NVIDIA Launches BlueField-4 DPU to Accelerate AI Factories and Data Center Performance
The NVIDIA BlueField-4 platform is expected to enter early availability in 2026 as part of the NVIDIA Vera Rubin platform, positioning itself as the foundational engine for the next era of AI factories and data-driven innovation.
At the NVIDIA GTC conference in Washington, D.C., NVIDIA introduced the BlueField-4 data processing unit (DPU), marking a major leap forward in AI infrastructure designed to meet the surging demand for trillion-token workloads and massive-scale computing.
BlueField-4 forms part of NVIDIA’s full-stack BlueField platform, purpose-built to accelerate gigascale AI data centers. The new DPU delivers up to 800 Gb/s of throughput, offering substantial computing performance gains and enabling AI factories up to four times larger than those supported by its predecessor, BlueField-3.
At its core, BlueField-4 integrates an NVIDIA Grace CPU and NVIDIA ConnectX-9 networking, providing six times more compute power and setting a new benchmark for high-performance, energy-efficient data processing. It delivers software-defined acceleration across AI data storage, networking, and security, transforming data centers into intelligent, secure, and fully automated AI infrastructures.
The platform also introduces a multiservice architecture supporting native service function chaining, allowing seamless integration and management of multiple network, storage, and security services within one unified environment. Enhanced by NVIDIA DOCA microservices, BlueField-4 enables scalable, containerized services that simplify AI deployment, operations, and orchestration while maintaining cloud elasticity and runtime security.
Security remains central to the platform’s design through NVIDIA’s Advanced Secure Trusted Resource Architecture, which ensures zero-trust tenant isolation and full software-defined control across virtualized and bare-metal environments.
BlueField-4 is built to deliver consistent performance across NVIDIA’s broader AI infrastructure portfolio, including RTX PRO Servers, HGX, DGX, GB200 and GB300 systems, as well as large-scale designs such as the NVIDIA Enterprise AI Factory and DGX SuperPOD. Supporting these systems, the ConnectX-9 SuperNICs provide ultralow-latency 800 Gb/s Ethernet networking to maximize efficiency and reliability in AI workloads.
The unveiling of BlueField-4 has been met with wide ecosystem support. Global server and storage leaders such as Cisco, DDN, Dell Technologies, HPE, IBM, Lenovo, Supermicro, VAST Data, and WEKA are integrating the technology into next-generation AI storage and compute platforms. Cybersecurity innovators including Armis, Check Point, F5, Palo Alto Networks, and Trend Micro are adopting BlueField architectures to deliver zero-trust and real-time threat protection.
Cloud and AI infrastructure providers such as Akamai, CoreWeave, Crusoe, Lambda, Oracle Cloud Infrastructure, Together.ai, and xAI are leveraging NVIDIA DOCA microservices to enhance networking, speed data movement, and improve AI runtime performance. Additionally, cloud-native software companies including Canonical, Red Hat, Nutanix, and SUSE plan to integrate BlueField-4 for automated, AI-ready cloud and data-center operations.
Global systems integrators Accenture, Deloitte, and World Wide Technology are collaborating with NVIDIA to deliver BlueField-powered infrastructure and cybersecurity services to enterprise and government clients worldwide.
The NVIDIA BlueField-4 platform is expected to enter early availability in 2026 as part of the NVIDIA Vera Rubin platform, positioning itself as the foundational engine for the next era of AI factories and data-driven innovation.
