Using NVIDIA Aerial CUDA-Accelerated RAN on Red Hat OpenShift to Accelerate Development of AI-Native 5G and 6G RAN Solutions

Why It Matters

It gives telcos a scalable, open‑source platform to accelerate AI‑RAN innovation, cutting hardware cycles and unlocking new edge‑service revenue streams.

Key Takeaways

• Open‑source Aerial stack runs on Red Hat OpenShift single‑node edge
• GPU‑accelerated PHY achieves target throughput with zero errors
• Zero‑touch GitOps provisioning simplifies far‑edge deployment
• Supports ARM and NVIDIA GPUs, including GH200 Grace Hopper
• Enables AI‑native 5G/6G services on a hybrid‑cloud platform

Pulse Analysis

The race to AI‑native 5G and the forthcoming 6G era is reshaping telecom infrastructure. Traditional radio access networks, once fixed hardware bundles, are now being re‑imagined as software‑defined, cloud‑native systems that can ingest machine‑learning models at scale. By open‑sourcing its Aerial CUDA‑accelerated RAN stack, NVIDIA gives operators and vendors a transparent code base that can be audited, extended, and integrated with any Kubernetes‑compatible platform. Coupled with Red Hat OpenShift, the stack offers a unified development environment that bridges high‑performance GPU compute with the orchestration agility required for edge deployments.

In a recent proof‑of‑concept, Red Hat deployed the Aerial stack on a single‑node OpenShift cluster powered by an NVIDIA GH200 Grace Hopper Superchip and a ConnectX‑7 NIC. The configuration leveraged zero‑touch provisioning through GitOps, delivering a fully automated bare‑metal rollout at the far edge. Performance metrics showed the cuPHY layer meeting its 1,544 Mbps downlink and 197 Mbps uplink targets without packet loss, confirming that GPU‑accelerated PHY can satisfy real‑time RAN timing constraints. The demonstration validates the feasibility of running AI‑intensive workloads alongside core telecom functions on the same hardware.

The collaboration signals a shift toward open, multi‑vendor ecosystems where telcos can iterate faster and lower capital expenditures. By abstracting RAN functions into containers, service providers can push updates via software releases rather than costly hardware swaps, accelerating time‑to‑market for new AI‑driven services such as predictive beamforming or edge analytics. Moreover, the compatibility with ARM and NVIDIA GPUs ensures that future hardware generations can be adopted without re‑architecting the stack. As operators chase higher spectrum efficiency and new revenue streams, AI‑RAN on OpenShift positions them to capitalize on the next wave of connected intelligence.