NVIDIA launches BlueField-4 to power the operating system of Artificial Intelligence factories

NVIDIA unveiled the BlueField-4 data processing unit to accelerate gigascale Artificial Intelligence infrastructure with 800Gb/s throughput and expanded compute. The platform brings software-defined acceleration for storage, networking and security to support trillion-token workloads.

NVIDIA announced the BlueField-4 data processing unit at GTC Washington, D.C., positioning it as the core operating system for next-generation Artificial Intelligence factories. BlueField-4 is part of the broader BlueField platform and is designed to accelerate gigascale Artificial Intelligence infrastructure, delivering high-performance inference, software-defined acceleration across storage, networking and security, and support for 800Gb/s throughput to meet rising demand for trillion-token workloads.

Technically, BlueField-4 pairs an NVIDIA Grace CPU with NVIDIA ConnectX-9 networking to provide what NVIDIA describes as six times the compute power and support for Artificial Intelligence factories up to four times larger than those possible with BlueField-3. The platform includes native support for NVIDIA DOCA microservices, enabling containerized services that secure, scale and simplify deployment and operations. BlueField-4 also supports a multiservice architecture with service function chaining, multi-tenant networking, rapid data access and cloud elasticity. ConnectX-9 SuperNICs are highlighted for ultralow-latency networking, optimized RoCE performance and predictable networking for demanding workloads when used with NVIDIA Spectrum-X Ethernet.

Security is anchored by the NVIDIA BlueField Advanced Secure Trusted Resource Architecture, which enables secure bare-metal compute instances with zero-trust tenant isolation and full software-defined infrastructure control. NVIDIA positions BlueField-4 as delivering consistent performance and security across its infrastructure portfolio, including NVIDIA RTX PRO Servers, HGX, DGX, GB200 and GB300 systems, as well as data-center designs like the Enterprise Artificial Intelligence Factory validated design, AI Data Platform, DGX SuperPOD and cloud partner reference architectures. A broad ecosystem of server and storage vendors, cybersecurity companies, cloud and Artificial Intelligence providers, platform software vendors and global systems integrators are planning BlueField-4 adoption. NVIDIA expects BlueField-4 to be available in early availability as part of Vera Rubin platforms in 2026.

58

Impact Score

Study finds widespread weaknesses in autonomous agents

A multi-institution study found that autonomous agents across several sectors are highly exposed to tool-chaining, goal drift, and memory poisoning attacks. The findings suggest agentic systems face broader and deeper security risks than stateless large language models.

Federal safety net unprepared for Artificial Intelligence job losses

Economists are warning that the federal system designed to support displaced workers is not equipped for a wave of job losses tied to Artificial Intelligence. Existing unemployment benefits and retraining programs are widely seen as too limited to manage broad disruption.

Chrome downloads Gemini Nano model locally without clear consent

Google Chrome is reported to download a 4 GB Gemini Nano model onto some PCs automatically when certain Artificial Intelligence features are active. The process happens without clear notice in browser settings and can repeat after the model is deleted.

Contact Us

Got questions? Use the form to contact us.

Contact Form

Clicking next sends a verification code to your email. After verifying, you can enter your message.