liberman returns to discuss the AI Data Platform, a new reference design for GPU-accelerated enterprise storage. The core idea is to bring the GPU to where the data lives rather than copying data to a GPU cluster, addressing what NVIDIA calls 'data gravity' — the challenge that enterprise data is large, growing, and expensive to move. By embedding GPUs directly into traditional storage systems, enterprises can continuously transform unstructured data into 'AI-ready' representations (chunks, embeddings, vector indices) as a background operation, without copying or moving the source-of-truth files. The design solves several enterprise pain points around agentic AI adoption. Traditional RAG pipelines require multiple copies of data, expanding attack surfaces and risking stale content or outdated permissions as data drifts. With in-place GPU processing, security and access controls are preserved alongside the source documents, and changes propagate immediately. Liberman also notes that spare GPU cycles in storage can be used to run AI agents directly within the storage system for tasks like automated classification, governance, and telemetry monitoring — an evolution he compares to letting agents 'work from home.' NVIDIA is delivering this through partner-based reference designs built on its AI Blueprints rather than direct products. Major storage vendors have already begun implementing the designs, often extending them in unexpected ways. Liberman frames the shift as a natural evolution of storage — parallel to how CPUs were once added to storage systems for compression, encryption, and search — and predicts that all storage will eventually be GPU-accelerated and that every AI factory will include an AI data platform to decouple compute and data scaling.
Original sources
Podcast NVIDIA AI Podcast
Episode:How AI Data Platforms Are Shaping the Future of Enterprise Storage - Ep. 281 ↗
AI Radar wrote this summary from the publisher’s notes with AI, not from a transcript. How it works.