Five things we learned about AI and climate change at PNW Climate Week
Image by Gerd Altmann from Pixabay
The rise of artificial intelligence has been accompanied by a worrying rise in power demand. AI systems require a massive amount of computing power. The data centers that provide that power draw enormous amounts of electricity.
The scale is troubling. The International Energy Agency (IEA) has estimated that AI could require up to 1,000 TWh annually by 2026. That’s equivalent to Japan’s annual electricity consumption.
Transparency Coalition’s Jai Jaisimha recently joined a panel of AI and climate experts to discuss the challenges and possibilities that AI brings to the climate crisis. The event was part of PNW Climate Week, a six-day conference that runs concurrently in Seattle, Portland, and Vancouver, B.C.
Here are five takeaways from the discussion.
PNW Climate Week panel on climate and AI, L-R: Moderator Jordan Bain, Heather Redman, Jai Jaisimha, Sailesh Chutani, Hansi Singh, Vikram Modgil. More on the panelists here.
Amazon Web Services just ‘bought’ a nuclear power plant
This head-turning fact was dropped by Heather Redman, managing partner at the AI-focused venture capital firm Flying Fish. AI’s demand for computing power last year drove a massive sales gain for AWS, Amazon’s cloud computing service, which is one of a handful of global companies with the infrastructure sufficient to meet the rising demand.
So great is that demand that these cloud hyperscalers—companies competing to provide global-scale compute, like AWS, Microsoft Azure, Google Cloud, and Meta—are now seeking ways to bypass traditional energy generation suppliers and generate megawatts themselves. This feels like the moment Amazon realized it was becoming too big for UPS and the US Postal Service and began to create its own delivery fleet.
Amazon didn’t buy a turnkey nuclear plant off the shelf, of course. In March 2024, the company purchased a data center from Talen Energy. That center happens to be located right next to Talen’s Susquehanna Station near Berwick, Pennsylvania. Susquehanna is the nation’s sixth-largest nuclear power plant, with a generating capacity of 2,228 megawatts. Amazon plans to build a 960-megawatt data center campus there in the near future. It’s not far-fetched to imagine the entire output of the plant eventually dedicated to powering the servers.
Meanwhile, Microsoft is looking to hire a nuclear energy infrastructure director to focus on powering its cloud service Azure.
2. beware the looming human-vs-server energy battle
As climate change drives more heat waves and powerful storms in places like Texas, Arizona, and California, these increasingly regular events result in energy demand spikes. When the temps hit 110 degrees for weeks on end, air conditioners roar and energy grids struggle to stay online.
Hansi Singh, a climate tech expert and founder of Planette.AI, foresees a coming competition between human resource needs and data center needs. “Human demand for energy is growing due to climate change,” she said. “Humans and their air conditioners are already competing with servers” for energy. “At a certain point there is an issue of human life and human well-being.”
Singh sees parallels with water fights in the American West. When drought hits, some farmers have their irrigation water turned off. There is a well-established system of priority water rights in the West. But there is no such system with regard to energy generation. If Amazon and Microsoft own more and more of the power grid, who will suffer rolling blackouts first—data centers or the residents of Phoenix?
3. water rights may be the key to future compute power
Transparency Coalition founder Jai Jaisimha took it one step further: Nvidia announced earlier this year that its next generation of CPUs will be liquid-cooled. Which means water may also enter the AI resource battle. “If you are concerned about water consumption at these data centers, their consumption” may soon increase with the coming of liquid-cooled chips,” Jaisimha said.
TCAI’s Jai Jaisimha (center): Companies need incentives to behave in the right way.
4. AI could increase energy efficiencies, but it may not be enough
The growth of AI-driven energy system startups like Phaidra may bring new efficiencies to data centers, reducing energy consumption by as much as 40%. But with data center growth moving at 2x and more, greater efficiency may be just one part of the solution. “The United States cannot deal with the energy demands of AI,” said Heather Redman. “We’ve gained efficiencies” over the past decades, “but we’ve not built out load growth.”
5. AI companies need an incentive to operate within ethical bounds
Will AI companies mitigate these problems themselves? Not without incentives.
“Self-regulation, third-party certifiers, and industry standards can emerge,” said Jaisimha. “Industry will not move in that direction unless there’s something over their heads that makes them move. They need an incentive, either financial or legislative.”
“Some kind of looming legislative effort can encourage people to behave in the right way,” Jaisimha added. “Privacy is a great example. There is no federal data privacy law. So seventeen different states passed their own privacy laws—and suddenly everybody’s at the table negotiating a federal privacy law. That kind of dynamic could play out in the larger AI space.”
NIST, the National Institute of Standards and Technology, “has put out a framework for what good housekeeping looks like” in the AI industry, said Redman. But it’s a standard without positive or negative incentives. “We do need regulation at the national and international level,” she added. ”I don’t know we’ll get it at the international level, but it may happen” at the national level.