The useful question is what changes for users, developers or buyers, and whether the announcement stays industry context or becomes something people can actually use.

AI workloads now require datasets and context windows that exceed system memory limits, pushing storage systems to handle thousands of concurrent operations securely and efficiently. NVIDIA highlighted benchmarks showing its Vera CPU, part of the Vera BlueField-4 STX, delivers up to 3.21x higher throughput than x86 CPUs in compression and encryption pipelines, enabling storage platforms to process AI data more effectively with less compute overhead.

At the Future of Memory and Storage conference, NVIDIA announced it is open sourcing the cuFile APIs and its vertical storage software stack, allowing GPUs to read from and write to storage directly. This approach leverages GPU threads and high-bandwidth memory to reduce data access times to microseconds, aligning with Linux security best practices and supporting AI-driven cybersecurity measures.

The company also introduced Storage-Next, an initiative with over 40 storage and flash vendors, including DDN, KIOXIA, and Micron, to standardize GPU-driven storage behavior and develop open industry standards. NVIDIA’s SCADA framework enables massively parallel GPUs to pull only necessary data directly from storage into high-speed memory, reducing bottlenecks in AI-native platforms like DDN’s Infinia.

NVIDIA emphasized that AI success depends on productive infrastructure use rather than sheer ownership, with its AI-native storage solutions—such as Vera BlueField-4 STX and CMX Context Memory Storage—designed to enforce continuous policy enforcement and support long-context, multi-turn AI inference. The company’s work includes the unified DOCA security stack to ensure safe, direct storage access without compromising system security.