Energy demand from data centers is steadily increasing, and AI development is a significant factor

This article was published by the Canada Energy Regulator on Oct. 2, 2024.

There are currently an estimated 239 data centers operating across Canada,⁽¹⁾ and the industry is expanding rapidly.⁽²⁾ Data centers and their data transmission networks consume a lot of energy. According to the International Energy Agency (IEA), in 2022, they consumed an estimated 460 terawatt-hours (TWh) globally, or roughly 1.4-1.7 per cent of global electricity use.⁽³⁾⁽⁴⁾ That amount of energy is roughly equivalent to 71 per cent of Canada’s electricity generation that same year,⁽⁵⁾ and the IEA projects global data center energy consumption will double by the end of 2026. The rapid growth of artificial intelligence (AI) applications, particularly large language models and machine learning, is a significant contributor to this rising energy demand, as AI workloads require substantial computing power. Multiple factors make Canada an attractive destination for data centres:

1. Relatively low electricity prices in some regions
2. Significant renewable and clean electricity resources
3. A relatively cool climate⁽⁶⁾

Figure 1: Data centers in Canadian cities

Canadian Data Center Landscape

Canada’s data center landscape is driven by low electricity costs, particularly in provinces like Quebec and British Columbia, where hydroelectricity is abundant and relatively affordable. The country’s clean energy grid, with over 80 per cent of electricity coming from non-emitting sources, can also make it attractive for companies looking to build new data centers with a low carbon footprint. Additionally, Canada’s cool climate further enhances the energy efficiency of data centers by reducing cooling costs.^(7)(8)(9)

In Canada, utilities and system operators are factoring data center electricity usage into their outlooks. For example, Hydro Québec anticipates an increase of 4.1 TWh in data center demand from 2023 to 2032 in its most recent electricity supply plan.⁽¹⁰⁾ The Ontario Independent Electric System Operator (IESO) lists data center electricity demand growth as a key driver in expected power use in Ontario’s commercial sector.⁽¹¹⁾ The Alberta Electric System Operator (AESO)’s latest outlook includes a High Electrification scenario that includes higher electricity demand from data centers, among other areas of demand growth.⁽¹²⁾

Figure 2: Data centers by province

Canadian data centers are increasingly adopting advanced waste heat recovery and cooling technologies to improve energy efficiency and reduce environmental impacts. For example, at the QScale Q01 data center in Levis, Quebec, nearly 100 MW of energy from waste heat is expected to be redirected to households by the end of 2024.⁽¹³⁾ Liquid cooling systems, which are more efficient than traditional air cooling, are gaining popularity. These systems capture heat generated by data centers and can raise water temperatures from 30-35°C to 70-80°C, which can then be used in district heating networks.⁽¹⁴⁾

Impact of large language models and artificial intelligence on energy demand

Global data center workloads nearly tripled between 2015 and 2019, while energy demand remained relatively stable at about 200 terawatt-hours (TWh) per year.⁽¹⁵⁾ This stability is largely due to continuous advancements in energy efficiency. However, since around 2020, these efficiency gains have slowed, contributing to a rapid increase in energy consumption from data centers.^(16)(17)

Large Language Models (LLMs), such as ChatGPT and Microsoft Copilot, have transformed the field of artificial intelligence (AI). As these models have expanded in size and capability, the computational resources, and consequently the energy required for their training, has grown exponentially. As these models are becoming mainstream, the energy demand for processing individual queries has also surged, and AI-enabled queries generally consume more energy compared to traditional and competing technologies. For example, an average ChatGPT query requires about 10 times as much electricity to process as a Google search.⁽¹⁸⁾ By 2030, AI is expected to drive a cumulative 160 per cent increase in data center energy consumption globally.⁽¹⁹⁾

Training increasingly complex AI models has seen computational energy demands doubling approximately every nine months in recent years, and this trend of growth will likely continue.^(20)(21) However, efforts to enhance AI efficiency, such as the use of specialized hardware and software optimizations, could mitigate some of this growth.