Nvidia Debuts STX Architecture, Boosting AI Throughput by 5x
On March 16, Nvidia announced its BlueField-4 STX reference architecture, a new design intended to reinvent the storage stack for artificial intelligence. The modular architecture enables cloud and AI providers to deploy accelerated storage infrastructure that can handle the massive context memory required by sophisticated AI agents. The company states the new system can deliver up to five times the token throughput and four times the energy efficiency of traditional CPU-based storage, directly addressing performance bottlenecks that can slow AI inference and underutilize expensive GPUs.
Traditional data centers struggle with the responsiveness needed for agentic AI, which requires real-time access to large, coherent datasets. The STX architecture keeps this data close and accessible at scale, allowing AI systems to operate with higher speed and responsiveness.
Agentic AI is redefining what software can do — and the computing infrastructure behind it must be reinvented to keep pace. NVIDIA STX reinvents the storage stack, providing a modular foundation for AI-native infrastructure that keeps AI factories operating at peak performance.
— Jensen Huang, founder and CEO of NVIDIA.
Broad Ecosystem of Tech Giants Adopts New Standard
The STX architecture launched with significant support from across the technology industry. Major AI labs and cloud service providers, including CoreWeave, Crusoe, Oracle Cloud Infrastructure (OCI), Mistral AI, and Vultr, have already committed to adopting the platform for their context memory storage needs. This broad initial adoption signals strong market confidence in the new standard's ability to solve critical AI infrastructure challenges.
On the hardware side, a formidable coalition of storage providers and manufacturers is building systems based on the new reference designs. Partners include industry leaders like Dell Technologies, Hewlett Packard Enterprise, IBM, NetApp, Supermicro, and VAST Data. Nvidia confirmed that platforms based on the STX architecture will become available from these partners in the second half of this year, providing a clear timeline for enterprise adoption.
STX Creates New Demand for Data Center Hardware
The introduction of the STX architecture is poised to trigger a new hardware upgrade cycle within data centers. The platform's emphasis on high-density, accelerated storage necessitates new server designs and components. Underscoring this trend, manufacturing partner Compal has already showcased a high-density AI server, the SG231-2-L1, built on the related NVIDIA HGX Rubin platform, which integrates the BlueField-4 DPU. This demonstrates that hardware manufacturers are prepared for the increased power and density requirements of next-generation AI.
This architectural shift also drives demand for the underlying components that enable this level of performance. The need for larger, faster context memory is expected to directly increase demand for high-speed NAND storage. The move away from general-purpose storage toward specialized, high-performance AI memory creates a significant opportunity for component suppliers positioned to serve the evolving data center market.
