TELUS, in partnership with HPE and NVIDIA, built Canada’s first fully sovereign AI factory in Rimouski, Quebec, combining HPE HPC stacks with NVIDIA H200 GPUs and Quantum-2 InfiniBand networking to minimize latency and avoid bottlenecks.

For CIOs, the playbook is to own the data perimeter for residency and governance, design for high compute density with unified architectures to cut latency and cost, and pursue sustainable scale through free cooling and renewable power to stabilize operating costs.

Advocates say enterprise AI demands sovereign infrastructure to protect data sovereignty, compliance, and IP when handling sensitive data and cross-border transfers.

The Rimouski facility uses 141GB of HBM3e memory per GPU and 4.8 TB/s GPU bandwidth, enabling handling of multi-billion-parameter datasets and scalable distributed training.

Early adopters include OpenText with sovereign Aviator AI for 1,600+ clients, Accenture in regulated verticals, League for healthcare data residency, and Mila leveraging domestic data controls.

Market demand has been strong, with initial capacity selling out in months; plans call for a distributed national AI grid, starting with a second factory in Kamloops featuring liquid cooling and heat recovery, then two facilities in Vancouver to reach over 150 MW and 60,000 GPUs by 2032.

The factory ranked 22.74 petaflops on TOP500, making it Canada’s fastest supercomputer, and it runs on 99% renewable hydroelectricity with advanced cooling and free cooling, achieving high energy efficiency and reduced water use.

The article argues the era of AI infrastructure tethered to external multi-tenant clouds is ending, giving way to dedicated, sovereign AI factories with ongoing collaboration among TELUS, HPE, and NVIDIA.

TELUS PureFibre provides secure, ultra-low latency edge connectivity with native access to NVIDIA AI Enterprise and NIM microservices to simplify deployment for enterprise customers.