A new wave of networking technology designed for complex artificial intelligence systems was unveiled at a recent developer conference. The announcements highlight a strategic push to create integrated hardware ecosystems that form the backbone of next-generation AI data centers.
A central introduction is the ConnectX-9 SuperNIC, a network interface card engineered specifically for high-demand AI operations. This component is reported to achieve a data transfer rate of 1.6 terabits per second and incorporates support for the latest PCIe generation. Its architecture is tailored for managing the substantial computational loads required for both training advanced AI models and executing large-scale inference tasks. Integrated hardware-level security features and programmable input/output capabilities provide a foundation for customizable infrastructure.
Simultaneously, the next iteration of the BlueField data processing unit (DPU) was revealed. This processor is positioned as a core component for managing AI data center operations. Performance specifications indicate a substantial generational leap, with throughput said to be twice that of its predecessor and computational power increased by a factor of six. This performance enhancement is attributed to an architectural shift to a more powerful, enterprise-grade CPU design.
The new DPU integrates the capabilities of the ConnectX-9 SuperNIC, aiming to consolidate and accelerate storage, networking, and security functions across AI infrastructure. By handling these tasks, the processor is designed to free up resources on the main server CPUs, thereby increasing overall system efficiency for AI workloads.
This hardware is slated for integration into the company’s upcoming rack-scale server systems and will also be available to third-party server manufacturers. The combination of the high-throughput network interface and the powerful data processor is expected to establish a new benchmark for performance and efficiency in AI-driven computing environments, enabling more scalable and powerful AI factory networks.
Nvidia Elevates AI Infrastructure with Advanced Networking Hardware
A new wave of networking technology designed for complex artificial intelligence systems was unveiled at a recent developer conference. The announcements highlight a strategic push to create integrated hardware ecosystems that form the backbone of next-generation AI data centers.
A central introduction is the ConnectX-9 SuperNIC, a network interface card engineered specifically for high-demand AI operations. This component is reported to achieve a data transfer rate of 1.6 terabits per second and incorporates support for the latest PCIe generation. Its architecture is tailored for managing the substantial computational loads required for both training advanced AI models and executing large-scale inference tasks. Integrated hardware-level security features and programmable input/output capabilities provide a foundation for customizable infrastructure.
Simultaneously, the next iteration of the BlueField data processing unit (DPU) was revealed. This processor is positioned as a core component for managing AI data center operations. Performance specifications indicate a substantial generational leap, with throughput said to be twice that of its predecessor and computational power increased by a factor of six. This performance enhancement is attributed to an architectural shift to a more powerful, enterprise-grade CPU design.
The new DPU integrates the capabilities of the ConnectX-9 SuperNIC, aiming to consolidate and accelerate storage, networking, and security functions across AI infrastructure. By handling these tasks, the processor is designed to free up resources on the main server CPUs, thereby increasing overall system efficiency for AI workloads.
This hardware is slated for integration into the company’s upcoming rack-scale server systems and will also be available to third-party server manufacturers. The combination of the high-throughput network interface and the powerful data processor is expected to establish a new benchmark for performance and efficiency in AI-driven computing environments, enabling more scalable and powerful AI factory networks.
Archives
Categories
Archives
Nvidia Elevates AI Infrastructure with Advanced Networking Hardware
November 19, 2025IBM Partners with Groq to Accelerate AI Inference Across Enterprise Systems
October 25, 2025Categories
Meta