VAST's unified AI data platform
Renen Hallak, CEO of VAST Data, on AI agents creating infinite storage demand
This is really a claim that AI infrastructure is converging into one shared data layer, not separate boxes for speed on one side and safety on the other. Old enterprise storage from Dell, NetApp, and Pure was built to keep business systems stable and compliant. HPC storage was built to feed very fast clusters. VAST is trying to combine both, so one system can keep huge GPU fleets busy while also handling replication, permissions, and production data management.
-
The concrete workflow difference is that AI teams need thousands of GPUs to read the same training sets, checkpoints, embeddings, and outputs at once. VAST positions its disaggregated shared architecture as a way to let any compute node hit one flash pool at very high parallelism, instead of splitting data across separate storage and analytics systems.
-
The comparison set breaks into two camps. WEKA and DDN are closer to the HPC world and optimize hard for throughput to GPU clusters. Pure, Dell, NetApp, and IBM are closer to the enterprise world and are easier fits for existing IT estates, but were designed around older file system assumptions and lower parallel access demands.
-
What makes the bet bigger than storage is that VAST is adding catalog, SQL, and compute on top of the same system. That lets a customer store raw files, search metadata, run queries, and process data without copying it into a separate warehouse or ETL stack, which expands the battle from storage budgets into database and analytics budgets.
The market is moving toward data platforms that look more like operating systems for AI factories. If VAST keeps winning large GPU cloud and enterprise deployments, the next step is not just replacing storage arrays, but becoming the default place where AI data is stored, indexed, queried, and served across training, inference, and agent workloads.