WD: 8-second AI video leaves 500MB back-end trace

A distinction between training and inference matters for how AI budgets are built. Training is finite: a run starts, burns through compute, and ends, so it can be budgeted and retired. Inference runs continuously, and every request leaves data behind. WD's measurement of an 8-second video makes that abstract point concrete: the user sees a 3.97MB 720p file, while the back end retains more than 500MB, a 130x multiplier. At enterprise scale, with longer or higher-resolution video, the back-end footprint can grow into the hundreds of gigabytes per clip. The piece also notes that retained data is taking on a new role, as archives that once sat untouched become inputs for retrieval-augmented generation and future inference. That shifts the planning question from raw capacity to how dynamically the stored data will be accessed.