NVIDIA Debuts BlueField-4 STX for Agentic AI, Claims Major Throughput and Efficiency Gains
At GTC, NVIDIA introduced the BlueField-4 STX platform targeting agentic AI workloads and memory-heavy models. The company claims the new design can provide up to 5x higher token throughput and 4x better energy efficiency versus current solutions, and noted that several major cloud providers have already committed to the hardware. NVIDIA framed the product as a way to move storage and data services closer to accelerators to reduce latency and improve sustained model performance.
If the performance and efficiency claims hold in production, BlueField-4 STX could lower operating costs and speed inference for large-scale AI deployments, making it attractive to cloud operators and enterprises running agentic systems. Real-world validation from signed cloud partners will be important for adoption and for assessing how much these gains translate into lower cost-per-token and reduced energy use in practice.