Who makes drives for AI data centers

This is the drive itself — the sealed NVMe unit that plugs into a compute tray or a storage node, as opposed to the NAND inside it or the controller that runs it. In an AI cluster it is the fast local tier: the place training data is staged so the GPUs are not waiting on a network filesystem, and the place checkpoints land during a long run, where write bandwidth and consistency matter more than raw capacity. Form factor is the part of this that has actually changed. Dense compute trays leave very little room and a lot of heat, which has pushed the industry toward the newer ruler and E-series formats designed for airflow and serviceability rather than the legacy drive shapes inherited from disk. A tray built around one of those formats is committed to it, so form factor is a design-in decision made years before the drives are bought. Demand here follows cluster count and checkpoint discipline rather than model size directly. What to look for is who actually makes the NAND behind a given drive — several of the names on this part are integrated memory makers selling their own flash, and several are module and drive builders buying it, which are very different businesses with the same product on the shelf. Endurance rating and sustained write performance separate an AI-tier drive from a general enterprise one.