This is a massive storage deployment. Given the I/O demands of LLM training, especially for checkpointing, moving to this scale of NVMe flash makes sense compared to traditional disk arrays.