Details
- NVIDIA unveiled BlueField-4 STX, a modular reference architecture for accelerated storage systems designed to support agentic AI workloads requiring long-context reasoning and real-time data access.
- The architecture combines the NVIDIA Vera Rubin platform, storage-optimized BlueField-4 processor, ConnectX-9 SuperNIC, Spectrum-X Ethernet networking, DOCA, and AI Enterprise software to bridge GPU compute and data storage.
- STX delivers up to 5x token throughput, 4x energy efficiency gains, and 2x faster data ingestion compared to traditional CPU-based storage architectures, with the CMX context memory storage platform as its first rack-scale implementation.
- Early adopters include CoreWeave, Crusoe, IREN, Lambda, Mistral AI, Nebius, Oracle Cloud Infrastructure, and Vultr; storage partners include Cloudian, DDN, Dell, HPE, IBM, NetApp, Nutanix, VAST Data, and WEKA; manufacturing partners include Supermicro and Quanta Cloud Technology.
- STX-based platforms will ship in H2 2026, positioning NVIDIA to expand beyond GPU markets into complete AI infrastructure as enterprises scale autonomous agent deployments.
Impact
STX addresses a critical infrastructure gap for agentic AI, where traditional storage cannot sustain the real-time context access and throughput demands of autonomous agents. Broad industry adoption signals NVIDIA's strategic pivot toward end-to-end AI data infrastructure, potentially reshaping competitive dynamics in enterprise storage as agentic AI transitions from research to production deployment.
