Arrcus and Canadian telecommunications provider TELUS have commenced a proof-of-concept to deploy a distributed AI inference network across Canada. The project uses the Arrcus Inference Network Fabric (AINF) to provide low-latency AI processing for mission-critical applications, including public safety and emergency response.

The initiative focuses on “sovereign AI,” ensuring that sensitive data and AI workloads remain within national borders. The AINF system functions as a control plane that evaluates business policies, such as latency targets and data sovereignty boundaries, to route traffic to the most efficient node. Technical data suggests the system can achieve a 60% reduction in “Time to First Token” and a 40% reduction in end-to-end latency.

The deployment integrates with NVIDIA BlueField-3 DPUs for line-rate encryption and NVIDIA Spectrum-4 Ethernet switches. This architecture allows for real-time video analytics and predictive threat detection at the network edge. It also supports SRv6 and Mobile User Plane (MUP) transport for programmable traffic management across 5G and wireline infrastructure.

“Public safety and mission-critical services demand AI that is fast, reliable and sovereign by design,” said Tim Fell, Vice-President of Wireline Technology & Services at TELUS. He noted the fabric provides the foundation to deliver AI inferencing at scale with necessary security and predictability.

The system is designed to be vendor-agnostic, supporting various hardware platforms to avoid lock-in while optimising performance for government and enterprise clients.