Climate Crunch: The power-sapping AI deepfake menace

Hot takes

➔ Climate literacy for the worried and the hopeless. Ottawa plans to invest14.4 million in 17 projects to boost environmental literacy among young people. It ranges from $1.8 million for BC Parks Foundation to inform students about biodiversity loss and several programs that combine climate science with traditional Indigenous knowledge. We are going to need all of it: A recent study shows Canadian youth are wallowing in “worry and hopelessness” in response to climate change.

➔ Carbon Upcycling is a step closer to cracking the cement emissions challenge. The Calgary-based startup recently broke ground on Canada’s first commercial-scale carbon capture and utilization facility at partner Ash Grove’s Mississauga cement plant. Opening in 2026, the $10-millionproject will capture kiln CO₂ and convert it, along with other industrial byproducts, into up to 30,000 tonnes a year of low-carbon cement materials. Funded by federal programs and venture firm CRH Ventures, the initiative aims to embed circular economy solutions into heavy industry. Also read our 2024 case study on Carbon Upcycling.

➔ UN plastics treaty talks are on the brink. There was limited consensus among delegates of 179 nations after a 10-day marathon on global plastic pollution that ends today. The scale of the problem is massive, with straws, cups and stirrers, carrier bags and microbeads among the single-use products clogging up oceans and threatening marine life. The 2,000-plus delegates are poring over 32 clauses in a draft text, but, some say, countries can’t even agree on the definition of “plastic pollution.”

AI deepfakes’ power problems

Saskatchewan Premier Scott Moe was the latest target of an AI deepfake video that falsely showed him promoting cryptocurrencies. It’s not just a public menace—it has implications for the climate.

Some AI models gobble up 20.4 million joules to generate a 30-second video, according to data extrapolated from an MIT study —that’s enough to power an electric vehicle for 18 kilometres. Eight million AI deepfakes could be shared online this year—doubling every six months, according to Open Fox, a law-enforcement consultancy. Add to it millions of meaningless, misleading and near-malicious AI-concocted videos that litter the Internet (countries as mythical monsters , anyone?). It’s not nothing: One 30-second video per day, for a month, would be the equivalent of adding 40% to a typical one-bedroom apartment’s utility bill, according to energy policy lead Shaz Merwat who wrote about AI’s power needs last year.

AI’s demand on power grids is already formidable:

➔ Datacentres will account for 20% of electricity demand growth to 2030 in advanced economies, the International Energy Agency estimates.

➔ Emissions from data centres could rise from 180 million tonnes today to 300 metric tonnes (Mt) by 2035 in IEA’s base case, with 500 Mt as a higher estimate—roughly 2.5x the emissions of Canada’s oil and gas sector.

➔ The United States now has the largest pipeline of gas-fired power plants in development globally, surpassing China—with a fifth to power data centres.

➔ Given the exponential use originating from image and video queries, most likely over time, we could see tiered pricing for heavy data users to match users’ data use for heavier AI tasks, Merwat said.

Felippa Amanta, who published a study on the AI deepfake challenge for Oxford University, goes a step further: “The deepfake’s effect on climate information and emotion will be much more significant than the direct energy to produce the deepfake.”