In this panel from the Supermicro Open Storage Summit 2025, theCUBE’s Rob Strechay sits down with Harish Arora (NVIDIA), Tony Asaro (Hammerspace), Peter Sjoberg (Cloudian), Ace Stryker (Solidigm) and Tian Lee (Supermicro) to explore how storage innovation is powering inference at scale in the age of generative AI. The discussion dives into the shifting role of AI factories, where inference joins training and data prep as a core workload. Arora highlights NVIDIA’s NeMo Retriever and Dynamo frameworks for scaling inference pipelines, while Asaro explains how Hammerspace eliminates data gravity with its AI Anywhere data platform. Sjoberg details Cloudian’s breakthrough with GPUDirect for object storage, accelerating unstructured data access into GPUs, and Stryker outlines Solidigm’s SSD strategy for balancing high-performance and high-density workloads. Lee emphasizes how Supermicro brings these innovations together into complete, turnkey infrastructure solutions. Key themes include balancing performance and cost in real-world inference pipelines, advancing disaggregated inference architectures, and building validated GPU-optimized platforms that scale across edge, cloud and data center environments. The conversation also touches on how enterprises are preparing for the rise of agentic AI by integrating storage, networking and compute in tightly aligned ecosystems.