Arrcus & TELUS test sovereign AI network in Canada

Arrcus has begun a proof of concept with TELUS for its Arrcus Inference Network Fabric.

The project will explore a national-scale network foundation for sovereign AI inferencing in Canada.

TELUS is assessing whether the system can support distributed AI workloads while keeping sensitive data and processing within Canadian borders. The work is aimed at uses including public safety, emergency response, enterprise services and government applications.

The trial reflects a broader shift in AI infrastructure from centralised model training to distributed inference, in which models are deployed closer to users and data sources. That shift places more emphasis on network design because requests may need to move across multiple sites while meeting strict requirements for latency, security and data location.

For telecoms operators and public sector customers, sovereignty has become a more prominent issue as governments and regulated industries seek to maintain control over data and digital infrastructure. TELUS is examining a network approach that can direct AI traffic according to operator policies on geography, performance and security.

How it works

Arrcus said its Inference Network Fabric acts as a control layer for distributed inference traffic. It is designed to evaluate policies such as latency targets, model choice, power constraints and data residency rules, then route requests to an appropriate node or cache across a distributed network.

The system is intended for use across edge, data centre and cloud environments. In the TELUS proof of concept, it is being combined with NVIDIA BlueField-3 data processing units for encryption and NVIDIA Spectrum-4 Ethernet switches for transport across multiple sites.

Arrcus also said the fabric works with the NVIDIA Dynamo framework for site-local large language model load balancing, while the Arrcus platform manages broader inference routing across the TELUS network. The release also cited integration with vLLM, SGLang, Triton and Kubernetes.

The companies described several intended functions for the trial, including geo-aware routing to enforce Canadian data residency requirements, traffic prioritisation for critical workloads, and support for Segment Routing over IPv6 and Mobile User Plane transport across 5G and wireline infrastructure.

Arrcus said the platform supports a range of merchant silicon and hardware platforms, which could help operators avoid dependence on a single supplier in mixed infrastructure environments.

Performance claims

Research cited by Arrcus said intelligent, policy-aware networking can reduce time to first token by more than 60%, improve throughput by 15%, cut end-to-end latency by 40% and lower cost per inference by as much as 30%. Arrcus attributed those figures to sources including Anyscale Ray Serve, Red Hat vLLM Semantic Routing and the AWS Machine Learning Blog.

No financial terms for the proof of concept were disclosed. The companies also did not say how long the evaluation would run or when TELUS might decide whether to move from testing to deployment.

Still, the project points to a growing commercial opportunity for network suppliers seeking a role in AI infrastructure beyond data centre switching. As AI applications move into public services and regulated sectors, suppliers are trying to show that the network itself can become part of the governance and routing logic for inference workloads.

That is especially relevant for applications where response times and security controls have operational consequences, such as emergency coordination, video analysis and threat detection. Operators are under pressure to support those services without sending sensitive data outside national boundaries or introducing delays through distant compute locations.

Tim Fell, Vice-President, Wireline Technology & Services, TELUS, outlined the rationale for the project.

"Public safety and mission-critical services demand AI that is fast, reliable and sovereign by design," said Tim Fell, Vice-President, Wireline Technology & Services, TELUS. "With AINF, Arrcus gives us the intelligent, policy-aware networking foundation to deliver AI inferencing at speed and scale across our network, with the data sovereignty, security and predictability that our public safety partners, government customers and enterprise clients require."

Arrcus Chairman and Chief Executive Officer Shekar Ayyar said the trial reflected a wider industry shift in how AI services are delivered.

"The era of centralized AI is giving way to distributed intelligence, and the network is becoming the control plane for performance, sovereignty and economics," said Shekar Ayyar, Chairman and Chief Executive Officer of Arrcus. "TELUS is building the kind of trusted, national-scale AI infrastructure that Canada needs, and AINF is the networking fabric that can make this possible. AINF's intelligent routing, real-time policy enforcement and hardware-accelerated security enable TELUS to deliver mission-critical AI to the right place, with the right guarantees at the right time."