Launched in April, the facility represents a monumental leap in AI development, fueling national research and innovation in Alberta
Canada is a global leader in artificial intelligence, a position we maintain through continuous advancements in our technological infrastructure. The Pan-Canadian AI Compute Environment (PAICE) project directly addresses the growing demand within Canada's research community by providing dedicated, national AI infrastructure. Building on the Government of Canada’s 2018 Pan-Canadian AI Strategy, PAICE is a collaborative national effort. It unites Canada’s three AI centres of excellence—Amii, Mila, and Vector Institute —with CIFAR (Canadian Institute for Advanced Research) and key host sites like the University of Alberta, Université Laval, and the University of Toronto. This extensive project is funded by Innovation, Science and Economic Development Canada (ISED) through the Digital Research Alliance of Canada (DRAC), with additional provincial support underscoring a strong national commitment to innovation.
Fueling Alberta's AI Breakthroughs
Amii worked closely with the University of Alberta to build Vulcan, a high-performance computing (HPC) site, which launched in April 2025. This large-scale facility will enable AI breakthroughs by providing the immense computational power needed to uncover novel applications. As a vital resource, Vulcan will bridge the gap between exploring new AI solutions and applying them to real-world problems. Whether for optimizing energy systems, developing new healthcare solutions, or enhancing logistics, the proximity of this advanced research to Alberta's industry leaders is accelerating the path from scientific discovery to commercialization and widespread adoption.
“AI is about understanding intelligence and demonstrating that understanding by building systems,” explains Michael Bowling, Amii Fellow and Canada CIFAR AI Chair.
Proper infrastructure ensures we have enough processing power to build, test and run these systems effectively. As Bowling notes, these systems "think" with computation, meaning significant computational power is essential for making real progress. This is especially true in rapidly evolving fields like natural language processing, computer vision, and reinforcement learning (RL), where the demand for high-performance compute infrastructure continues to grow in response to increasingly complex programming.
Building an HPC data centre from scratch
Through the PAICE project, each national AI centre – Amii, Mila and Vector Institute – worked with a designated host site (University of Alberta, Université Laval, and the University of Toronto, respectively) to build compute data centres. These data centres are uniquely tailored to each site’s research focus and existing infrastructure.
Robert Craig, Amii’s IT Director, leads the project for Amii, alongside IT Project Manager Jinal Kothari, who helped lay the groundwork for PAICE in its early stages. Together, they worked closely with a cross-functional team from the University of Alberta — including staff from IST, procurement, finance, and cybersecurity as the initiative scaled.
“This is not by any means a small project,” says Craig. “We’re building a high-performance compute AI infrastructure from scratch. To ensure we maintain our leadership in AI, it is important to have the computing infrastructure in place that can run high-powered AI models,” explains Craig.
The University of Alberta undertook significant upgrades to its existing infrastructure to support the deployment of a high-performance computing platform. These enhancements include improvements to both power delivery and cooling systems, ensuring the facility can reliably accommodate the increased demands associated with advanced research computing workloads.
Tailored for RL Excellence
As a global hub of reinforcement learning (RL), the Alberta site needed specific customization. To determine the data centre’s precise requirements, Craig and the team formed a dedicated Researcher Advisory Committee. The committee included Amii Fellows, University of Alberta researchers and Canada CIFAR AI Chairs Michael Bowling, Marlos C. Machado, Bei Jang, and Lili Mou, among others.
“One pretty unique thing about Amii research is that we often don't work with fixed static datasets,” said Bowling. “Reinforcement learning and other areas of particular emphasis at Amii involve taking action in real or simulated environments, and so the algorithms themselves are generating their data.”
This focus on self-generating data in reinforcement learning demands specialized compute resources. To address this, the team discovered that a powerful combination of CPUs and GPUs working in tandem would be most ideal for the Alberta site.
While GPUs excel at parallel processing—performing many calculations simultaneously—CPUs are critical for collecting experiences from the environment and managing the overall system. In RL algorithms, CPUs handle task distribution, data flow, and communication between different components, accelerating intensive computations and leading to faster learning and decision-making. Though certain components of reinforcement learning, such as neural network training, benefit from GPU acceleration, the overall RL process can be more CPU-intensive. This is particularly true concerning agent interactions with the environment, complex data processing and management, and sequential-logical decision-making—all components unique to RL.
An AI Catalyst for a Prosperous Future
The establishment of Vulcan marks a pivotal milestone. The PAICE Compute Project, now in its fourth year, has already completed its first $14 million HPC procurement and is preparing for a second this fiscal year.
The completion of Vulcan represents a transformative moment for AI in Canada, firmly cementing Alberta's position at the forefront of AI research and commercialization. Vulcan's ability to facilitate rapid iteration and experimentation with complex AI models will lead to more robust, accurate, and versatile AI solutions made in Alberta that can address real-world challenges.