What We Heard Report: Consultations on AI Compute

Table of Contents

• Executive summary
• Introduction
• Overarching findings
• Access: Ensuring availability and affordability of AI computing resources
• Talent: Attracting and retaining AI talent
• Data: Securing high-quality data and data sovereignty
• Other themes
• Conclusion
• Annex A. Online survey overview and questions

Executive summary

Budget 2024 announced the Government of Canada's commitment to launch a new Canadian Artificial Intelligence (AI) Sovereign Compute Strategy and AI Compute Access Fund. This is a $2 billion investment that will provide Canadian researchers and AI companies with the tools needed to be competitive in a rapidly advancing global AI landscape.

Between June and September of 2024, ISED conducted consultations through roundtables and online input to inform the design and implementation of the two initiatives: a new AI Compute Access Fund and a Canadian AI Sovereign Compute Strategy. Stakeholders from large, medium and small businesses, industry associations, Indigenous peoples, not-for-profits, post-secondary institutions, and research institutions across Canada attended the series of in-person and virtual roundtables. There were more than 1,000 submissions provided via the online survey and email. This report provides a summary of the input collected through the roundtable discussions, combined with written submissions received through the online consultation.

Overall, we heard consensus during the consultations that affordable access to computing resources, AI talent, good quality data, and national infrastructure are essential to Canada's ability to stay competitive in the AI landscape. This report is organized into the three main themes which emerged during the roundtable discussions and online submissions:

• Ensuring accessible and affordable AI computing resources
• Attracting and retaining AI talent
• Securing high-quality data and data sovereignty

The following overarching findings emerged regarding AI Compute in Canada:

• There is an urgent need to invest in AI compute to lay the groundwork and secure supply chains for the AI ecosystem, given the rapid technological advancements in AI.
• For industry, high cost is the most cited issue when accessing AI compute. Many companies get locked in to a specific provider when accessing compute and can face steep cost increases when trying to scale up.
• Affordability aside, access to AI compute is less of an issue for industry, except in cases where data needs to reside in Canada or when smaller companies struggle to secure short-term contracts for computing resources.
• Access to AI compute is a growing concern for researchers due to lack of publicly available high-performance computing (HPC) resources to meet demand.
• There is a strong demand for open, good quality datasets to support AI development as it impacts AI models' performance, accuracy and reliability.
• Enhancing data security and privacy and ensuring compliance with regulations to protect sensitive information should be prioritized.
• Access to cutting-edge computing infrastructure is crucial to develop, attract, and retain talent.
• Talent development is a cornerstone of advancing AI compute systems, directly impacting both the immediate workforce needed to build and manage these systems and the long-term pipeline of talent from academia to industry.

Introduction

Background

Canada has a world-leading AI ecosystem – from development, to commercialization, to safety. Canada was the first country in the world to introduce a national AI strategy and has invested over $2 billion since 2017 to support AI and digital research and innovation. Building on this foundation, we can make sure Canadian values and Canadian ideas help shape this transformational technology.

Estimates suggest that AI could contribute $15.7 trillion USD to the global economy and boost North America's GDP by 14.5% in 2030. Its transformative influence is already evident in industries such as healthcare, finance, and manufacturing, reshaping not only how we work but also how we live. Canada, a home to over 1,500 firms specializing in AI and 20 public AI research labs, is competing for its share of the global AI market, which is expected to grow over $2 trillion CAD by 2030.

As AI continues to drive innovation and economic growth globally, there is increased demand for AI computing infrastructure. AI compute refers to the computational resources required to train, develop, and deploy AI models. This includes HPC systems, specialized hardware such as Graphics Processing Units (GPUs) and Tensor Processing Units (TPUs), and cloud-based platforms that provide scalable resources. Access to AI compute is crucial as the more complex and data-driven AI models become, the more computational power they require to function and deliver results. However, Canada is lagging all G7 nations in HPC resources. Investing in national AI computing capacity will ensure Canada's world class researchers and innovators have the tools they need to drive frontier research and AI solutions for adoption by all sectors. Such an investment will advance economic growth by building domestic compute capacity and supply chain resilience.

Budget 2024 announced an investment of $2.4 billion to secure Canada's AI advantage. This includes:

• $2 billion to support access to computing capabilities and build technological infrastructure for Canada's world-leading AI researchers, start-ups, and scale-ups, including:
  • an AI Compute Access Fund that will provide support for compute power for Canadian AI researchers, start-ups and scale-ups; and
  • a Canadian AI Sovereign Compute Strategy to develop sovereign AI infrastructure to respond to both near-term and long-term needs of researchers, industry, and government.

These initiatives will enable Canada to secure its globally competitive position by ensuring that both our AI industry and researchers have access to affordable and cutting-edge infrastructure to support the growing AI ecosystem. The purpose of this report is to summarize the consultations that were undertaken to inform the design and implementation of the AI Sovereign Compute Strategy and AI Compute Access Fund.

Consultative process/methodology

Roundtables

A series of roundtables were held from June to September 2024, to seek input from a wide range of AI stakeholders. In-person roundtables were hosted by the three national AI Institutes – Amii in Edmonton, Mila in Montreal, and the Vector Institute in Toronto – and other key industry players – Atlantic Canada Opportunities Agency (ACOA), Canadian Institute for Advanced Research (CIFAR), U15 Group, Canada's Tech Network, Digital Supercluster, Ocean and ScaleAI Superclusters, and Communitech. Virtual roundtables were also conducted with provincial governments, Indigenous groups, academic institutions, and other stakeholders to gather feedback on AI compute.

Consultation survey

ISED conducted a public online survey through the ISED website. The survey was officially launched on June 26, 2024, and closed on September 6, 2024. More than 1,000 submissions provided input via the survey link or email (aicompute-calculia@ised-isde.gc.ca). Participants included Canadian data centre firm/cloud providers, multi-national firms, industry associations or business organizations, Canadian AI firm/developers, entrepreneurs, Canadian AI researchers, post-secondary institutions, research institutions, civil society organizations, government department/agency, not-for-profit or government organizations, interested Canadian, and others.

The survey asked thirteen questions under three themes: 1) access to compute; 2) developing sovereign and sustainable compute capacity; and 3) priorities for longer-term compute infrastructure. Respondents were encouraged to answer the questions of most interest and where they had the most expertise. As such, the number of responses for each question varied. The full list of survey questions are in Annex A of this report.

Indigenous engagement

ISED supports the Government's commitment to renewing its relationship with Indigenous peoples, which is based on the recognition of rights, respect, cooperation and partnership. Through this consultation, ISED engaged with Indigenous partners through both virtual and in-person bilateral meetings, conducted with the consent of the Indigenous organizations. This engagement was to help shape the strategic framework on AI compute and gain a better understanding of the concerns and needs of Indigenous partners. Out of the nineteen Indigenous organizations consulted, eleven participated.

Findings

ISED analyzed the feedback from the roundtables, responses to the online survey and email input, and identified some overarching themes and findings. These are captured under "overarching findings" and are also discussed in greater detail under the different themes.

Note: This report does not attempt to interpret respondents' feedback or translate it into policy solutions, but rather, to reflect it as it was articulated.

Overarching findings

A number of overarching themes emerged throughout the consultations.

For industry, high cost is the most cited issue when accessing AI compute. Many companies get locked in to a specific provider when accessing compute and can face steep cost increases when trying to scale up.

Generally for industry participants, affordability of AI compute is more of an issue than accessibility. Hyper-scalers provide reliable services, but some innovative startups are priced out of reaching their full potential quickly. Accessing AI compute is expensive, especially for startups and smaller companies that rely on AI, limiting their growth.

Access to AI compute is a growing concern for researchers due to lack of publicly available HPC resources.

While researchers have access to the Canada's public computing infrastructure, many researchers noted that the supply available is insufficient to meet demand. Researchers suggested that acquiring more GPUs and expanding the compute capacity provided through the Digital Research Alliance of Canada (DRAC) should be prioritized to increase availability and scalability of resources.

There is a strong demand for open, good quality datasets to support AI development as it impacts AI models' performance, accuracy and reliability.

High-quality, open, and accessible data is essential for advancing AI research and innovation, with data quality and availability directly impacting the validity and scalability of AI models. Robust, curated datasets reduce bias and enhance generalizability, while open data fosters collaboration, supporting new discoveries and maximizing research impact.

Enhancing data security and privacy and ensuring compliance with regulations to protect sensitive information should be prioritized.

Stakeholders highlighted data sovereignty and security as one of the top benefits of Canadian-owned and controlled computing infrastructure to reduce reliance on foreign entities and to be compliant with national data protection laws. Particularly in sectors with heightened privacy sensitivities, such as healthcare and defense, data sovereignty is a pressing concern.

Access to cutting-edge computing infrastructure is crucial to develop, attract, and retain talent.

Investing in cutting-edge computing infrastructure, such as GPUs, cloud resources, and specialized AI hardware, is essential for attracting and retaining top talent and fostering the development of a skilled workforce. With global competition to attract and retain AI talent, Canada must ensure it is equipped with advanced computing infrastructure.

Talent development is a cornerstone of advancing AI compute systems, directly impacting both the immediate workforce needed to build and manage these systems and the long-term pipeline of talent from academia to industry.

The process of building and operating AI systems is highly technical and interdisciplinary, requiring talent not only proficient in AI-specific skills but also versed in data management, network optimization, and high-performance computing. These highly qualified professionals are responsible for designing scalable architecture, managing massive datasets, maintaining system performance, and ensuring cybersecurity. Without a robust talent pool, these essential tasks could become bottlenecks, slowing innovation and limiting the capability of AI infrastructure to meet increasing demands. A sustainable pipeline of talent flowing from academia to industry is essential to Canada's continued AI leadership.

Access: Ensuring availability and affordability of AI computing resources

Consultations highlighted high cost and limited availability – especially for domestic compute resources – as significant barriers to accessing AI compute. Concerns about access varied between different stakeholder groups, as resource preferences for accessing compute differed by group. Industry respondents, who predominately use cloud-based platforms, commonly raised concerns about the high cost of cloud services with 45% of Canadian AI firms/developers in the online survey identifying it as a barrier. On the other hand, availability was a bigger concern than affordability for the research community with 43% of online survey respondents identifying it as a barrier in accessing compute.

Canadian researchers primarily access their compute though DRAC's public compute infrastructure. DRAC plays a central role in supporting Canada's advanced research and innovation landscape by providing essential digital research infrastructure (DRI), including HPC, data management, and research software. Established as part of Budget 2018 to unify and expand Canada's DRI capacity, DRAC ensures that Canadian researchers and institutions have access to the technology needed to conduct world-class research across various disciplines. DRAC's infrastructure includes a network of supercomputers located at five sites across the country (McGill University, University of Victoria, Simon Fraser University, University of Waterloo, and University of Toronto), offering computational power essential for projects that involve large-scale data processing, complex simulations, and high-level machine learning tasks.

Industry

The most common method of accessing compute is through hyper-scalers like Amazon Web Services, Microsoft Azure, and Google Cloud Platform. Industry respondents in the online survey indicated that they primarily rely on cloud services for scalable GPU and TPU access, with some using on-premise infrastructure or hybrid cloud solutions consistent with the types of workloads and model trainings.^{Footnote 1}

Although cloud services are expensive, startups and small businesses rely on hyper-scalers for reliable computing resources because there are no cost-effective alternatives. The three commonly identified barriers to accessing AI compute among industry participants were: high costs, data security and privacy concerns, and insufficient computing resources. From the online survey, high cost was a key challenge noted by industry respondents, including both Canadian AI firms/developers and Canadian data centre firms/cloud providers. Canadian AI firms/developers raised concerns about the high cost of cloud services, while Canadian data centre firms/cloud providers were more concerned about the costs associated with operating data centers in Canada.

Some startups raised concerns about the difficulties in accessing the necessary computing resources (e.g. availability of short-term contracts of advanced GPUs, restrictions on access to virtual models at times). Limited availability of HPC resources during peak times exacerbates competition, as smaller companies often struggle to compete with larger entities that have more computing options and can leverage economies of scale.

Online respondents who identified as entrepreneurs noted that accessing cloud services was a challenge due to limited availability caused by pricing tiers and queues dominated by larger organizations. They also pointed out restrictions on model customization, as well as the difficulty of finding short-term contracts (which often come with a large premium). This commitment model of cloud services tends to favour tech giants and well-funded corporate entities that can absorb these costs. High costs strain early-stage ventures, forcing them to divert funds from other critical areas like talent acquisition.

Researchers

The three most commonly identified barriers to accessing compute resources by researchers were limited availability due to supply constraints, high costs, and concerns over data security and privacy.

Canadian AI researchers predominantly access computing resources through DRAC as highlighted via the online survey with approximately 40% of the respondents from the research community identifying it as their method of compute access. This was followed by researchers accessing compute through cloud services, locally on personal devices, and at their respective post-secondary institutions.

Canada's public compute infrastructure is available to researchers at no-cost, and thus researchers are less likely to buy access via a hyper-scaler. For researchers, limited supply was identified as the biggest barrier via the online survey, and cost was less of an issue (25%) compared to the industry respondents (45%). Many participants have called for an expansion of DRAC's public computing capacity due to the following highlighted limitations of its HPC resources:

• Long wait times and performance issues

  The demand for public compute infrastructure often exceeds supply. This is especially evident during peak times in the research/academia calendar when the demand for computing resources spike. Some participants noted that this could lead researchers to modify their projects based on the limited compute available.

• Scarcity of the latest GPUs for cutting-edge AI research

  Access to HPC systems, particularly GPUs available via public compute, is limited, making it difficult to run large projects effectively. Although major upgrades to GPC and CPU supply are currently underway to address demand in the very near term, they may be insufficient given the rapidly growing demand for research computing resources. Given the rapid advancement in AI, Canada's public compute infrastructure, as it stands, was not specifically designed for AI intensive workloads, as they often feature only a small number of GPUs per node. Researchers emphasized the pressing need for access to high-memory GPU computing resources.

Overall, there is demand to grow and expand Canada's public computing infrastructure capacity. Participants noted that Canadian-owned and controlled computing infrastructure could provide more economical resources and enhance performance for AI compute.

Talent: Attracting and retaining AI talent

Canada's AI sector is a key job creator and driver of productivity, innovation, and economic growth. Canada is home to over 1,500 companies specializing in AI and 20 public AI research labs. In 2022, Canada had 3% of the world's top-tier AI researchers, ranking sixth globally, and since 2019, Canada has ranked first in the G7 for the number of AI-related papers published per capita. Canada is rich in AI talent and research, but it must continue to invest in the resources and programs to both retain its current AI talent and continue to attract in the face of global competition.

Common themes have emerged during consultations, including recommended approaches to fostering attracting and retaining AI talent.

• Invest in AI compute infrastructure

  Stakeholders have emphasized that in order to attract and retain top AI talent, access to robust AI computing infrastructure is essential. Attracting the best candidates without offering top-tier equipment is difficult. Since the availability of cutting-edge GPUs is crucial for large language model (LLM) researchers in particular, one key solution to retaining AI talent is to ensure easy access to affordable GPU infrastructure. Developing state-of-the-art research facilities and providing access to the technology is critical to positioning Canada as an attractive destination for top AI talent to work and conduct research.

• Financial incentives

  Industry players seek financial incentives from the government to retain and attract top talent, because competitive compensation packages are critical in the global market and high costs of hiring AI talent become a growing concern when scaling up technology. High salaries and benefits are necessary to secure skilled professionals, particularly in the fields like technology and AI, where demand far outstrips supply. Some stakeholders mentioned that government incentives to businesses that hire AI professionals, such as tax credits, grants, or subsidies, can help companies offset these costs, enabling them to offer attractive employment packages without compromising their financial stability.

  Participants from the research community highlighted the need to provide funding for graduate students in AI to attract and retain top talent, ensuring they can access to competitive incomes and cutting-edge technology. By offering substantial research funding, grants, scholarships, and fellowships for AI projects, Canada can encourage researchers to stay and help create a pipeline of skilled talents that will contribute to the AI ecosystem.

• Education and skills training

  Participants highlighted the need to invest in educational programs to develop a strong pipeline of AI talent capable of optimizing resources and adopting new technologies. This includes offering specialized degree programs, certifications, internships, and continuous learning opportunities in advanced studies in AI and related fields. Scholarships and funding can be provided to accelerate this.

  Roundtable participants noted a challenge posed by the gap between academic training and industry needs. This gap is particularly problematic for startups with limited resources, as it leads to longer onboarding times for new employees. Aligning academic curricula with industry needs in AI is essential, along with opportunities that incentivize talent growth and training to bridge this gap. Furthermore, both startups and SMEs need support programs on how to optimize the use of compute resources and adopt AI technologies, as expertise in AI requires continuous learning and adaption to keep pace with rapid technological advancements.

• Research efficiency and support

  Consultations also emphasized the importance of providing training for researchers, particularly those who want to incorporate AI into their work but are not specialized AI researchers themselves. There is a concern that these researchers encounter barriers due to lack of necessary support and training, which prevents them from leveraging AI in their research. Therefore, providing targeted training programs and resources, along with access to advanced AI computing infrastructure, is crucial to ensuring these researches in Canada can remain competitive in the landscape.

Data: Securing high-quality data and data sovereignty

Consultations highlighted the need for data sovereignty and security in the development of Canada's AI compute capacity. Respondents from industry and research highlighted data privacy and security as key benefit of Canadian-owned and controlled computing infrastructure.

Common themes of challenges related to data have emerged from consultations.

• Data security

  Discussions revealed that a Canadian-owned and controlled computing infrastructure could reduce the risks of external government influence, minimizing the threat of data breaches or unauthorized access. With enhanced data sovereignty and security in Canada, firms and researchers can mitigate geopolitical risks.

  While data sovereignty and storage locality were not the main concerns in accessing compute for most consultation participants, some noted that they use on-premise solutions with enhanced security for handling sensitive data. Similar to industry respondents, research sectors with heightened privacy concerns (e.g. healthcare and defense) raised reservations about storing sensitive data on computing resources provided by third-party cloud services. The risk of cyber-attacks or data leaks is higher for entities in finance and insurance sectors due to the high sensitivity and value of their data.^{Footnote 2} These sectors stressed the importance of strict data sovereignty measures to protect sensitive information effectively.

• Regulatory compliance

  Stakeholders noted that while private cloud services offer robust security, managing regulatory compliance across multiple jurisdictions adds complexity. In particular, for businesses with strict data residency requirements (e.g. sciences, defense, geospatial data), data security was one of the primary concerns when accessing compute. These sectors face greater constraints and regulations in order to scale efficiently within Canada and effectively integrate AI into existing technology. A Canadian-located and controlled infrastructure ensures that sensitive data used for AI training and inference remains under Canadian jurisdiction, under Canadian ownership, thereby enhancing data sovereignty. Such investments would align with Canada's regulatory frameworks, enabling industry players to more easily comply with local regulations and standards related to AI ethics, data privacy, and security.

• Intellectual property (IP)

  Sufficient Canadian-owned infrastructure, such as data centres, is essential to ensure that IP developed within Canada remains protected from foreign ownership. Consultations noted that valuable datasets can be kept within the country by enabling innovators to train AI models on domestic infrastructure. This would support both the safeguarding and generating of IP within Canada. Supporting the growth of IP is critical for Canada's global competitiveness, as it secures both economic and technological sovereignty.

• Barriers in policy

  Research participants also noted that one of the main barriers enabling data use is the complexity of the policy landscape and lack of a unified approach to data sharing and utilization in Canada. For example, health data is often governed by different regulations at the institutional, provincial, and federal levels.

Key considerations for advancing data sovereignty

Four key considerations emerged for the government to prioritize when advancing data sovereignty and security in Canada.

• Establishing a regulatory framework for data management

  Creating a data regulatory framework tailored to Canada's unique needs is essential to improve data access and aggregation. Developing policies to unlock and utilize Canadian data sources could also attract companies and foster innovation. Canada has the opportunity to claim global leadership in sectors where it possesses valuable data (e.g. healthcare). For example, creating a national strategy for health data sharing and utilization would enable better collaboration between public healthcare providers, researchers, and businesses. Currently, there is no unified approach to data sharing and utilization across provinces and between public and private entities. To address this gap, some stakeholders suggested that a standardized framework across jurisdictions could be established for secure data sharing, with governance structures in place to ensure the ethical use of data. This would not only facilitate better data integration but also enhance collaboration and innovation across sectors.

• Addressing geographical disparities in infrastructure

  Several participants highlighted the need to expand compute infrastructure in rural and remote areas to ensure that opportunities for innovation are not limited to major cities. Canada faces geographic disparities in its AI computing infrastructure, which can hinder resilience and innovation. This is a result from the concentration of resources in established AI hubs in Ontario and Quebec, driven by access to skilled talent, academic institutions, and lower-cost renewable energy, while underserved regions face challenges related to lower demand and limited investment. Currently, some stakeholders noted that researchers and firms in Western Canada using domestic data centres often rely on facilities located in Ontario, creating a vulnerability where an unplanned shutdown could disrupt projects. Moreover, real-time data processing is often being hindered by latency due to the physical distance from these data centres. To mitigate such risks, it is important to distribute resources more evenly across regions, ensuring that if one area is compromised, other can compensate. It is also important to recognize that some Indigenous lands lack the necessary connectivity redundancy to support data centres.

• Growing Canada's compute capacity to train domestic AI models

  By investing in local infrastructure and resources, Canada can ensure that sensitive data remains within its borders, reducing reliance on foreign computing services. Innovators would be able to effectively utilize Canadian datasets and generate IP. This approach not only safeguards data but also promotes technological advancement within Canada's AI ecosystem.

• Developing shared public datasets

  It is recommended to develop shared public datasets and environments for AI training and testing to support research and innovation. A key priority should be ensuring data lineage and provenance when making datasets available. Establishing and maintaining a consistent and secure solution for dataset accessibility is important and should be done in close collaboration with the research community. This approach would ensure transparency and fairness, allowing all stakeholders to benefit without favouring specific technologies or discouraging future innovators from entering the field.

Other themes

Other themes outside of access, talent, and data were highlighted during consultations.

Environmental considerations and energy

One of the recurring themes across responses was the environmental impact of AI infrastructure. Participants called for more sustainable and energy-efficient solutions to address this challenge to mitigate the carbon footprint associated with AI infrastructure. Training large AI models are highly carbon-intensive due to the infrastructure consuming vast amounts of electricity and relying on energy grids powered by fossil fuels. Data centres, in particular, contribute to carbon emissions, electronic waste, and water usage due to intensive cooling requirements.

Canada could leverage its unique advantages, such as clean energy, to build a competitive computing infrastructure and climate that reduces the need for cooling in data centres. Investments in the national infrastructure should prioritize locations with clean electricity sources, as developing a domestic computing capacity will put increased pressure on energy usage. These strategic investments could also align with government priorities to use clean and sustainable energy sources.

Commercialization and AI adoption

Consultations highlighted that a larger deficit in Canada's AI ecosystem is the commercialization and adoption of AI. It is often attractive for startups from AI-related fields to move to other countries in part due to easier access to investment capital and higher compensation opportunities for commercialization. To address this, Canada must create an environment in which it is more economically viable to build and scale high-tech businesses to compete globally.

Canadian firms also identified the high costs of AI technologies and workforce skills gaps as significant barriers to AI adoption. Developing talent, increasing funding, and expanding infrastructure resources would positively impact the implementation of AI technologies and the scaling of research and development occurring at Canadian institutions and startups.

Public-private partnership

Public-private partnerships in AI computing go beyond expanding capacity or providing access to expertise; they are crucial for fostering innovation, developing talent, and driving strategic advancements. While the private sector has its established mechanisms in innovation, AI researchers can bridge capability gaps at innovation hubs that offer opportunities for collaboration between industry and academia. The creation of state-of-the-art AI innovation hubs is particularly important in an experiment-driven field like AI, where collaborative environment is essential for success. These partnerships accelerate scientific discovery and expand the frontiers of knowledge, benefiting Canada's national security, scientific leadership, and industrial competitiveness.

Supply chain stability

Participants highlighted concerns about global supply chain disruptions and how Canadian-made solutions could safeguard against these risks. A Canadian compute infrastructure is necessary to protect our economy from future disruptions and ensure continuous access to AI computing resources for innovation.

Government

The two most powerful computing systems in the Government are optimized for weather and climate modeling, which are operated by Shared Services Canada (SSC) on behalf of Environment and Climate Change Canada. SSC also provides computing resources and data storage to science-based departments and their collaborators with the General Purpose Science Clusters. Additionally, National Research Council Canada owns and operates several clusters and has AI research capacity in house.

Overall, there is a demand from departments for more access to HPC infrastructure to advance research, innovation, and development. There is also a demand for training to help users effectively utilize public cloud services, which can help reduce costs while ensuring compliance with regulatory requirements.

Own the podium

Seize the opportunity to grow domestic AI champions and ensure Canadian innovations and tech suppliers are embedded in the compute supply chain. It is important to support the development of Canadian AI companies across various sectors and the growth of a diverse AI ecosystem, from startups to national champions.

Canada can allocate resources strategically to maximize impact in chosen areas. For example, the semiconductor industry is an area where Canada has advantage with our expertise in photonics and compound semiconductors that could contribute to building Canada's AI ecosystem. Consultations also noted that Canada is seen as a global leader in "ethical AI" and could leverage this to attract top talent and new foreign direct investment.

Quantum

Quantum Industry Canada groups semiconductors, AI, and quantum computing under the umbrella of "frontier computing". While the Canadian quantum community recognizes that quantum computing is currently less mature than AI, participants noted that Canada has existing quantum computing infrastructure that could be expanded to provide more computing power to the research community. Developing secure, edge computing capabilities with quantum-safe cryptography could be crucial to meeting AI needs, supporting advancements in fields like healthcare, finance, and cybersecurity while ensuring data protection and resilience against privacy threats.

International collaboration

There was interest in formalizing partnerships between Canada and other likeminded countries to foster stronger global networks. Participants noted that Canada's AI compute capacity could be used to facilitate partnerships with nations that share similar values on issues like data privacy and ethical AI. This would enhance joint research capabilities and create opportunities for collaborative projects on a global scale.

Conclusion

AI computing has emerged as a critical issue in the Canadian AI landscape, marked by challenges related to the affordability of computing resources, retention and attraction of AI talent, and risks associated with the usage of data. Through roundtable discussions, survey responses, and email input, valuable insights were provided related to the challenges in accessing AI computing resources in Canada and how they could be addressed moving forward. These insights will inform the development of the Canadian AI Sovereign Compute Strategy and AI Compute Access Fund, announced in Budget 2024.

Annex A. Online survey overview and questions

Overview

A consultation survey was officially launched on June 26, 2024 and closed on September 6, 2024. Respondents were encouraged to respond to the questions of most interest and where they had the most expertise. As such, the number of responses for each question varies throughout the consultation.

• A total of 942 responses were received via the online survey with an overall completion rate of 22%.
• 49% of respondents identified as industry representatives, 20% as researchers or affiliated with post-secondary institutions, and 12% as government or not-for-profit representatives. Additionally, interested Canadians accounted for 13% of the responses.
• 23% of the participants representing businesses were from small business (10-99 employees), followed by 17% from micro-business (fewer than 10 employees), and 15% from larger business (500 or more employees).
• Nearly half (45%) of the businesses were located in Ontario, 16% in Quebec, 15% in British Columbia, 8% in Albera, with the remaining provinces and territories each representing under 2%. There was no representation from Northwest Territories and Nunavut.

Questions

Access to compute

• How do you currently access compute? What are the most pressing issues you face in accessing AI compute in the ecosystem currently (i.e., level of access, cost, security etc.)?
• What level and type of compute access do you require to grow and scale your business or conduct research? Are there existing programs or support mechanisms that could be leveraged in the short term?
• What new approaches should the Government consider to support researchers and industry access to AI compute in the short term?

Developing sovereign & sustainable compute capacity

• What is the benefit to you of having computing infrastructure that is Canadian-owned and controlled?
• What are the short-term opportunities or initiatives that would help to expand existing computing infrastructure in Canada?
• In your view, what are the opportunities to incorporate Canadian-made computing hardware and software as part of Canadian AI Sovereign Compute Strategy?
• How can we leverage this investment to both retain and attract AI talent in Canada?
• Are there specific collaboration models that would help to support industry/academic partnerships?
• Which model, or combination thereof, do you think would achieve results most efficiently and effectively?

Priorities for longer-term compute infrastructure

• What would you view as the priority components or elements of National AI Compute Infrastructure for the next five years?
• How can public infrastructure be leveraged to support both the needs of researchers and industry? Are their unique requirements to these stakeholder groups which should be considered?
• What level of support to compute infrastructure providers would incentivize an increase in the supply and availability of compute in Canada and over what length of time could an increase in supply be expected?
• How can Canadian compute be leveraged to collaborate and expand relationships and networks with likeminded countries?