NVIDIA (NASDAQ: NVDA) announced the BlueField-4 STX storage architecture, designed for artificial intelligence workloads requiring long-context reasoning capabilities. The company made the announcement at its GTC conference.

The STX reference architecture provides up to 5x token throughput and up to 4x energy efficiency compared to traditional storage systems, with 2x faster data ingestion for enterprise AI data, according to the company's press release.

Early adopters planning to use STX for context memory storage include CoreWeave, Crusoe, IREN, Lambda, Mistral AI, Nebius, Oracle Cloud Infrastructure and Vultr.

The architecture incorporates the NVIDIA CMX context memory storage platform, which the company states expands GPU memory with a high-performance context layer. STX uses the BlueField-4 processor that combines the NVIDIA Vera CPU with ConnectX-9 SuperNIC, along with Spectrum-X Ethernet networking.

"Agentic AI is redefining what software can do — and the computing infrastructure behind it must be reinvented to keep pace," said Jensen Huang, NVIDIA's founder and CEO. "AI systems that reason across massive context and continuously learn require a new class of storage."

Storage and manufacturing partners developing infrastructure using the STX designs include AIC, Cloudian, DDN, Dell Technologies, Everpure, Hitachi Vantara, HPE, IBM, MinIO, NetApp, Nutanix, Supermicro, Quanta Cloud Technology, VAST Data and WEKA.

STX-based platforms will be available from partners in the second half of this year, according to the announcement.