
Artificial intelligence has become a big part of our lives. It’s embedded in the new iPhone, we engage it through chat bots, content generation and apps like Instagram and WhatsApp. Canada is a global leader in AI development, with the federal government pledging $2 billion to develop the country’s AI infrastructure.
“AI data centres are driving unexpected and insatiable demand for data centres across Canada and the U.S.,” says CBRE’s Scott Harper, who leads the Toronto data centres group. “So much so that we’re starting to see modular data centre facilities being built to shorten the delivery time.”
AI is a massive energy hog. Generative AI queries use up to five times more energy than conventional browser searches and training AI models is highly energy-intensive. Amid a global energy shortage, the data centre industry faces big challenges in getting the power it needs to keep feeding the AI beast.
Evolving Technology
Soaring demand for data centres and some major limitations to powering them have forced innovation and adoption of new technologies that store and process data. Chips are more resilient, handling more heat and greater amounts of data so that less cooling is required. Racks (the structures that hold chips), servers and other IT equipment can handle higher wattage. A few years ago, one rack could handle about 5kW in thermal loads; these days that figure is closer to 30kW, and even more efficient racks are in the works.
“Having more efficient equipment means we can process larger quantities of data quicker, in the same square footage as before,” Harper explains. “But each rack still produces heat equivalent to lighting several BBQs. So cooling them down requires a lot of energy.”
Cooling solutions have advanced and new data centres are making the switch from air to liquid cooling. Liquid cooling takes up a third of the space compared to conventional air cooling and the use of liquid ensures the racks stay at a constant temperature. This is done through chilled doors or by bringing cool water directly to the chip. Some data centres even use immersion cooling, putting servers into a liquid to keep them from overheating.
Despite these improvements, cooling still makes up 30% to 50% of the power consumption in data centres. “We process much more data than before with these new technologies,” says Harper. “The question is: where are we going to get the energy to power them?”
Energy Required
More power is needed fast, says Harper, but selecting sustainable sources is essential. “Using natural gas would increase the carbon footprint of data centres ten-fold compared to solar power. When one data hall uses more energy than 500 homes, finding green energy sources is critical.”
Renewables may be the answer. Harper is currently helping big power users turn to renewable energy sources by connecting them with solar, wind and hydrogen fuel cell developers that can sell green energy directly for data centre use.
In some cases, data centre users purchase this clean energy during downtimes and store it for later use. This avoids overloading the grid during its busiest times and enables companies to acquire green power at more affordable rates.
However, Canada’s energy demand is so great that power from non-renewable sources might have to be part of the solution. Harper expects to see an increase in natural gas project announcements, especially in Western Canada.
“We’re very aware that the data centre industry is competing for power at the moment and that’s only going to intensify with the rise of AI,” says Harper. “But the industry is changing quickly and working hard to find sustainable solutions.”
Who knows, maybe AI will be able to help with that, too.
Recent Insights
Stay In The Know
Subscribe today and join hundreds of professionals who get the latest blogs delivered straight to their inbox.