I stood in the cavernous hall of a new data center outside Seattle last month, the whirring of cooling systems creating a constant hum that reminded me of standing near a waterfall. My guide, an operations manager named Ravi, gestured toward row after row of humming servers.
“Each rack consumes about as much electricity as 25 households,” he explained, his voice competing with the mechanical symphony around us. “And we’re just one facility among thousands worldwide.”
This particular data center runs largely on hydroelectric power—a regional advantage in the Pacific Northwest. But as I’ve discovered through months of reporting, this is the exception rather than the rule in our rapidly expanding digital infrastructure landscape.
The global artificial intelligence boom is creating an unexpected beneficiary: fossil fuels. As data centers multiply to power ChatGPT, generative AI, and other computational behemoths, their enormous energy needs are driving new investment in oil and gas infrastructure at a time when climate scientists have warned we should be rapidly transitioning away from carbon-intensive energy sources.
According to recent analysis from the International Energy Agency, electricity consumption from data centers could double by 2026, representing more new demand than the total national electricity consumption of countries like Australia or Spain. Tech companies purchased a record 46 gigawatts of renewable electricity in 2023, yet demand is outpacing the availability of new clean energy projects.
“We’re seeing a profound disconnect between tech companies’ climate pledges and the reality of how quickly we can build renewable capacity,” says Dr. Helena Wong, energy systems researcher at the University of British Columbia. “This gap is being filled with fossil fuels in many regions.”
The numbers tell a stark story. The electricity consumed by a single ChatGPT query requires approximately 10 times more energy than a standard Google search. Training a large language model can generate as much carbon as five cars emit during their entire lifetimes, according to research published in Science last year.
In Alberta’s oil sands region, where I traveled this spring, energy companies are seeing renewed interest from data center developers attracted by reliable natural gas supplies and existing energy infrastructure. Similar patterns are emerging across Texas, where gas-powered electricity generation is expanding to meet growing computational demands.
“It’s somewhat ironic,” says James Rivera, an energy transition analyst with Climate Policy Initiative, “that the same technologies that could help optimize our energy systems are currently increasing fossil fuel consumption because we’re deploying them faster than we can build clean energy.”
When I visited the Fort McMurray region in March, I met with Darlene Cardinal, an Indigenous environmental advocate from the Athabasca Chipewyan First Nation. We stood at the edge of a vast tailings pond as she pointed to new construction in the distance.
“They tell us this boom is different—it’s for the digital economy, not just for cars and plastics,” she told me, the wind whipping across the industrial landscape. “But the impacts on our water, our air, our traditional lands remain the same.”
Cardinal’s community has spent decades documenting the health and environmental effects of fossil fuel extraction. Now, they’re watching a new justification emerge for continued development.
The situation presents a complex challenge for climate policy. Tech companies have made ambitious climate commitments—Microsoft aims to be carbon negative by 2030, while Google has pledged to operate on 24/7 carbon-free energy by 2030. Yet their energy demands are growing faster than renewable infrastructure can be built.
Environment Canada estimates that without significant policy intervention, data center emissions could account for up to 3.5% of the country’s total carbon footprint by 2030, up from less than 1% today.
“We need to have an honest conversation about the energy footprint of AI,” says Melissa Chen, director of Digital Sustainability Now, a Vancouver-based nonprofit. “Many of these models are being developed and deployed without full consideration of their climate impacts.”
Some promising approaches are emerging. In Stockholm, where I visited earlier this year, excess heat from data centers is being captured and redirected into the district heating system, warming thousands of homes during harsh Nordic winters. Meanwhile, companies like Borealis Computing are developing specialized AI chips that require significantly less energy than conventional hardware.
Dr. Wong from UBC believes we need more sophisticated policies. “Carbon pricing alone won’t solve this. We need regulations that encourage efficiency in algorithm design, promote heat recovery, and require transparency about AI’s energy footprint.”
Back in Seattle, as I prepared to leave the data center, Ravi showed me one final area—a testing zone for liquid cooling technology that could reduce energy needs by up to 40% compared to conventional air cooling.
“This is where we’re heading,” he said, his face reflecting in the blue-tinted cooling fluid. “But the question is: can we deploy solutions like this quickly enough?”
It’s a question that hangs in the air like the persistent hum of the servers themselves—a digital economy racing forward while our energy systems struggle to keep pace in a climate-compatible way.
As someone who has reported on both environmental issues and technological change for over a decade, I find the collision of these worlds particularly revealing. The AI revolution promises to transform countless aspects of our lives, but its physical footprint reminds us that even our most ethereal technologies remain tethered to the material world—and to the climate consequences that result.
