Image credit: Chad Davis/Flickr
It’s now impossible to avoid hearing about “artificial intelligence” (AI), a broad category of automation in which computers process large amounts of data to detect patterns, build models, and complete tasks without specific programming (often referred to as “machine learning”). AI is most commonly encountered through chatbots (like ChatGPT and Gemini) and image generators (like Midjourney and DALL·E), but is also being applied at far larger scales, such as in the oil and gas industry and military operations.
The Manitoba government is trying to attract AI data centres, with Premier Wab Kinew himself showing a particular interest in AI. Last October, Premier Kinew pledged that “Manitoba is ready to punch above its weight when it comes to the new AI frontier” and that “you’ll see servers and data centres in Manitoba in the future,” following the release of a report by Manitoba’s Innovation and Productivity Taskforce.
But the new “AI frontier” poses huge threats to Manitoba’s climate and environmental commitments. Recent polling showed strong opposition to AI use and data centres in Manitoba, with two-thirds of respondents expressing concern about their environmental impacts. This article provides a brief overview of Manitoba’s pursuit of AI data centres and the risks they pose, particularly regarding electricity.
What Makes AI Data Centres Different
First, it’s important to distinguish between different kinds of data centres. Data centres have existed for decades as the physical backbone of the internet, including cloud computing and storage, video streaming and image sharing, online shopping, social media, and search engines. As of mid-2024, there were more than 7,000 data centres built or under development worldwide.
There are currently 10 data centres in Manitoba—nine in Winnipeg and one in Winkler—most of which are owned by local or Canadian companies such as Les.Net, Bell MTS, and Manitoba Hydro. (The exception is one facility owned by California-based data centre giant Equinix.) None of these, however, are the enormous “hyperscale” facilities being built at a feverish pace by Amazon, Microsoft, Meta, and Google—and that are triggering widespread controversy and protest around the world.
There has been an explosion in the number of new data centres being built specifically for AI training and operations. Unlike conventional data centres, AI data centres use advanced graphics processing units (GPUs) and other AI accelerators that are much more energy-intensive, consuming about 10 to 15 times as much electricity as conventional units.
AI Data Centres Proposed for Manitoba
Jet.AI and Consensus Core recently proposed a massive new data centre just south of Winnipeg. Powered in part by a jaw-dropping six fossil gas-fired turbines—which would make it one of the largest polluters in the province—this AI data centre would occupy 350 acres of formerly agricultural land in the Rural Municipality of Ritchot. More than 4,200 people have signed an online petition opposing the project.
The seriousness of the proposal is difficult to assess because publicly available information beyond the project proponents’ corporate documents is limited. Jet.AI is a penny-stock company that claims to specialize in “private jet charter artificial intelligence” (and has two former senior Israeli Air Force pilots on its board). However, the other project partner, Consensus Core, has deployed a team of lobbyists from Ottawa-based StrategyCorp (including former NDP MP and BC MLA Nathan Cullen) to move the project ahead, meeting with political heavy-hitters including Finance Minister Adrien Sala, the premier’s chief of staff, and both the chair and president/CEO of Manitoba Hydro.
Even if the Ritchot project doesn’t proceed, AI data centres are likely to set up shop in Manitoba. The Province recently introduced legislation to give Manitoba Hydro greater ability to reject large load additions to the grid, seemingly to anticipate potentially undesirable data centre applications. On the other hand, Premier Kinew and Minister Sala continue to advocate for “sovereign data and AI” and the supposed job creation these projects will bring. Manitoba’s position seems aligned with the federal government’s recent offer of government support for “sovereign, large-scale AI data centres with total planned capacities greater than 100 megawatts.” Premier Kinew has also had several meetings in the last year with representatives of Bell Canada, which is building data centres across the country and could be a likely proponent for one in Manitoba.
The main reasons Premier Kinew has cited for wanting to bring AI data centres to the province include data control, job creation, and near-ideal conditions (including cold winters that can reduce cooling demand and low-emissions hydroelectricity). The provincial Innovation and Productivity task force has also called for “sovereign strategic investment and governance” of data and AI so that “Manitoba can lead by owning the underlying data assets, setting the rules for their use, and capturing the economic returns here at home.” Notably, this report did not advocate for options such as public data trusts.
However, there are countless reasons to oppose the development of AI data centres in Manitoba. First, there are the direct impacts of data centres: significant local impacts such as noise and air pollution; minimal long-term job benefits or regional economic spinoffs; the looming e-waste challenge; and potentially massive freshwater consumption (newer cooling systems require much less water but are far more expensive). Then there are the broader societal risks and impacts of AI: the destruction of jobs; potentially calamitous effects on childhood cognitive and social development; the lack of safety regulations to prevent violence and exploitation; constant copyright violations and privacy issues; the concentration of power by a handful of AI monopolies; and much, much more.
The question we’re most concerned with, however, is the intense demand that AI data centres will place on Manitoba’s electricity grid and/or the resulting greenhouse gas emissions.
The Massive Energy Demand of AI Data Centres
The hallmark feature of AI data centres is their astronomical energy demands, to the point that the industry explicitly discusses and compares them by power consumption. Much of this demand comes from operating the advanced servers, with additional power needed for cooling, networking, and data storage. Skyrocketing electricity demand from AI processing has reversed recent data centre efficiency improvements, with much of the added demand met by fossil gas- and even coal-fired generation (which requires enormous water inputs for cooling), significantly increasing annual emissions from the biggest corporate players like Amazon, Google, and Microsoft.
The U.S.-based Electric Power Research Institute has reported that a 100 MW data centre consumes as much electricity as 80,000 households, while a 1 GW facility would be equivalent to 800,000 homes. In total, data centres consumed about 1.5% of global electricity consumption in 2024: more than countries such as Indonesia or Iran. The International Energy Agency’s (IEA) base case projects that this demand will double to around 3% of global demand by 2030, or nearly equivalent to the entire continent of Africa in 2024. However, its “Lift-Off Case” predicts that data centre demand could exceed the combined electricity generation of Germany, France, the UK, and Italy in 2024.
While the latter scenario may seem far-fetched, tech companies are directly integrating AI into everything imaginable—search engines, email accounts, software, social media, and devices themselves—and “leading labs are racing us toward a world where AI ‘agents’ perform tasks for us without our supervising their every move.” Newer models are also likely more energy-intensive than previous iterations; video and image generation is also especially bad on this front.
The IEA has noted that electricity demand growth from data centres in its base case will account for only about 10% of global growth by 2030, far outpaced by the electrification of industry, transportation, and buildings. However, this rate of increase varies significantly depending on where in the world we’re looking. In the US, this growth will account for almost 40% of the increase in electricity demand between 2022 and 2030, reaching 8% of the country’s total load. AI processing will account for an ever-increasing share of data centre demand. Goldman Sachs expects the AI electricity demand to increase from 14% to 27% of total global data centre demand between 2025 and 2027. Whether using existing or new electricity generation, transmission, and storage infrastructure, all of this represents power that countries cannot use for alternative purposes.
Data Centres and Manitoba’s Electricity Supply
In Manitoba, we must consider all AI data centre proposals in the context of a looming peak electricity crunch. According to Manitoba Hydro and the government, the province will need much more electricity in the coming years and decades, especially to ensure sufficient power for electric space heating (which accounts for about half of all building heating in Manitoba). Hydro’s overriding concern is meeting peak electricity demand during the coldest and darkest days of the year, hence the supposed need for the proposed gas-fired power plant in Brandon. But various demand response measures and energy storage tools may help smooth these peaks throughout the day and minimize periods of extreme grid stress.
There may be potential for Manitoba Hydro to add AI data centres to its Curtailable Rate Program, which essentially pays industrial power consumers to reduce their electricity demand during peak periods. But this option is unlikely, given that data centres are marketed based on “minimum uptime guarantees” of being online 99.9% or even 99.99% of the year; further, AI tools and agents are increasingly integrated into complex systems, including the electrical grid, making them hard to curtail. Given that new data centres will push to operate 24/7 to maximize service reliability and profits, their operations would increase the burden on the provincial grid and/or require company-owned on-site generation. In both cases, these peaks would likely be met with emissions-intensive gas-fired power, compromising Manitoba’s emissions reduction goals.
Even if the Province somehow forced data centres to participate in the Curtailable Rate Program, they would still likely add significant demand to the grid. A recent Jet.AI corporate presentation described its proposed data centre location south of Winnipeg as a “‘Goldilocks’ site for data center development” due to its “unique confluence of abundant hydropower and natural gas capacity,” adding that Manitoba Hydro’s supply is “cost-effective, reliable, abundant” and “ideal for sustainable data center operations.”
Such projects would consume significant amounts of power that Manitoba otherwise needs to electrify transportation, buildings, and industry, and could risk driving up costs for ratepayers. While on-site generation wouldn’t have the same effect on grid capacity, it could well have knock-on effects by driving up the costs of the materials, labour, and fuel needed to build new generation capacity. Combined, this would mean higher costs, more emissions, and slowed electrification of Manitoba—all to line the pockets of AI companies.
What’s the Path Forward?
The first thing to come to terms with is that, contrary to the claims of tech billionaires like Sam Altman and Elon Musk, AI is not inevitable. Nor is AI data centre development in Manitoba. Governments are under no obligation to concede to the fantastical claims of AI companies about the supposed benefits of big warehouses of advanced servers. While there may be some merit to having increased domestic storage of sensitive data, the same does not hold for AI processing.
Some jurisdictions have limited the amount of additional electricity demand that new AI data centres can consume. Other analyses have recommended developing regulations mandating minimum efficiency and sustainability standards, renewable energy requirements, and restrictions on the use of AI in oil and gas extraction.
While these are worthwhile policy considerations, Manitoba could take a stronger position by banning AI data centres, as is being pursued in US states including New York and Georgia. Tech giants have made clear that they will vehemently fight regulations, such as the recently defeated effort in Washington State to “pay additional utility charges, comply with clean energy requirements, and shut off power at times of peak demand on the grid.” It is far easier to refuse such companies access to the province now than attempt to force them to accept profit-squeezing regulations after the fact. A ban on AI data centres could include exceptions for smaller-scale conventional data centres specifically designed to house sensitive data.
At minimum, we would hope to see the province put a moratorium on all new AI data centre development until Manitoba Hydro’s long-term energy planning and the Province’s net-zero action plans are complete, with a solid understanding of how to meet growing demand for electrification across sectors. Governments should also not provide any public money or tax incentives to AI data centres, given the tremendous costs to communities and the province more generally.
Rather than hanging Manitoba’s economic future on the highly dubious prospect of AI data centres, a far better path forward would be economic development kick-started by big public investments in climate solutions, funded by progressive tax reform: for example, replacing electric resistance heating in rural areas with much more efficient ground-source heat pumps. Unlike AI data centres, such investments would significantly reduce peak electricity demand, create good, permanent jobs, reduce provincial emissions, and improve thermal comfort and affordability for Manitobans.