Arrcus announced support for NVIDIA’s new BlueField-4 platform at GTC Washington, D.C., combining its ArcOS software with NVIDIA’s most advanced DPU to drive performance, efficiency, and security for gigascale AI factories. The collaboration targets next-generation AI data centers, where optimized fabrics are essential for large-scale training, inference, and cloud-scale operations.
BlueField-4 delivers a sixfold increase in compute power over its predecessor and up to 800 Gb/s of networking throughput. By integrating the NVIDIA Grace CPU with ConnectX-9 networking, the platform introduces a unified infrastructure acceleration framework designed for high-performance computing and AI workloads. Arrcus’ ArcOS runs natively on BlueField-4, enabling dynamic offload of network-intensive functions such as IPSec, NAT, routing, EVPN/VXLAN, and ACLs—freeing CPU resources for AI compute while maintaining secure, multi-tenant performance.
The combined solution provides a scalable architecture for AI data centers and emerging “NeoClouds” that demand composable, cloud-ready infrastructure. Through NVIDIA’s DOCA microservices, operators gain advanced networking, data acceleration, and cybersecurity features to power distributed AI deployments. The solution is also designed to open new revenue models, such as Inference-as-a-Service, leveraging the efficiency of the joint platform.
• BlueField-4 DPU offers 6× compute performance over BlueField-3 and 800 Gb/s network speed
• ArcOS enables hardware offload of routing, IPSec, NAT, and EVPN/VXLAN to boost AI fabric efficiency
• Integration of NVIDIA Grace CPU and ConnectX-9 networking for unified compute and networking acceleration
• DOCA-based microservices extend automation, observability, and secure connectivity at scale
• Supports new monetization opportunities for NeoCloud and AI-as-a-Service operators
“Together, Arrcus and NVIDIA are redefining the foundation of gigascale AI infrastructure,” said Shekar Ayyar, CEO of Arrcus. “By combining the flexibility of ArcOS with the acceleration of BlueField-4, customers can deploy a secure, high-performance fabric that scales seamlessly from core to edge.”
🌐 Analysis: This announcement reinforces Arrcus’ strategy to serve hyperscale and NeoCloud operators deploying GPU-dense AI clusters. By optimizing ArcOS for NVIDIA’s BlueField-4, Arrcus joins a growing ecosystem—including Red Hat, Dell, and Cisco—building composable AI infrastructures centered on NVIDIA’s DPUs and Grace architecture. The partnership positions Arrcus as a key enabler in the transition to high-performance, software-defined AI networks.