As AI Increases Demands on Memory, Storage Steps Up

At the Future of Memory and Storage (FMS) conference, NVIDIA announced new storage advancements to address the growing demands of AI, which require massive datasets and context windows that exceed system memory. The company highlighted that storage systems must efficiently encrypt, compress, verify and reconstruct data, which can become bottlenecks when thousands of AI agents access storage simultaneously. NVIDIA's Vera CPU, part of NVIDIA Vera BlueField-4 STX, delivers up to 3.21x higher throughput than an x86 CPU in a two-stage compression and encryption pipeline, allowing storage platforms to handle AI data floods more efficiently. NVIDIA also announced it is open sourcing its cuFile APIs, part of GPUDirect Storage, enabling GPUs to read and write to storage directly with microsecond access times. Additionally, NVIDIA introduced Storage-Next, an initiative with over 40 storage and flash vendors such as DDN, KIOXIA, and Micron, to develop interoperable standards for GPU-driven storage. The initiative includes SCADA (scaled, accelerated data access), a framework that lets massively parallel GPUs pull only necessary data directly into their high-speed memory, securely splitting the access process to maintain security. NVIDIA also mentioned NVIDIA CMX Context Memory Storage, an AI-native context tier for long-context, multi-turn, agentic AI inference built on NVIDIA STX.