Metrum AI and Solidigm use SSD offload and DiskANN to cut memory costs while delivering efficient AI inference at scale, ...