.jpg)
The networking demands of gigascale AI factories require a fundamental shift in infrastructure to eliminate bottlenecks and maximize efficiency. Arrcus, a hyperscale networking software company, has announced support for the newly launched NVIDIA BlueField-4 platform. The combination of Arrcus ArcOS running natively on the BlueField-4 DPU is designed to deliver new levels of acceleration, security, and performance for AI training, inference, and cloud-scale workloads.
Arrcus ArcOS now runs natively on the new NVIDIA BlueField-4 DPU.
The collaboration aims to power high-performance networking for gigascale AI factories.
ArcOS offloads resource-intensive network functions like routing and security to the DPU.
This offload frees up CPU cycles for AI workloads, increasing overall efficiency.
BlueField-4 offers 6x the compute power of its predecessor and 800 Gb/s networking.
The solution enables new monetization models for service providers, like Inference-as-a-Service.
Arrcus ArcOS is engineered to run natively on NVIDIA BlueField-4, allowing customers to offload resource-intensive network functions directly to the DPU. This includes IPSec, NAT, Routing, EVPN/VXLAN, and Access Control Lists (ACLs). By moving these tasks off the main CPU, the solution increases network throughput and system-wide efficiency, ensuring that valuable CPU cycles are dedicated to AI compute workloads while maintaining secure, multi-tenant connectivity.
The NVIDIA BlueField-4 platform provides a significant leap in capability, introducing six times the compute power of the previous generation and supporting 800 Gb/s networking speeds. It integrates the NVIDIA Grace CPU and ConnectX-9 networking, creating a powerful foundation for advanced infrastructure acceleration and cybersecurity that is critical for modern AI data centers.
Together, ArcOS and BlueField-4 deliver accelerated performance across AI fabrics for both training and inference. They extend AI fabric visibility, automation, and secure connectivity, enabling composable, cloud-ready infrastructure. For service providers and "NeoClouds," this combination unlocks new monetization opportunities, such as Inference-as-a-Service, by providing a highly efficient and optimized network fabric purpose-built for gigascale AI operations.
This partnership addresses the critical networking challenges presented by the scale of modern AI. By pairing the software-defined flexibility of ArcOS with the raw acceleration power of BlueField-4, Arrcus and NVIDIA are providing a foundational element for the next generation of AI infrastructure—one that is secure, high-performance, and designed to scale effortlessly to meet the demands of future AI innovation.
Arrcus is a leading provider of networking software solutions that empower businesses to achieve unparalleled scalability, performance, and reliability in their infrastructure. Arrcus is disrupting the industry with disaggregated solutions that deliver innovative, agile, and cost-effective networking, allowing enterprises to break free from traditional, monolithic systems and embrace a more flexible, efficient, and scalable approach to modern networking. The Arrcus team consists of world-class technologists who have an unparalleled record in shipping industry-leading networking products, complemented by industry thought leaders, operating executives, strategic partners, and top-tier VCs. The company is headquartered in San Jose, Calif.
TechIntelPro empowers B2B technology decision-makers with the latest industry trends, deep insights, leadership perspectives and actionable intelligence to drive both organizational and professional milestones—and thrive on the tech pulse.