AI technology is ever-expanding, and so are the large-scale data centres that keep them running. The resources it takes to fuel these centres call for vast areas of land — and many countries are signing up to construct new AI data centres, as the market for them is expected to grow from $236.44 billion USD to $933.76 billion USD by 2030, according to Yahoo News. This includes Canada, which is far behind the thousands of centres planned or already built in the US, and another oncoming powerhouse of AI data centres — China.In this article, the Western Standard will reveal the number of large-scale AI data centres being made or already exist in the three countries.However, prior to this, it is important for readers to understand how exactly these centres work, and why they require so many resources to build and maintain them..AI centres are multi-purpose, used to program and keep AI models running.They are unlike regular data centres, which can be located in city centres, since these do not require large amounts of resources, and are often used for running things like business software, web pages, and storage files, among other computing jobs.AI data centres, on the other hand, "...basically need ... a lot of cheap land so you can imagine this is going to be far away from city centres," stated Lyndsey Rolheiser, an urban economist and assistant professor at the Schulich School of Business at York University, to CBC News. "They need access to energy and access to water so this is kind of pushing them out into the outskirts within [Canada's] provinces.” .As explained by Medium, this is because AI centres are running large neural networks, that are connected to one another.Regular data centres do not work like this — their software has discrete tasks that are often not connected. AI centres run on a higher-performing software, using graphics processing units (GPUs), instead of central processing units (CPUs), which normal data centres use. CPUs are like generalists, who easily manage a list of instructed tasks.However, CPUs are incompatible for AI programming since training AI requires performing the same task across enormous amounts of data and requires a different type of chip, GPUs, which in analogous terms, are designed to work like an army of identical workers. .GPUs also require far more physical space for their IT infrastructure, including advanced storage, networking, energy, and cooling needs, which explains the centres' larger scale.With AI's requirement for powerful computational software, advanced networking, and vast storage systems, AI data centre consume massive amounts of electricity.AI data centres usually consume three to five times the amount of power that normal data centres do; to put this into perspective — one AI server rack requires 50-150 kilowatts of electricity, compared to 10-15 kilowatts for normal computing. This means the centres are also bound to overheat — necessitating cooling systems.Liquid cooling systems, ie, using water to cool IT infrastructure, are necessary for AI centres because the heat generated by electricity usage is so intense that air based cooling is not effective..This means centres often need to be near a large water source.Not only is a large amount of land required, but because of the amount of electricity needed — cheap land is preferable.Now for what readers came here for, starting with Canada which currently has five AI data centres in their arsenal.Reported by CBC News, two of them are located in BC, one in Ontario, Quebec, and New Brunswick..According to York University research, there are a whopping 96 AI data centres being planned throughout Canada.Alberta will be home to 90% of these projects once built, though not all are confirmed to be built, but many are considering Alberta to be the place to do it.This is because unlike many other provinces, Alberta does not have a regulated electricity market, and with its added cheap gas-powered electricity, and a government that hopes to grow the sector by fast-tracking many of the proposed AI data centre projects — it has become an attractive prospect. "The most dramatic finding is that the places where energy is relatively cheap to generate, and more generation is happening from renewables, are the places where it is becoming more difficult for data centres to be situated,” stated Rolheiser..This means places like Quebec, BC, and Ontario who have legislated laws restricting electricity grid access to new power users, are making it more difficult for AI data centre projects to even consider locating in those provinces.Previously reported by the Western Standard, one of the latest data centre proposals wish to settle in west-central Alberta, located on Crown Land next to Tourmaline's existing Banshee Gas Plant 40 km south of Edson in Yellowhead County.The centre is being proposed by Emerald Energy Technologies, a wholly owned subsidiary of the oil company, Tourmaline Oil Corp.The company hopes to begin production as early as 2027, however, they are still searching for a tech company to operate the facility. .Alberta AI boom continues as Tourmaline proposes massive data centre near Edson.Next, the US, where according to data collected by the activist, Erin Brockovich, who launched a crowdsourced map tracking all the data centres in the country, there are over 4,000. According to the Guardian, 809 data centres are being planned in the US — though they do not disclose how many of them are AI-focused.As for the companies with the most data centres in the country, these include but are not limited to, Amazon Web Services, Meta, and Microsoft Azure, though not all are AI-centred.As for China, there is no official number published for AI data centres, though according to the China Communications Industry Association Data Center Committee, at least 150 newly built AI data centres were finished and running by the end of 2024. .The one that stands out as the coolest centre being built, an underwater AI data centre with obvious temperature-cooling benefits.Reported by Scientific American, AI data centres usually cycle through gallons of water per day to cool down its software, China is planning on making this system extremely efficient, by putting the data centre in the ocean.The construction for the centre started in June, with a wind-powered underwater data centre located almost 10 km off the coast of Shanghai, one of China's AI hubs.The company whose data centre it is, HiCloud, says after consulting with the China Academy of Information and Communications Technology, the project is expected to use at least 30% less electricity than on-land data centres, thanks natural cooling..Reported by AI Weekly, Beijing is reportedly accelerating roughly $295 billion worth of national AI data centre infrastructure projects to be built from 2028 to 2030. Beyond the announcements of AI infrastructure spending by the Chinese government, something a bit disappointing lurks below.Based on research by MIT Technology Review, the AI large language model (LLM) craze has lost momentum in recent years — along with AI data centre plans.As early as 2024, 144 companies had registered with the Cyberspace Administration of China, the country's internet regulator, to develop their own LLMs.As reported by the Economic Observer, only 10% of those companies were still actively investing in large-scale model training (and the data centres needed to do the training) by the end of that year..China's AI infrastructure is also being used as a tool for boosting politicians' reputations, since China's government is highly centralized, local governments move up through regional appointments, with many local leaders gaining political favor through economic projects that present quick results.So, in 2023, AI was introduced, with companies beginning to partner with local government, to capitalize on the AI trend.AI infrastructure began being used as a way to boost political reputation, justify companies' business expansion, and boost stock prices.With the AI push so strong, many data centres had little regard for actual demand or technical feasibility.This meant many projects ended up being built by executives and investors with a limited knowledge of AI infrastructure, resulting in centres falling short of industry standards..By the end of 2024, China's AI centre boom was now looking bleak, and was serving disappointing results, mainly due to many companies unwilling to rent GPUs, the type of chips used to train AI models.This resulted in an underutilization of Chinese AI data centres, though despite this, in 2025 and 2026, China has announced plans to spend billions on AI infrastructure, which includes, AI data centres.
