Storage leaders target AI's unstructured data

The discussion at the Supermicro Open Storage Summit highlights a shift in how storage vendors view unstructured data. Previously, managing such data focused on archiving and backup, but now the goal is actively preparing it for AI consumption. This involves not just storing data, but also curating it, describing it, and automating its movement to where it is needed for model training and inference.

The companies present are approaching this from different angles. Supermicro provides high-density hardware platforms and Context Memory servers for KV cache tiers. Cloudian offers S3-native object storage with a strong emphasis on governance and security. Hammerspace focuses on creating a unified global namespace and data orchestration, including a new Model Context Protocol (MCP) layer to help AI systems understand distributed data. Seagate contributes hard drives for the later stages of the data lifecycle.

A key takeaway from the executives is the need to identify the value of data to determine the right infrastructure balance. Seagate's El-Batal advised to "value your data and don't go cheap on your infrastructure, while at the same time being efficient at it." The goal is to prepare data not only for training but also for compliance audits, debugging, and understanding AI outputs, making data preparation a critical step for effective AI deployments.