Supermicro’s upcoming Open Storage Summit video series highlights a major shift in storage architecture for agentic AI. Running from Aug. 11 to Sept. 3, the seventh annual event will feature 12 sessions focused on enterprise AI, inference infrastructure, and data management. Industry veterans from Supermicro’s ecosystem—including AMD, DDN, Intel, Nutanix, and Solidigm—will discuss how companies are building the hardware needed to support these workloads.
Storage architecture is becoming a critical component in AI factories, particularly as the industry moves from chatbots to AI agents. Agents require context memory—the relevant information needed to understand and process tasks—which has led to the creation of new storage tiers dedicated to KV caches. Supermicro has developed systems based on Nvidia’s architecture to address this need, offering greater token throughput and improved efficiency.
When Nvidia introduced the BlueField-4 STX reference architecture at GTC in March, it included the CMX context memory storage platform. This system is optimized for storing and serving KV caches for large language models. Supermicro joined with storage partners like Vast Data and Nutanix to announce AI data platform solutions based on this architecture. Zeus Kerravala, principal analyst at ZK Research, noted that this move brings storage into the AI factory as an integrated component rather than just a storage-optimized hardware refresh.
Storage architecture is evolving to handle the specific demands of AI agents. As these systems perform longer, multi-step tasks, they generate organizational knowledge that has value across business lines. To preserve this memory, MinIO introduced AIStor Memory, an enterprise-controlled platform that retains agent memory and workspaces. This builds on the September 2025 launch of AIStor-powered pods on Supermicro systems, which provide preintegrated, ready-to-deploy object storage for rapid AI deployment.
With agentic AI workloads, the demand for infrastructure extends beyond GPUs to include CPUs. Sub-agents are reasoning and executing tasks across compute cycles, increasing the importance of CPU orchestration. Supermicro’s H15 server portfolio, powered by sixth-generation AMD EPYC 9006 Series processors, reflects this shift. The systems are optimized for next-generation GPUs and connected through AMD Pensando networking, with redesigns available for Hyper, CloudDC, and Grand Twin platforms.
MinIO’s AIStor Memory and other solutions aim to provide persistent memory that supports these complex workflows. By integrating storage directly into the AI stack, companies can better manage the data generated during reasoning and execution. This approach aligns with the broader trend of building infrastructure that supports active reasoning rather than long-term archiving. Google reorganizes AI leadership as key executives depart, showing how the industry adapts to these new demands.