Combined solution will steer traffic intelligently and securely for faster application response, lower latency, improved power efficiency and reduced cost per inference

SAN JOSE, Calif.: Arrcus, the leader in distributed networking infrastructure today announced at NVIDIA GTC integration between the Arrcus Inference Network Fabric (AINF) and NVIDIA AI infrastructure, including the NVIDIA Dynamo framework, NVIDIA BlueField-3 DPUs, NVIDIA Spectrum-X Ethernet networking and GPU platforms.

AINF powered by NVIDIA creates a secure, intelligent and policy-aware inference fabric spanning edge, data center and cloud environments. With this combined solution, operators like Lightstorm can deliver next-generation Physical and Agentic AI applications with lower latency, improved GPU utilization, enhanced power efficiency and reduced cost per inference.

“AI is entering its inference era, where networking becomes the control plane for performance and economics,” said Shekar Ayyar, Chairman and CEO of Arrcus. “By integrating AINF with NVIDIA AI technologies, we are enabling operators and enterprises to intelligently route inference traffic, maximize GPU utilization and deliver real-time AI services at global scale.”

As AI shifts from centralized training to globally distributed inference, infrastructure demands are changing rapidly. Agentic AI workflows increasingly chain dozens of inference calls across multiple models and tools to complete tasks. These workflows require intelligent model resolution, priority classification and policy enforcement at the moment a request enters the network. Real-time AI applications, ranging from robotics and autonomous systems to video analytics and agentic workflows require:

• Ultra-low latency
• Sovereign, geo-aware routing
• Intelligent model selection
• Efficient GPU utilization
• Secure, multi-site connectivity

“AI inferencing at scale across Asia‑Pacific demands reliable, low‑latency connectivity across vast WAN distances. Lightstorm is enabling hyperscalers, neoclouds and enterprises with the network foundation required for this shift, and by leveraging Arrcus’ AINF solution powered by NVIDIA, we’re excited to make real‑time, large‑scale inferencing into a deployable reality in the region,” said Amajit Gupta, Group CEO & MD, Lightstorm.

Traditional networking and load balancing approaches lack the policy awareness and dynamic routing required to support such requirements at global scale. With today’s announcement, Arrcus introduces tight integration between AINF and NVIDIA AI technologies to create a seamless inferencing fabric from edge to data center to cloud addressing the requirements above.

AINF acts as the central conductor for agentic AI, leveraging intelligent LLM classifiers running on NVIDIA AI infrastructure to determine the best-fit model and route requests to the optimal site and network path in real time.

This enables:

• Lightweight, specialized models at the edge
• Centralized execution for complex reasoning workloads
• Differentiated routing for latency-sensitive applications like voice, video and gaming
• Network slicing and traffic prioritization for service-level guarantees

AINF and NVIDIA Dynamo

AINF with NVIDIA Dynamo serves as the global routing layer for distributed inference, intelligently steering requests to the optimal site based on model availability, service tier, geofencing policies, site capacity and real-time network health. AINF works alongside the NVIDIA Dynamo framework, which handles site-local LLM load balancing, while the AINF router manages global inference routing across distributed environments.

• NVIDIA Dynamo provides GPU telemetry such as queue depth, KV-cache pressure and replica health
• AINF ingests these signals via its Site Agent into the AINF control plane
• AINF computes model-aware routing decisions across sites

AINF determines which site should serve an inference request, while NVIDIA Dynamo determines which replica within that site responds, ensuring optimal utilization of compute resources across large-scale deployments.

AINF and NVIDIA BlueField and Spectrum

AINF also integrates with NVIDIA BlueField-3 DPUs to secure inference traffic across locations. NVIDIA BlueField-3 enables line-rate encryption at up to 400 Gb/s with zero CPU overhead, ensuring sovereign, multi-site inference traffic remains encrypted without impacting performance. Combined with NVIDIA Spectrum-4 Ethernet switches and GPU platforms, the Arrcus solution delivers an end-to-end inference fabric engineered for performance, efficiency and security.

Arrcus will showcase the Arrcus Inference Network Fabric and its integration with NVIDIA AI-Grid at NVIDIA GTC, March 16–19, at booth #3019. To learn more or schedule time with the Arrcus team, visit the Arrcus GTC meeting page.

Additional Resources

• Arrcus introduces AI-policy aware Arrcus Inference Network Fabric
• Arrcus Harnesses NVIDIA BlueField-4 to Power Gigascale AI Factories with Superior Network Performance and Efficiency
• Arrcus delivers unprecedented cloud-scale networking and operational simplicity with NVIDIA DOCA Platform Framework
• Arrcus Powered by NVIDIA BlueField DPUs Enables Scalable and Efficient Networking for Enterprise AI

About Arrcus

Arrcus is a leading provider of networking software solutions that enable organizations to achieve unmatched scalability, performance, and reliability across their infrastructure. The company is transforming the industry with disaggregated solutions that deliver innovative, agile, and cost-effective networking, empowering businesses to move beyond traditional, monolithic systems and adopt a more flexible, efficient, and scalable approach to modern connectivity.

Arrcus’ team includes world-class technologists with exceptional track records in delivering industry-leading networking products, alongside recognized specialists, operating executives, strategic partners, and top-tier venture capital investors. The company is headquartered in San Jose, California. For more information, visit the official website or follow Arrcus on LinkedIn and Twitter/X.

Fonte: Business Wire