Arrcus PoC with TELUS Highlights AI Networking and Sovereignty

AI is raising the stakes for networking in service provider networks. A new deal between distributed cloud networking startup Arrcus and Canadian service provider TELUS demonstrates how networking may require more sophisticated distributed connectctivity, flexibility, and policy-driven features to keep up with AI demands.

Arrcus, a serial member of our Futuriom 50 annual list of hot startups, today announced a proof-of-concept (PoC) with TELUS. The goal is to accelerate secure AI infrastructure using the Arrcus Inference Network Fabric (AINF).

TELUS is using Arrcus to add a number of capabilities to its network, among them adding high-performance communications for mobile networks, improved security, as well as the capability to provide sovereign data segmentation.

CSPs Look for Network Upgrades

It turns out the massive investment in next-generation 5G infrastructure may not be enough for most CSPs. They now have to deal with the challenges of AI services, which require specialized security, low latency, and cloud data access. Another rising trend is the need to keep some network traffic in specific geographic areas to meet data sovereignty requirements.

Although this is a challenge for service providers on the infrastructure side, it’s an opportunity on the networking business side. Global networks are positioned to help provide better capabilities for AI services, including inferencing, which often happens on the edge of the network closer to the consumer, where AI training data is used to generate results from users' requests, whether that’s from humans or AI agents.

“With the influx of AI-related infrastructure demand, much of the market focus has been on training datacenters,” said Shekar Ayyar, CEO of Arrcus. “However, we are seeing more attention and white space around AI inferencing. Inference networks are more widely distributed than training networks, and they require more diversified form factors. They also need stronger policy controls.”

Arrcus sells to service providers, cloud providers, and large enterprises. But Ayyar believes the CSP market is a growing area of interest in AI infrastructure, due to the size of the infrastructure and the potential role they can play in enabling faster access to data, as well as sovereign services.

In the TELUS PoC, AINF will be used to deliver sovereign, distributed AI inferencing at national scale. TELUS says it will deliver secure, low-latency AI to mission-critical applications, from public safety and emergency response to enterprise and government services, while keeping sensitive data and workloads within Canadian borders.

Will this be a simple “cut-and-paste” for Arrcus to sell to many carriers? It may not be that easy, but Ayyar says there is ample market demand with “10 to 15 opportunities” in the pipeline. Arrcus already has deals in the CSP market, including with SoftBank and Liberty Global. The company earlier this year announced record bookings in 2025.

It’s a Matter of Policy

One of the key features of AINF is high-performance routing that can be assigned specific network policies. This could be helpful, for example, if a customer needs to know which AI agent traffic is required for a given task. It could also be possible for emergency public services. Networking policy can be applied around data access, business security requirements, and operational constraints. Geopolitical concerns are also driving more interest in data sovereignty boundaries.

Ayyar says this policy-based approach can be further integrated using NVIDIA’s Dynamo software. Arrcus partners with NVIDIA on the hardware side, delivering networking gear based on NVIDIA’s Spectrum-4 switches.

“Dynamo can be used to insert policy and intelligence into the network infrastructure," said Ayyar. "We can bring this capability into Layer 3 and work with Layer 2, implementing it at the hardware level. The hidden backbone of AI is the network. The network can directly affect GPU ROI."

Key parameters used in policy assignment might include data sovereignty, latency, and economic factors such as the cost of the networking traffic. TELUS will also be able to use these policy-based routing features to protect specialized services, such as low latency services or emergency channels.

The network could also optimize routing for agents and determine where computation should take place, said Ayyar. Traffic can be directed to achieve the right behavior, similar in concept to caching in traditional content delivery architectures.

This also applies to mobility infrastructure and wireless connectivity architectures within telecom environments such as Telus. These networks can create channels based on latency, security, and other policy requirements.

The goal, overall, is better economics and performance on the network.

Here are some of the key AINF capabilities announced with Arrcus for the TELUS Network:

Sovereign, Geo-Aware Routing: Enforcing geofencing and data sovereignty at ingress, supporting regulatory complaince and data residency within Canada
AI Policy-Aware Routing: Dynamically directing inference to optimal sites based on models, policy, capacity and network conditions
Mission-Critical Latency: Supporting real-time AI with up to 60% faster TTFT and 40% lower latency
Line-Rate Security: Providing up to 400-Gb/s encryption with zero CPU overhead using NVIDIA BlueField DPUs
Network Slicing & SLOs: Prioritizing critical workloads with operator-defined service levels for latency, throughput, power and cost
Distributed Edge Intelligence: Enabling edge-based inference while routing complex workloads to centralized resources
Open Integration: Integrating with NVIDIA Dynamo, vLLM, SGLang, Triton and Kubernetes for flexible, scalable deployment
SRv6 & Mobile User Plane (MUP) Transport: Leveraging Segment Routing over IPv6 (SRv6) and the Mobile User Plane (MUP) architecture for programmable, scalable transport of AI inference traffic across TELUS’ 5G and wireline network infrastructure
Silicon Diversity: Supporting a broad ecosystem of merchant silicon and hardware platforms across the PoC, enabling TELUS to build a flexible, vendor-agnostic infrastructure that avoids lock-in and optimizes performance and cost across heterogeneous deployment environments

Futuriom Take: The TELUS announcement with Arrcus AINF gives a compelling argument for the role that service providers can take in improving the performance, security, and control of the network for advanced and demanding applications such as AI inference, as well as many other services.