SAIG founders Sasha Luccioni and Boris Gamazaychikov at the Insectarium de Montréal on April 23.Clara Lacasse/Supplied
Everything about artificial intelligence is big these days, from the models trained on huge quantities of information pulled from the internet, to the data centres and energy infrastructure powering it all.
But Sasha Luccioni and Boris Gamazaychikov argue that’s not the only approach. The focus on scale in AI today comes with trade-offs like surging electricity consumption and a reliance on fossil fuels to get data centres online quickly. “The trajectory AI is on today is unsustainable from an environmental perspective,” Mr. Gamazaychikov said.
The pair are launching a new research and consulting firm in Montreal called the Sustainable AI Group (SAIG). The goal is to advise large companies on AI adoption through an environmental lens and use data to inform decision-making, building on Dr. Luccioni’s research into the energy and climate effects of the technology. By equipping companies with knowledge, SAIG hopes that will put pressure on large AI developers to adopt more sustainable approaches, too.
Opinion: Canada built the foundation of AI. Now let’s own what comes next
“If a company truly wants to understand and decarbonize AI, they need more information about the different use cases,” Dr. Luccioni said, “and currently that doesn’t exist.”
Since earning a PhD in AI from the Université du Québec à Montreal in 2016, she has become one of the world’s leading experts on the energy and environmental footprint of AI. This month, she stepped down as the AI and climate lead for New York company Hugging Face, which provides tools for building machine learning applications. While there, she launched a program to score AI models based on energy efficiency. Mr. Gamazaychikov, meanwhile, was most recently the head of AI sustainability at Salesforce Inc. He will serve as chief executive officer while Dr. Luccioni assumes the role of chief scientific officer at SAIG.
The two said that the timing is right for such an effort, and that the focus on AI and sustainability is unique among advisory firms.
Electricity demand from data centres surged by 17 per cent last year, according to the International Energy Agency, outpacing overall global consumption, which grew by 3 per cent. The power consumption per AI task is falling at an unprecedented rate, the IEA noted, but use of the technology is growing overall. That’s why the agency estimates that electricity demand from data centres will double by 2030.
How a Toronto AI startup hopes to make the keyboard obsolete
As AI adoption grows, many companies in the United States and Canada are turning to natural gas to power data centres, which emits carbon. “This is not going in the right direction,” Dr. Luccioni said. “That doesn’t make much sense because in terms of new technologies, they’re supposed to be green.”
Big technology companies like Microsoft, Google, Amazon and Meta have extensive programs to address their environmental footprints through efficiency improvements, carbon offsetting and water reduction initiatives. But the AI race is compromising these goals. Microsoft’s total carbon emissions have risen by 23 per cent since 2020, “due to growth-related factors such as AI and cloud expansion,” according to its latest sustainability report. Meta’s emissions have surged by 64 per cent over the same time period.
AI model developers such as OpenAI and Anthropic, meanwhile, are focused on building large, general-purpose models that can complete a wide range of tasks. But these models can be more energy-intensive and expensive than deploying small, tailored versions. “You don’t need that kind of model for a lot of the enterprise tasks,” Dr. Luccioni said. “You don’t need a Swiss army knife. You need a specific wrench.”
That concept still hasn’t fully spread among companies adopting AI. “They’re taking the Big Tech playbook at face value without a more nuanced approach,” Mr. Gamazaychikov said.
The same is true of data centres, as some developers are prioritizing speed-to-market over sustainability. The government of Alberta, for example, is trying to attract developers to take advantage of the province’s natural gas reserves. Bell Canada is building a 300-megawatt data centre complex in Saskatchewan, where the energy grid is predominantly fuelled by natural gas and coal power.
Provinces that have traditionally housed data centres and rely on renewable energy, such as Quebec, Ontario and British Columbia, are facing supply constraints. Still, Dr. Luccioni said Canada should think strategically about how to allot available clean power to new developments, rather than build quickly with fossil fuels. Telus Corp., for example, said earlier this week that its data centre expansion will be powered with 98 per cent clean energy from BC Hydro.
“The question we’re asking is, how do we do this as fast as possible? Given we’ve painted ourselves into that corner, natural gas seems to be the decision,” Dr. Luccioni said. “But if we asked ourselves some different questions, we would have different answers.”