NVIDIA BlueField security and acceleration arrive on the Red Hat AI Factory with NVIDIA and Red Hat OpenShift

Agentic AI inference and 5G RAN workloads are pushing data center networking to its limits, yet up to 30% of host CPU power is wasted simply routing packets and enforcing security policies. When infrastructure services run directly on host CPUs, they create an "infrastructure tax" – consuming valuable CPU cycles that should be dedicated to core business applications, 5G core, radio access network (RAN) network functions, or resource-intensive AI models.

To help service providers and enterprises modernize without sacrificing profitability, Red Hat OpenShift now offers general availability support for NVIDIA BlueField—delivering line-rate acceleration, in-silicon security, and lower total cost of ownership across hybrid networks. By combining Red Hat OpenShift with NVIDIA DOCA Platform Framework (DPF) and BlueField DPUs, service providers and enterprises can offload, accelerate, and isolate infrastructure services directly on purpose-built hardware. The benefits include reclaimed CPU cycles, improved workload density, and hardware-enforced security boundaries.

From the service provider edge to AI factories

The integration of NVIDIA BlueField into OpenShift addresses critical infrastructure challenges across multiple environments. For telecommunications service providers, meeting strict Service Level Agreements (SLAs) requires deterministic latency and high-throughput packet handling. Offloading Open vSwitch and OVN-Kubernetes processing directly to the BlueField’s silicon achieves line-rate networking without consuming host CPU resources or forcing costly switch fabric reconfigurations. For power- and space-constrained edge sites, offloading infrastructure processing to BlueField maximizes compute density, lowers site operational costs and enforces tenant isolation in hardware.

NVIDIA BlueField also serves as a foundational component in the Red Hat AI Factory with NVIDIA.AI triggers massive volumes of North/South and East/West network traffic for model training, long-running agents, and real-time inferencing. By offloading data movement to specialized hardware, service providers eliminate network bottlenecks and ensure peak GPU pipeline utilization for enterprise AI services.

Offload, isolate, accelerate

Integrating NVIDIA BlueField into service provider Red Hat OpenShift environments delivers three key architectural advantages:

1. Infrastructure services offload: Moving OVN-Kubernetes network stack handling to BlueField’s silicon reclaims host CPU cores, returning raw compute capacity back to application workloads.
2. Security and isolation: Running the infrastructure control plane on the DPU physically separates it from the tenant workload domain. Even if a tenant application space is compromised, the underlying network control plane remains secure and isolated.
3. Performance acceleration: Purpose-built hardware-accelerated processing enables deterministic network performance and high-speed throughput up to 400Gb/s, bypassing host software bottlenecks.

The general availability of NVIDIA BlueField support on OpenShift represents a significant milestone in production-ready DPU deployment, transitioning from earlier technical previews through rigorous hardening, integration, and provisioning optimizations. OpenShift together with NVIDIA DOCA establishes a clean operational division between application compute and infrastructure services—reserving high-performance x86 hosts for application logic while dedicated Arm cores and hardware accelerators execute networking, security, and other infrastructure services. 

Get started today 

Ready to eliminate the infrastructure tax and accelerate your OpenShift workloads with NVIDIA BlueField? Review the NVIDIA and Red Hat telco AI checklist and learn more about the benefits of deploying NVIDIA BlueField on OpenShift.