SAN JOSE, California – Nvidia unveiled BlueField-4 STX, a modular reference architecture designed to enable enterprises, cloud, and AI providers to deploy accelerated storage infrastructure.

Unveiled at its flagship GTC event, the STX architecture marks yet another of Nvidia’s agentic-related offerings. It’s designed to provide the high-performance storage capabilities necessary to power AI agents and overcome the slowdown in AI inference caused by traditional data center storage as context grows.

“Agentic AI is redefining what software can do – and the computing infrastructure behind it must be reinvented to keep pace. AI systems that reason across massive context and continuously learn require a new class of storage," Nvidia CEO Jensen Huang explained. “STX reinvents the storage stack, providing a modular foundation for AI-native infrastructure that keeps AI factories operating at peak performance.”

Nvidia claims the architecture enables storage systems built atop it to provide four-times higher energy efficiency compared to traditional systems, while being capable of ingesting two-times more pages per second for enterprise AI data. The press release outlining STX further claims it provides up to five-times tokens-per-second compared with undefined traditional storage solutions.

The rack-scale implementation, meanwhile, features Nvidia’s context memory storage platform, or CMX. Unveiled back in January, the offering pits its BlueField-4 data processing unit as the effective backbone to extend GPU memory capacity while enabling high-speed sharing of context across clusters.

Some of the biggest names in storage have signed up as partners to build AI infrastructure atop STX, including Vast Data, Weka, Hitachi Vantara, and the recently rebranded Everpure (formerly Pure Storage).

Manufacturing partners pledged to STX-based systems include AIC, Supermicro, and Quanta Cloud Technology (QCT).

STX-based platforms will be available from partners in the second half of 2026.