The sovereign imperative: Why enterprise AI mandates a new infrastructure playbook

For enterprise Chief Information Officers, the honeymoon phase of artificial intelligence is over. As organizations move past baseline experimentation and begin anchoring generative AI and agentic workflows into core operational stacks, we are hitting a collective wall. That wall isn’t defined by a lack of use cases or algorithmic capability; it is defined by the harsh realities of data gravity, compliance, and foundational infrastructure.

When scaling models that manipulate proprietary IP, sensitive financial data, or highly protected personal health information (PHI), standard public clouds introduce existential risks. The moment data crosses international borders or becomes subject to foreign legal frameworks—such as the U.S. CLOUD Act—data sovereignty evaporates.

As technology leaders, we cannot close our productivity gaps by consuming AI built entirely on someone else’s terms, governed by someone else’s rules. To capture the real economic returns of this technology, enterprise intellectual property must remain local, secure, and under domestic control.

This is the exact challenge that triggered a massive architectural shift, leading to the creation of Canada’s first fully sovereign AI factory. Developed by TELUS in close strategic partnership with HPE and NVIDIA, this initiative provides a powerful example of how IT leaders can balance computational capacity with uncompromised data integrity.The infrastructure challenge: Beyond virtual machines

Building an enterprise-grade AI architecture requires moving past the concept of commodity virtual machines (VMs). Today’s workloads—encompassing massive foundational model training, intricate fine-tuning pipelines, and high-throughput real-time inference—demand an entirely different caliber of hardware and network integration.

When TELUS set out to establish its flagship facility in Rimouski, Quebec, the primary goal was to eliminate the technical bottlenecks that traditionally plague distributed AI systems: network latency, thermal throttling, and fragmented software layers.

To achieve this, the architecture was co-designed using the HPE AI Factory with NVIDIA. By integrating high-performance computing (HPC) stacks from HPE with high-density clusters of NVIDIA H200 Tensor Core GPUs, the system unlocked the high-density compute capacity needed to eliminate processing bottlenecks. Armed with 141GB of HBM3e memory and an explicit bandwidth of 4.8 TB/s per GPU, these compute clusters are architected specifically to handle multi-billion parameter datasets.

However, raw GPU power means little if the interconnect cannot sustain the data exchange. The Rimouski facility eliminates inter-node bottlenecks by leveraging NVIDIA Quantum-2 InfiniBand networking. This provides the ultra-low latency and predictable throughput required for massive parallel training jobs, allowing distributed nodes to operate efficiently as a single, unified computing cluster.

Backing this localized infrastructure is the TELUS PureFibre network, which provides a highly secure, ultra-low latency edge communication pipeline straight into the factory’s core. Because TELUS operates as an official NVIDIA Cloud Partner (NCP), enterprise customers gain native access to a thoroughly validated software stack. This includes the NVIDIA AI Enterprise platform and NIM microservices, removing the manual integration friction that can stall AI deployment.

Verifiable performance and sustainability results

The structural benefits of this specialized blueprint are clearly reflected in the system’s empirical metrics. Upon its launch, the TELUS Sovereign AI Factory was formally audited and recognized by the global TOP500 list as the fastest and most powerful supercomputer in Canada, delivering a record-breaking 22.74 petaflops of performance. For technology leaders, this level of scale turns operations that previously took weeks or months into tasks completed in hours.

Yet, as any CIO managing a data center budget knows, high-density compute can come with a punitive environmental and financial cost. Running thousands of high-TDP accelerators creates unprecedented thermal loads.

The Rimouski facility answers this challenge by decoupling scale from carbon output. Operating on 99% renewable hydroelectricity, the facility is ranked among the world’s most efficient systems on the Green500 list. It features an advanced cooling infrastructure that capitalizes on its geographical location, utilizing intelligent “free cooling” mechanisms for 98% of the year. This design slashes mechanical cooling dependencies down to just roughly 40 hours annually, reducing net water consumption by over 75% and operating at three times the energy efficiency of standard enterprise data centers.

The market response to this combination of security, scale, and sustainability has been decisive. The initial capacity of the Rimouski supercomputer sold out within months of its commercial availability.

To meet this aggressive demand, TELUS, in coordination with the Government of Canada’s Enabling Large-Scale Sovereign AI Data Centres initiative, is rapidly scaling this architecture into a distributed national AI grid. A second sovereign AI factory in Kamloops, British Columbia, is coming online, boasting advanced closed-loop liquid cooling and heat-recovery systems. This will be followed by two major new installations in Vancouver (the M3 facility in Mount Pleasant and the 150 West Georgia facility), scaling the cluster’s collective footprint to over 150 megawatts (MW) and up to 60,000 GPUs by 2032.

Telus sovereign AI facility roadmap

• RIMOUSKI, QC | Live and operational 22.74 Petaflops | TOP500 #78
• KAMLOOPS, BC | Launching Fall 2026 with liquid cooling and heat recovery
• VANCOUVER, BC | M3 facility late 2026 -> 150 W. Georgia 2029
• TOTAL TARGET | 150+ MW capacity | 60,000+ NVIDIA GPUs by 2032

The sovereign advantage in practice

True validation of any architecture lies in the workloads it protects. By establishing an environment where 100% of data processing, operational logging, and storage reside within national borders under local legal jurisdiction, the factory has become a magnet for highly regulated entities.

Several global early adopters are already leveraging the TELUS Sovereign AI Factory to drive sophisticated, large-scale enterprise deployments:

• OpenText: The enterprise information management giant uses the Rimouski facility to host a dedicated sovereign configuration of its Aviator AI platform, providing bulletproof data residency assurances to more than 1,600 Canadian corporate and public sector clients.
• Accenture: The global systems integrator leverages the factory’s high-density compute environments to build and deploy complex, domain-specific AI models for highly regulated verticals, including financial services and public infrastructure, where cross-border data transit is a compliance disqualifier.
• League: A prominent healthcare consumer experience platform utilizes this Canadian-controlled perimeter to build out personalized, predictive healthcare solutions, ensuring sensitive patient health records remain entirely isolated from foreign access.
• Mila (Quebec AI Institute): The renowned research facility partners with TELUS to fuel deep-learning research breakthroughs, ensuring that the resulting intellectual property and foundational research remain distinct assets within the domestic economy.

Strategic takeaways for CIOs

If your current AI strategy relies purely on standard, multi-tenant public cloud infrastructure, it is time to reassess the long-term viability of that model. As you plan your capital allocations and architecture designs for the coming fiscal years, keep these three peer-driven principles in mind:

1. Own the data perimeter: Compliance is no longer just a checkbox handled by your legal team; it is a foundational architectural element. Evaluate your workloads based on data residency requirements. If a model handles proprietary customer data or core operational logic, it belongs in a sovereign environment where third-party backdoors and extraterritorial data access laws are structurally impossible.
2. Architect for compute density: Do not underestimate the network demands of multi-node AI clusters. Ensure your infrastructure partners are utilizing unified architectures—like the custom integrations developed by HPE and NVIDIA—to minimize latency and optimize total cost of ownership (TCO) at the hardware layer.
3. Prioritize sustainable scale: AI execution is an energy-intensive endeavor. Choosing data centers that utilize native free cooling and renewable power sources isn’t just an exercise in corporate social responsibility—it directly insulates your operational budgets from fluctuating grid costs and carbon penalties as your computational needs scale.

The era of building AI on infrastructure governed by someone else’s rules is coming to an end. By shifting toward dedicated, sustainable, and fully sovereign AI factories, forward-thinking technology leaders can confidently build an intelligent enterprise that is secure, compliant, and engineered to endure.

To learn more about HPE AI Factory with NVIDIA, visit hpe.com/ai.

**************

As AI becomes increasingly central to economic competitiveness, scientific advancement, and national priorities, organizations require infrastructure that balances performance with security and sovereign control. Together, HPE and NVIDIA co-engineer rack-scale AI systems that integrate AI computing, high-performance networking, and supercomputing expertise to support large-scale AI workloads. This provides enterprises, governments, and research institutions with a trusted foundation for sovereign AI initiatives while maintaining control over critical data, models, and operations.