VAST as AI Infrastructure Control Plane
VAST Data
VAST is trying to turn storage from a hardware budget line into the control plane for AI infrastructure. The important shift is that data no longer has to be copied out to separate warehouse, catalog, and compute systems before teams can query it or run preprocessing jobs. VAST now bundles file and object storage, metadata indexing, SQL style access, and event driven execution so one cluster can ingest data, organize it, and feed GPUs from the same software layer.
-
This changes the buyer and the deal size. Instead of replacing only a storage array, VAST can replace storage, parts of the data lake stack, and some ETL and preprocessing infrastructure. That is why average new customer commitments have crossed $1.2 million, with some customers above $100 million in total commitments.
-
The NVIDIA tie in matters because BlueField DPUs let VAST run data services inside or next to GPU servers, not just on separate storage controllers. In practice, that means a GPU cluster can pull training data, access metadata, and increasingly run data path functions with less CPU overhead and fewer bottlenecks.
-
That puts VAST in a different lane from WEKA, Pure Storage, and Dell, which are still primarily selling faster storage, and closer to Databricks and Snowflake, which sell the place where data gets organized and worked on. The difference is that VAST brings that stack on premises and into dedicated GPU clouds like CoreWeave.
The next step is for VAST to become the default data substrate under private and sovereign AI factories. If it keeps proving that database, pipeline, and inference data services can run in the same footprint as storage, it moves from being an infrastructure component to being the operating layer that GPU clouds and large enterprises standardize on.