As news of the two looming large-scale data centres coming to 150 West Georgia Street and 111 East 5th Avenue, and the subsequent protests they\u2019ve provoked, continues to make headlines, Vancity Lookout took a step back to establish some basics.<\/p>\n
While the federal government has supported the projects in the name of establishing a stronger foundation for Canadian data sovereignty, critics have flagged concerns ranging from environmental impacts to noise pollution.<\/p>\n
There are still several unknowns regarding the two centres in Downtown and Mount Pleasant. However, there is much context to be gleaned from understanding how these centres work, what sorts of regulations may be needed and how novel what is being proposed actually is for the city.<\/p>\n
Generally speaking, a data centre is a facility designed to process and handle enormous volumes of digital information and house rows of computers for storing, processing, and transferring data.<\/p>\n
Depending on who you ask, there are generally four to six main types of centres<\/a>, including \u201centerprise data centres\u201d owned by a single company and \u201ccolocation data centres\u201d that share space rented by multiple tenants.<\/p>\n There are also \u201ccloud data centres\u201d that rent cloud storage space, generally on a subscription basis to clients, and \u201cedge data centres,\u201d which are smaller and geographically closer to users to facilitate connections that need real-time, rapid connection to the internet (think self-driving cars, hospital equipment, or industry alert systems).<\/p>\n Finally, there are \u201chyperscale data centres.\u201d While there is no single, universally agreed-upon definition of \u201chyperscale,\u201d generally, a hyperscale centre will house at least 5,000 servers<\/a> and is physically very large (upwards of 10,000 square feet). The vast majority of the world\u2019s 700 (and growing<\/a>) hyperscale data centres are owned by tech giants like Google, Amazon, Microsoft and, increasingly in Canada, companies like Bell and TELUS.<\/p>\n It is important to note that not all data centres are built to facilitate AI. Still, the majority of the newer centres (specifically hyperscale data centres) are generally constructed with this technology in mind.<\/p>\n From what we know about the proposed TELUS data centres in Vancouver, they are both hyperscale data centres built for AI.<\/p>\n Well, yes and no. Technically, there are already a number of data centres in and around Vancouver.<\/p>\n Most of these, however, are either not AI-specific data centres, or those that do have AI capabilities are not nearly as large as the proposed TELUS centres will be.<\/p>\n Despite asking several experts, it was difficult to nail down exact specifics on the city\u2019s existing data centre landscape. This is in part because, as Mohammad Shahrad, assistant professor at the UBC Department of Electrical and Computer Engineering, pointed out, not all data centres are publicly announced.<\/p>\n \u201cBusinesses could operate their own data centres,\u201d he said. \u201cBut things at the [scale of the TELUS data centres] require lots of permitting and involve [consultations with] BC Hydro, provincial [and national] governments, and the city.\u201d<\/p>\n To understand the difference between an AI data centre and a non-AI data centre, we need to zoom in on the technology that powers them.<\/p>\n Historically, computers have run on CPUs, or Central Processing Units. Within a CPU, a \u201ccore\u201d is essentially the brain<\/a> of the computer. Each core handles its own tasks and information independently. More cores generally mean a computer can perform more complex tasks.<\/p>\n An ultra-basic personal laptop may have two to four cores<\/a>, though more advanced computers can have upwards of two dozen cores<\/a> that switch between tasks sequentially.<\/p>\n The two TELUS data centres proposed for Vancouver would eventually scale<\/a> to include over 60,000 combined GPUs that consume a lot of energy, generate a lot of heat and require a ton of water for large-scale air conditioning and other cooling systems.<\/p>\n According to the International Energy Agency<\/a>, one large-scale GPU data centre, like many built in the United States, consumes the same amount of energy annually as powering 400,000 electric vehicles.<\/p>\n According to TELUS, all three centres combined will initially draw 85 megawatts of power, scaling to 150 megawatts by 2032. The downtown and Mount Pleasant centres will draw 100 megawatts and 26 megawatts, respectively.<\/p>\n To put this into perspective, a standard household microwave uses about 1000 watts of power. Therefore, the downtown data centre would pull up to the same amount of power as running 100,000 microwaves simultaneously.<\/p>\n TELUS claims that it will use up to 90 per cent renewable energy to power its centres. This is an estimate that CCPA senior researcher Simon Enoch said they can make, in part, because they plan to plug into Vancouver\u2019s existing power grid, which is powered mostly by hydroelectric energy.<\/p>\n From what we know so far, the answer is most likely yes. Depending on the local electricity grid, a data centre may draw power from a different combination of sources. For example, in Alberta, AI data centres draw power from a grid largely powered by fossil fuels.<\/p>\n In part because the proposed data centres will be connected to Vancouver\u2019s already comparably \u201ccleaner\u201d hydroelectric power grid, the downtown and Mount Pleasant centres will be more environmentally sustainable than their majority fossil-fuel powered counterparts.<\/p>\n The TELUS data centres are also advertised as more environmentally sustainable due to their \u201cclosed-loop\u201d cooling system. But what does that actually mean?<\/p>\n The energy generated by the systems at work \u201chas to go somewhere,\u201d said Shahrad, otherwise the metal parts of the computer may overheat, warp, and become physically damaged.<\/p>\n Traditionally, Shahrad said, data centres have cooled the air via HVAC systems to keep server rooms from overheating. They do this much like traditional air conditioning, using water evaporation to cool indoor air while releasing excess heat into the outside atmosphere.<\/p>\n TELUS has instead proposed a \u201cclosed-loop cooling system.\u201d Pipes with cold water flow behind the servers to remove heat from the air and server components. As the servers\u2019 heat is absorbed into the water in the pipes, the hot water is transferred to different areas of the facility.<\/p>\n Ordinarily, Shahrad said, companies might not plan to build in urban areas because real estate costs are higher than in more rural locations. However, he noted a benefit of these urban data centres is their ability to use the captured heat to warm surrounding areas.<\/p>\n \u201cOne advantage of [building in] Vancouver is we already have 20-plus kilometres of underground pipelines for distributing heat,\u201d Shahrad explained that with proper coordination and design, the heat generated from the IT infrastructure can then be fed into district energy systems so that the hot water can be used by neighbourhoods.<\/p>\n \u201cIf you want to make the claim that this is sustainable, this is one of the best ways to reuse heat generated as a byproduct in a meaningful way,\u201d said Shahrad.<\/p>\n Enoch and other experts Vancity Lookout spoke with said they are concerned that these large companies won\u2019t be held accountable, given there is currently no robust legislation in place regarding the regulation and construction of AI data centres.<\/p>\n In December, B.C. passed Bill 31,<\/a> which included limits on AI and other tech companies\u2019 ability to draw from B.C.\u2019s power supply. Though this is a \u201cgood start,\u201d Peter A. Allard School of Law doctoral graduate Oludolapo Makinde said that, across the country, this type of legislation is severely lagging amid the push for AI innovation.<\/p>\n \u201cWe need legislation that looks at the environmental and social impacts of AI data centres and how we can mitigate those risks and harms, rather than putting our head in the sand and pretending that those risks do not exist,\u201d she said, adding that there are few existing regulatory barriers<\/a> for these centres outside of B.C.<\/p>\n \u201cThis stuff needs to be regulated before it\u2019s built\u2026 not after the fact,\u201d Enoch added.<\/p>\n With the rapid development of more AI data centres across the country, Shahrad expressed concern over how replicable TELUS\u2019 comparably sustainable centres may actually be. \u201cA big question is to what extent can future data centres of the same class be built to [be as sustainable as those proposed in Vancouver], because there\u2019s only so much infrastructure for distributing that district energy,\u201d he said.<\/p>\n He added that while the Vancouver data centres will have an advantage in terms of sustainability, operationally, they are less competitive because data centres in cities tend to be more expensive.<\/p>\n \u201cIf you want to sell computational resources that are sitting in this data centre versus another one, in say, rural Alberta, a company wanting to rent some GPU\u2019s would probably go for the cheaper option. They wouldn\u2019t care how sustainable it is unless we have proper regulation, which we don\u2019t have right now,\u201d he added.<\/p>\n Another concern is water consumption. As it stands, Enoch said that big tech companies have not been particularly forthcoming about their \u201cactual hard data on water consumption and other environmental indicators.\u201d<\/p>\n \u201cThe more industry hides information from citizens, the more they think there\u2019s a reason to hide,\u201d he said, adding that a provincial regulator would be \u201cinvaluable\u201d in addressing some of the distrust that springs up around the development of AI data centres.<\/p>\n A recent report by The Logic uncovered 309 data centres<\/a> across the country in total, with many more hyperscale AI-data centres specifically in the works.<\/p>\n Currently, the largest proposed AI-data centre in the country that has broken ground is in Regina.<\/a> It\u2019s a 300-megawatt facility, where Enoch said \u201cthey are digging the foundation as we speak,\u201d but added that it won\u2019t remain the largest facility for long.<\/p>\n Another large-scale data centre<\/a> in Olds, Alberta, that is still in the proposal stage, will be an \u201cabsolutely enormous\u201d 1,000 megawatt facility that will cost roughly $10 billion to construct, said Enoch. Another gargantuan 1.9 gigawatt centre<\/a> for the area west of Red Deer has also been proposed, though it remains unclear whether it will be approved.<\/p>\n Bell has proposed at least seven other data centres<\/a> as part of its Bell AI Fabric plan, including four facilities in Kamloops and Merritt, as well as several others planned \u201cacross the country.\u201d<\/p>\n However, when it comes to promises of environmental sustainability, recent documents obtained by<\/a> Greenpeace Canada show that the federal government is aware of – and even counting on – the largest of the AI data centres increasing demand for fossil fuels.<\/p>\n Experts have flagged concerns around how robust the protections for these centres will be if they have international (specifically, US-based) \u201ctenants.\u201d<\/p>\n