By Hessie Jones
In Canada, the requirement for a modern energy system is met with more urgency today. The energy demands fueled by artificial intelligence and the appetite for data centers cannot be satisfied with our current energy infrastructure. By 2050, the demand for electricity will increase by 75%. 15% of this increase is attributed to compute powered by data centers.
This momentum to build an independent, more sovereign Canada, has resulted in strong discussions about what independence really means and to what extent we need to ready ourselves to meet the demands of our future, without a willing and steadfast neighbour. For Canada how we can grow GDP and sustain an economy with our own resources and new global partnerships? A recent Toronto Board of Trade discussion tackled this issue.
Canada, the Growth of Data Centers and the need to Expand Energy Infrastructure
Chuck Farmer is the Chief Energy Transition Officer at the Independent Electricity System Operator (IESO), which balances the grid between demand-side energy and procurement, and plans energy needs of the province. This 75% growth, as per the chart, has been driven by population growth, the emergence of the digital economy and decarbonization goals. The demand growth ‘had’ been 60% as of 2023, however the rise in data centers accounts for the incremental forecast to 75% by 2050.
The growth of the data center market is expected to grow at a compound annual growth rate of 16%, with AI-focused data center demands “growing significantly faster at 50% CAGR through 2030. “According to the International Energy Agency (IEA), in 2022, [data centers] consumed an estimated 460 terawatt-hours (TWh) globally, or roughly 1.4-1.7% of global electricity use…. equivalent to 71% of Canada’s electricity generation.”
Farmer suggests that 1800 data centers are expected to be built in Ontario in the next 10 years and to support that demand, new generation and storage injection into the grid will need to be in place by 2028. He points to the refurbishment of nuclear stations, currently underway, and the need for additional nuclear facilities to support the load. The transmission projects in Ontario will supply energy needs to areas in northern and southern Ontario with interconnections to Manitoba, Quebec, New York and Niagara grids.
Data Centers Now Compete with Cities for Energy
To put things into perspective, one server can consume 100-600 W of power. “A fully populated server rack, housing 42 1U servers, can consume anywhere between 4 kilowatts (kW) and 25 kW of power.” The largest data centers — called hyperscalers — in comparison, will have between 10k and 100k servers, will be more power hungry and can consume over 100 MegaWatts (MW) of power.
Farmer points out that the largest customer is the Greater Toronto Are (GTA) which currently can consume 300 MW of power. This enormity means that for just one hyperscaler this is the equivalent to power in one metropolitan area like Toronto. At this scale, the anticipated 1800 data centers, mentioned earlier, must include strategies for efficiency and renewable energy use.
Heidi Bredenholler-Prasad is the VP Commercial Strategy and Business Development at Enbridge Gas. She argues that the company is not only a stop gap but provides the reliability long term for data center power needs — immediately, and as backup transmission to other energy sources. Andy Fenton is the VP Sales & Marketing at Telehouse Canada, which provides colocation and connectivity solutions in data centers. He reasons that these data centers are strategically located. His portfolio of clients is in downtown Toronto — connectivity is the closest point to the population.
The specific needs could range from crypto exchanges, which can accommodate intermittent energy; battery plants, and EV manufacturers to start. However, Fenton is clear that reliable uninterruptable energy will be required to power AI as inferences become the norm. He emphasizes that until two years ago, inference cloud players did not exist. LLMs have enabled inferences in real time, processes leveraging trained models to make predictions or generate outputs based on new and unseen data. The compute required to process the query to generate the results needs to happen at the edge of the device.
For Pat Dalzell, VP, Corporate Affairs at Bruce Power, the return on investment is clear. Growing demand will accelerate options for large-scale power generation — options which are urgent to keep the lights on. However, he emphasized that both reliability and affordability are necessary. Demand is an unknown factor that will determine the level of affordability. For context, current demand per server rack = 130kW. He contends that for Nvidia, a 700kW demand is ‘mindblogging.’ Attempting to develop for this level of demand is impossible. The needs of AI today did not exist two years ago and there was no feasible way to accurately anticipate the impact of changes on infrastructure.
Who will bear the cost?
Ultimately Dalzell points out that supply and demand will bear out cost, while they plan the power needs in Ontario. Farmer adds that they will have a better idea of demand over time, but they must update their decisions to manage the risk ahead. While the nuclear investments in refurbishments have gone ahead, the decision whether to build them is still pending. Currently there is competition for available power but to support population growth in the next decade, power will be a scarce resource until the larger power infrastructure is in place.
Bredenholler-Prasad of Enbridge addresses the need for policy and industry collaboration to establish alignments, especially for economic development. Farmer, however, is skeptical that regulation will play a role in the short term. In the meantime, they realize that it will take time to plan, review, and get those plans commissioned.
The key for these energy providers is to ensure that they are able to meet the demands of the system. However, if demand paces above forecast there will be undue stress on the available space. Fenton, of Telehouse intimated that the data centers demand the necessary energy, and likely will be inflexible, as he states, “Data load systems are running towards peak load while models continue to churn… 27/4 and will not settle down.”
He argues that data centers will be unwilling to make agreements that would compromise their energy needs: “Workloads can go anywhere so these companies will go to where they can get power to consumers.”
Ultimately, as per Dalzell of Bruce Power, “We don’t want to challenge the system to drive up the cost to the consumer. We don’t know where the load is coming from but there will be substantial demand in the future — that is clear. We’re looking at our site as a potential demand center and the potential for these demand centers will evolve with the growth of the data centers.
The question of sustainability and how an energy infrastructure that inevitability will put more stress on our resources, impact our environment and net zero emissions needs to be addressed. Dalzell drew attention to Canada’s seasonal shifts and suggested the need for flex demand for seasonal structures, especially in hotter weather — to lower the cost of the system. With the global investment favouring development of AI infrastructure, it is unlikely that data and inference center cloud players will be willing to compromise. Farmer, however, impressed upon the need for data centers to change the operating reserves. If there is less compromise, he asserted that perhaps AI will need to modify processes to fix its own problem. Data centers are leapfrogging their own growth and at the same time, decarbonization goals will mean moving away from the reliability of natural gas, especially in new developments.
Dalzell of Bruce Power noted there are duplicative processes where they can work with government regulators to streamline and speed up the process. In addition to the nuclear refurbishment, they are recapitalizing the supply chain to optimize the buildout of the refurbishments. Farmer added that it takes less time to deploy efficiencies and that should be leveraged in the short term. And while it is possible to add demand and load without increasing rates, the minute you build infrastructure to support the added demand, the cost to the end consumer will increase What Fenton emphasizes is the need for a common voice to represent industry to mobilize on the common issues.
What is clear is the tension between the demands of the energy system and Canada’s mandate towards its Net zero emissions goals. With AI being the catalyst that fuels every aspect of society’s digital future, as Farmer put it, those powering these AI systems must also address these energy requirements and how to minimize the environmental impact and impending resource constraints. This will be critical as the climate crisis becomes the near-term threat and longer-term existential risk to human life.
Altitude Accelerator’s sweet spot is helping startup companies within the environmental and cleantech space. Our entrepreneur-in-residence, Bryan Duarte is a seasoned professional focusing on clean technology. Hear from Bryan Duarte in our post, Building Cleantech Success in Canada for Startup VISA Founders and Empowering Change: Shaping the Future of Venture Capital through Diversity, Equity, and Inclusion. If you are interested in learning more, please contact us info@altitudeaccelerator.ca.
