Supermicro announced a KV cache storage solution with IBM and NVIDIA on November 5, 2026, designed to move reusable AI inference context into shared storage beyond GPU memory. The companies report up to 56 times faster time to first token on a 130,000-token prompt in joint testing. The offering combines Supermicro Petascale all-flash storage servers using AMD EPYC 9005 processors, IBM Storage Scale software, NVIDIA’s Dynamo inference framework and Spectrum-X Ethernet networking. It is aimed at…