Canada's Mixed Signals on AI and Cloud Could Doom Both
Prime Minister Carney's government have signaled great interest in both AI and cloud data centres. Do they have the stomach to do both?
Canadian Cyber in Context is sponsored by
All views expressed belong to Canadian Cyber in Context and do not reflect the position of any sponsor.
During Canada’s 2025 federal election for the 45th Canadian Parliament, Prime Minister Mark Carney and the Liberal Party made investing in the overall artificial intelligence (AI) supply chain a core part of its platform. In particular, Mr. Carney promised to build infrastructure, including AI data centres, communication networks, and a digital supply chain, that would support this industry. Foundational to all of this is developing Canadian sovereign capacity that also benefits the Canadian economy. However, what was lacking from the platform was any commitment to building cloud data centres. This was likely not at the top of Prime Minister Carney’s thoughts despite the Business Council of Canada and others advocating for greater investment in both Canadian sovereign cloud and AI. I even had an article get quite popular talking about the risks of reliance on US-based hyperscalers and cloud computing under the US CLOUD Act. And Canadian public servants are increasingly concerned with data sovereignty and how to protect the government’s data. Suffice to say, there are many reasons south of the border that are leading Canadians and the government to recognize the rising political and economic risks of relying upon American AI and hyperscaler firms.
As a result of these increasing voices, it came as little surprise that during the announcement of the Major Projects Office, Prime Minister Carney stated:
“Finally, we will task the Major Projects Office with assisting in the development of a Canadian Sovereign Cloud. This would build the compute capacity and data centres needed to underpin our competitiveness, protect our security, and boost our sovereignty. This will give Canada independent control over advanced computing power, while reinforcing our leadership in AI and quantum.”Although significant details are lacking, Prime Minister Carney’s statement that the Major Projects Office will support cloud and sovereign cloud sends signals that the Prime Minister’s Office and cabinet may not yet realize that this potentially runs counter to their stated plans on AI. The investment and development of sovereign AI and cloud, particularly in relation to cloud and AI data centres, will have significant effects on each other, as they both share overlapping supply chains and infrastructure. Despite this, it must be stressed that AI and cloud data centres are not the same and to approach investment into non-descript “data centres” with the idea of helping both sovereign AI and cloud is unlikely to help either.
AI and cloud data centres are not the same, and trying to sell Canada on doing both is either naive or misinformed. Potentially both. Information from the Prime Minister’s Office and his cabinet is sparse regarding sovereign cloud at this time, but a significant amount of work has already been invested in AI, including the Canadian Sovereign AI Compute Strategy.
But first, what is the difference between AI and cloud data centres?
AI vs Cloud Data Centre - What is the Difference?
I am not being picky here about details, but such ambiguity is already causing confusion. A recent Betakit article appeared to suggest that cloud and AI data centres functionally serve the same purpose, or that cloud data centres “powers computations, including AI training and reasoning processes, through servers house in data centres throughout the world.” While cloud data centres can be used for these purposes, they are not efficient for this task, which is why specific AI data centres are constructed to handle the significant computing required for AI workloads and processing. This confusion about the types of data centres is fueled by the government’s mixed messaging, which potentially sends the wrong signals to Canadians and industry about what to expect from the federal government concerning digital infrastructure developments. I can only assume that the federal government does indeed understand the difference between these types of data centres, but if they do not, this means the federal government is making major decisions without understanding its impact.
When we talk about data centres, most of us picture massive, nondescript buildings filled with rows upon rows of racks, each top to bottom with servers containing storage drives to host vast amounts of data. Often overlooked is that such data centres still require high-performance CPUs with substantial processing power to handle the data, along with high-quality error-correcting RAM to minimize the risk of data corruption. A significant amount of infrastructure accompanies this, including cooling systems and the power required to run all these power-intensive activities. All of this means that electrical costs are among the largest operating expenses for data centers; however, this primarily refers to traditional or cloud-based data centres. AI data centres are much, much worse.
Traditional or cloud data centres are predominantly concerned with data storage and handling, as well as ensuring the integrity and security of that data. AI data centres are concerned about the processing of that data, and are comparatively less concerned with data storage at rest. One of the most significant differences is that AI data centres make heavier use of graphics processing units (GPUs) for additional computing power that is required to train AI models and process GenAI systems. Despite their name, GPUs have been used to provide additional computing power across a range of activities for many years now, which is why Nvidia has skyrocketed to being one of the most profitable companies with its dominance in building GPUs specifically for AI processing.
While there are many nuances to consider when it comes to data centre usage, particularly in the context of virtualization, the bottom line is that the hardware and power requirements of AI data centres are significantly different. As a result, the Government of Canada cannot hedge its bets and invest broadly in “data centres” with the expectation that it will support both sovereign cloud and AI. While cloud data centres can indeed support AI by providing the data storage needed to train AI models or other related activities, these are still distinct activities that rely upon different systems to handle significant amounts of data storage versus those that can process that data. One of the easiest ways to understand these differences is to look at the power demands of cloud data centres versus those of an AI data centre.
1.21 Gigawatts!? - The Power Requirements
The actual power and costs of cloud data centres can vary based on size and location, with local electrical costs often playing a significant role in where data centres are constructed. As a result, rather than focusing on direct costs, many focus on the power demands of facilities and the servers. Regardless of the type of data centre, there is a standardized format for servers and server racks used worldwide. This allows us to readily compare AI and cloud/traditional data centres based on their approximate power use per server rack.
Average cloud data centre power consumption
Several estimates of the average power consumption of cloud data centers exist, so I have taken an approximate mean of these reports to provide a rough idea of the power consumption of cloud data centers.
On average, a single server rack requires 7-10 kW of power. In full facilities, this can equal approximately:
Small: 5-10 MW
Medium: 10-70 MW
Large/hyperscaler - 70 - 100+ MW
In 2023, on average, data centres accounted for approximately 1.5% of global electricity used, or nearly 50 gigawatts. This is expected to grow exponentially to meet the demands for AI compute capacity.
Average AI data centre consumption
On average, a single AI server rack requires ~17 kWs, which is expected to rise to 30 kWs by 2027. However, these rates often apply to much less intensive operations, and the power requirements can increase significantly. One such example is that the training of models, such as those for ChatGPT, can consume as much as 80 kW per rack. Server racks with NVidia GB200 chips could reach as much as 120 kW. This makes it clear that we are not talking about double or even triple power demands, but rather exponential power demands that countries are struggling to meet.
This provides a brief snapshot of the significant additional energy requirements and the distinct nature of AI-specific infrastructure. This is not just double the power requirements; we are talking about exponential growth in power demands. Researchers estimate that one of the popular large language models, such as ChatGPT, requires five times more energy than a traditional internet search. This is one of the reasons why many advocate for more caution, as the five times increase in energy demands does not always equate to a five times improvement in output, productivity, or quality.
As a result, it is important to ensure that the deployment and investment into AI and machine learning systems are done thoughtfully and holistically to ensure that they meet the objectives of growing Canadian sovereign AI and cloud capacity. There is a risk if the Government of Canada confuses or miscommunicates about AI and cloud data centres it may have a negative impact on both industries. Although I will not get into all of these dimensions, there are unique political and economic constraints and considerations for both supply chains that cannot be equally met with a de facto broad “data centre” approach.
Canadian Sovereign AI Compute Strategy
The Government of Canada’s Canadian Sovereign AI Compute Strategy implicitly acknowledges the differences between cloud and AI data centres despite the current mixed signals. The strategy commits approximately $2 billion to support the growth of access to computing power, which in many cases means access to AI data centre computing power. In particular, the strategy acknowledges the immediate and long-term need for computing power. Any current investments in building AI infrastructure will take years to provide benefits, so the strategy aims to provide investments for Canadian firms to purchase access to computing power and also invest in the long-term growth of Canadian computing power. The strategy provides:
$700 million for “mobilizing private sector investment” through the AI Compute Challenge, which is meant to increase domestic capacity for “commercial AI-specific data centres” and support the broader AI ecosystem.
$1 Billion for “building public supercomputing infrastructure” to build
This includes $705 million for the AI Sovereign Compute Infrastructure Program and $200 million to help address immediate needs.
$300 million for the AI Compute Access Fund to “support the purchase of AI computer resources by Canadian innovators and businesses.”
This strategy shows that the government at least understands the key needs of AI computing and the differences from cloud data centres and cloud computing generally. The strategy was broadly well received and many credit good leadership and consultation that went into the strategy. Although the strategy was released under the Trudeau government, a lot of the same people that developed the strategy are still in government and Carney’s government is currently looking to do even more in AI based on the groundwork done by the Trudeau government.
As a result, this likely indicates the current mixed messaging related to AI and cloud data centres is likely attributed to a lack of planning or policy direction from cabinet. There was a significant amount of work done on AI already and it formed a centerpiece to Carney and the Liberal Party’s platform, but cloud did not receive much attention. Calls for data sovereignty and sovereign cloud have only intensified since the beginning of January 2025, but their focus up until this point has been on AI, AI data centers, and the AI supply chain. As a result, there is currently a lack of information on exactly where the current government stands regarding sovereign cloud and related investments. Policy statements and positions up to this point increasingly fall back on phrases which claim such data centres support soverign AI, cloud, and quantum. Throwing out buzzwords to signal your intentions can only get you so far and an absence of more information means Canadians and industry are left grasping at the limited information provided for any indication of policy direction.
It is easy for Minister Evan Solomon to speak to AI, but as the Minister of Artificial Intelligence and Digital Innovation, this likely means that sovereign cloud data centres would also fall under his mandate. However, the Major Projects Office, which operates under the Privy Council, is already being provided direction to work on sovereign cloud. This means that the President of the Privy Council, Dominic LeBlanc, and the CEO of the Major Projects Office currently have the most influence on the direction of sovereign cloud policy and investment. Ultimately, it is unclear to what degree the government is looking at growth of cloud data centres independently of AI data centres beyond the AI Compute Strategy. As the term “sovereign cloud” will increasingly become politicized, there are many reasons the government needs to provide better messaging and consultations on sovereign cloud.
Major Cross-Project Efficiencies
Further, despite the shared infrastructure that can support both types of date centres, it is unclear to what degree the government is seeking to align such activities. The potential for this is seen with one of the projects selected for review by the Major Projects Office. The project in particularly chosen for further review by the Major Projects Office is the Darlington New Nuclear Project, which would aim to create an operational small modular nuclear reactor. The Darlington New Nuclear Project is already under construction in the municipality of Clarington, Ontario, but this technology or any future power production projects should be paired with a project to build a sovereign AI or cloud data centre. Small modular nuclear reactors already have significant potential to revolutionize nuclear power, which can be paired with the expansion of AI and cloud data centres in Canada to position Canada as a destination of choice for green energy for data centres.
In 2024, the Canada Energy Regulator (CER) released a good Market Snapshot on the growing energy demands in Canada associated with the rising demand for AI. They note that despite significant advancements in energy efficiency, these efficiency gains have slowed and may be insufficient to offset the greater demands for AI unless equal effort is placed in improving AI efficiency. Approximately 60% of Canada’s electricity was generated by hydroelectricity, and over 80% of electricity comes from non-emitting sources. This low carbon footprint and high renewable energy use are marketable, supply chain advantages. However, despite this strong position, there is a risk that this will not last long if Canada does not scale its power appropriately alongside investments in Canadian AI and cloud data centres.
So What? - What’s the defence angle?
The federal government’s current messaging appears to be that it wants to have its cloud cake made with AI and eat it too. While it is feasible to focus on investments and policy to support sovereign AI and cloud, it cannot be done without specific concern to dual use infrastructure that can support both versus what are the unique to AI and cloud respectively. The economic and political needs to support the growth of sovereign AI and cloud in Canada are not the same and policies need to appropriately reflect this.
The Government has already committed billions for AI compute capacity alone and to develop a similar capacity for cloud data centres could be as expensive. It is unlikely the government will invest a similar amount into broad Canadian cloud data centres, but will be focused on sovereign cloud to meet specific needs such as defence.
Both cloud and AI are central to the future of warfare for the Canadian Armed Forces (CAF). The Department of National Defence and CAF (DND/CAF) AI Strategy is unambiguous about the defence imperative for AI. DND/CAF views AI as foundational to digital modernization and necessitates a need for the CAF to be an “AI-enabled organization.”
The latest Canadian Army Capstone Operating Concept cites AI as a critical tool for use across the Canadian Army and also identifies it as a potentially disruptive technology for adversaries against Canada. There are growing economic and national defence demands for secure cloud and AI. Growing Canadian sovereign AI and cloud is not just about growing Canada’s economy or jobs, but is also about ensuring a reliable, secure capacity for AI and cloud that will server Canadian needs first over others.
To have a sovereign capacity for AI and cloud means more than having access to such technologies within the borders of Canada. Sovereign AI and cloud can mean many things, but fundamentally is about ensuring Canadian access to these technologies free of the potential risks of foreign interference. Canada could invest in many firms that are dedicated to developing AI or cloud applications and software, but if Canada does not have the infrastructure to sustain and support these, then any insistence on sovereign AI and cloud are fundamentally flawed. Sovereign AI and cloud mean more than just than having physical capacity in Canada, but about ensuring that such capacity and access is aligned with Canadian interests.
Your company’s logo could be here!
Interested in advertising on Canadian Cyber in Context? Check out our sponsors page.