NVIDIA unveiled new storage advancements at the FMS conference, emphasizing the growing demand for storage infrastructure driven by AI. Its Vera CPU delivers up to 3.21x higher throughput than x86 CPUs in compression and encryption pipelines, making storage an active part of accelerated computing.
SynthePulse Insight · AI deep readingMembers
Version 1 · 1 source
At FMS 2026, NVIDIA unveiled Vera CPU performance data, open-sourced the cuFile API, and introduced Storage-Next and SCADA frameworks, aiming to transform storage from a passive repository into an active data path to tackle the massive data challenges posed by AI agents.
AI agents consume massive amounts of data, and GPUs can directly issue storage requests, generating thousands of concurrent operations that put immense pressure on storage systems. In traditional storage architectures, critical data services such as encryption, compression, validation, and reconstruction can become bottlenecks, especially when thousands of agents access simultaneously.
NVIDIA points out that simply increasing storage capacity is insufficient; the key lies in efficient, secure storage architectures that enable AI factories to produce useful and accurate insights. Storage is no longer a passive place to store data but an active part of the data path.
This article primarily draws information from NVIDIA's official blog, which is vendor-published content. Performance data (such as the 3.21x figure) comes from NVIDIA's own benchmarks and has not been independently verified. The Storage-Next member list and open-source initiative details are subject to official announcements.
Primary report
Primary source