NVIDIA Unveils BlueField-4 STX Storage Architecture, Gaining Widespread Industry Adoption
NVIDIA has introduced its BlueField-4® STX modular reference architecture, designed to enhance storage infrastructure for enterprises and AI providers. This architecture addresses the growing demands of agentic AI, which relies on context memory for efficient data handling.
NVIDIA BlueField-4 STX Storage Architecture Overview
The STX architecture promises significant improvements, including:
- Up to 5 times token throughput
- Up to 4 times higher energy efficiency
- 2 times faster data ingestion
These advancements position NVIDIA’s solution at the forefront of storage technologies suited for AI applications.
Key Features of BlueField-4 STX
The NVIDIA BlueField-4 STX leverages several cutting-edge technologies to optimize performance:
- Accelerated by the NVIDIA Vera Rubin platform
- Storage-optimized NVIDIA BlueField-4 processor
- NVIDIA ConnectX®-9 SuperNIC for fast networking
- NVIDIA Spectrum-X
Ethernet networking - NVIDIA DOCA and AI Enterprise software integration
These technologies work together to create a responsive storage solution that meets the needs of agentic AI systems, allowing for real-time data access.
Adoption of BlueField-4 STX in the Industry
Leading storage providers and manufacturing partners are rapidly adopting the BlueField-4 STX architecture to build next-generation AI infrastructure. Early adopters include:
- CoreWeave
- Crusoe
- IREN
- Lambda
- Mistral AI
- Nebius
- Oracle Cloud Infrastructure (OCI)
- Vultr
Manufacturing partners engaged in STX system development include:
- AIC
- Supermicro
- Quanta Cloud Technology (QCT)
Impact on AI and Data Centers
Traditional data centers often lack the responsiveness required by agentic AI technologies. The BlueField-4 STX architecture provides a solution. It allows storage infrastructure to remain close and accessible, ensuring high throughput in AI tasks.
Jensen Huang, NVIDIA’s founder and CEO, emphasized the need for innovative storage solutions to support the evolving capabilities of AI. He stated, “AI systems that reason across massive context and continuously learn require a new class of storage.”
As the industry progresses, STX-based platforms are expected to be available from partners in the latter half of this year, marking a significant step forward in AI storage technology.
