- Telus launched a PoC with Arrcus to test its Inference Network Fabric as the foundation for sovereign, distributed AI across Canada
- The fabric steers inference traffic in real time based on policies covering latency, data sovereignty, model preference and power constraints
- Arrcus positions itself against Cisco, HPE and others by arguing incumbents are forcing legacy tools onto a problem that demands new architecture
Telus is testing a new approach to sovereign AI networking, partnering with Arrcus on a proof-of-concept (PoC) that uses the Arrcus Inference Network Fabric (AINF) as the networking foundation for distributed AI inferencing across Canada's national infrastructure, Arrcus said.
The PoC targets mission-critical applications such as public safety, emergency response and government services where AI workloads must run with ultra-low latency while staying within Canadian borders, Arrcus said. AINF, which was introduced in February, functions as an intelligent control plane for distributed inference, dynamically evaluating operator-defined policies covering latency, data sovereignty, model preference and power constraints, then steering inference traffic to the optimal node in real time.
How the fabric works
At the technical core of the Telus deployment, AINF integrates with Nvidia BlueField-3 data processing units for line-rate encryption at up to 400 Gb/s with no CPU overhead, and with Nvidia Spectrum-4 Ethernet switches to create an inference fabric spanning edge, data center and cloud, Arrcus said.
Intelligent, policy-aware networking can deliver more than 60% reduction in time to first token, 40% lower end-to-end latency and up to 30% reduction in cost per inference, according to Arrcus.
"Inferencing requires a smarter network, and the smarter network needs to know how to direct traffic based on policies," said Shekar Ayyar, chairman and CEO of Arrcus. The PoC aims to give Telus an environment where mission-critical applications have AI "as a first-class citizen," while maintaining data sovereignty within Canadian borders, the Arrcus CEO told Fierce.
AINF gives Telus "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," said Tim Fell, VP of wireline technology and services at Telus, in a statement from Arrcus.
A crowded race for inference infrastructure
The Telus PoC comes as carriers face a growing field of vendors competing to define what AI inference infrastructure looks like at national scale. Cisco recently introduced its own AI fabric, and HPE, Extreme Networks, Vast Data and others are all competing to offer inference infrastructure.
Arrcus contends that incumbents are applying the wrong tools to the problem. "The incumbent approach, not surprisingly, is to take what you have as your tool in the toolkit, your sledgehammer, and throw that at the problem," Ayyar said.
Cisco offers routers, F5 offers load balancers — but those tools weren't built for the policy complexity that AI inference demands. Arrcus, along with DriveNets and IP Infusion, is among the few vendors that "genuinely have a disaggregated infrastructure architecture that allows the customer freedom of choice" across the full stack from application to operating system to silicon, the Arrcus CEO said.
The PoC is getting underway immediately, though one component — AINF's inference router — remains in development and is expected to be ready by year-end. Telus has not disclosed whether it is evaluating other vendors alongside Arrcus.
Arrcus said it is also in discussions with other national carriers focused on sovereignty, and is fielding early inquiries about extending its architecture to space-based data centers — a use case that adds complexity around satellite, mobile and fiber connectivity integration. The company grew revenue 3x last year, expects comparable growth this year and is targeting an IPO in 18 to 24 months.
We have an inquiry into Telus for its perspective; the company had not responded at press time.
As carriers weigh where to place their infrastructure bets, our Fierce Network Research report examines what the shift to distributed AI actually demands of the network: AI and the automated network: Designing telco infrastructure for the age of inference