Report: data centers may exponentially increase greenhouse gas emissions
A new report from the Center for Biological Diversity foresees the climate change troubles these AI-driven centers present
Artificial intelligence is here, and it’s not going away. AI is presently being applied to advanced robotics, navigation, and even healthcare, and will soon be involved in many more industries. Of course, as most folks know, AI is used everywhere. Students use AI to write term papers, and teachers use it to grade term papers. Donald Trump even used AI to make a cartoon of himself wearing a crown, flying a jet, and dropping poop on people. (I would provide a link to this but it’s too disgusting.)
More and more industries and individuals are using AI; that means a corresponding increase in data centers needed to support it. That also means an increase in the amount of energy needed to power the data centers that power AI. Data centers already use LOTS of electricity. On average, a data center uses as much electricity as 100,000 American homes. The ones being constructed today, the mega-data centers, will consume 20 times as much.
The Center for Biological Diversity researched the projected growth of data centers, and in late October published a report: Data Crunch: How the AI Boom Threatens to Entrench Fossil Fuels and Compromise Climate Goals.
The report states that at current growth rates, and if nothing is done to mitigate emissions, by 2035 U.S. data centers are projected to emit more than 350 million metric tons of CO2 annually. This is approximately equal to the entire amount of emissions from Italy in 2023.
The AI-driven explosion in computing, which is expected to be powered by fracked gas in the U.S., could account for as much as ten percent of economy-wide emissions and 44 percent of power sector emissions by 2035.
The report’s authors found that if renewable energy were used to power this AI surge, it would comprise only four percent of power sector emissions and a very slight amount of economy-wide emissions allowable to meet the targets.
The U.S. leads in data center emissions
The U.S. already exceeds the rest of the world when it comes to data centers. Last year, U.S. data centers accounted for nearly half (45%) of the electricity used by data centers globally. Americans have the highest per-capita data center energy consumption, nearly three times that of the rest of the world. All this to create AI images of Richard Nixon dancing with a polar bear and using AI to write music and lyrics for songs. By 2030, data center energy use may be responsible for 12 percent of America’s total energy use.
“Feeding data centers with fossil fuels is taking the climate crisis we have now and blowing it up like the Incredible Hulk,” said John Fleming, Ph.D., a scientist at the Center’s Climate Law Institute and co-author of the report. “A gas-fed AI boom is going to hurdle us past any chance of keeping to our climate goal or maintaining a safe and healthy future for our planet. To the extent that data center buildout is needed at all, it should be powered only by clean, renewable energy.”
To make their analysis and projections, the team from the Center For Biological Diversity relied on the International Energy Agency (IEA) projections of data center energy demand through 2035. In their projections, the IEA used four scenarios of world energy use, basically from low to high. Most of the electricity produced for data centers is expected to come from fracked natural gas, not as polluting as coal, but still a fossil fuel.
The Center found data center emissions could triple by 2035. Given the political climate, with Donald Trump pulling the U.S. out of the Paris Agreement and calling climate change a hoax, the prediction may unfortunately become reality. “Trump is determined to feed the voracious AI vortex with more dirty fossil fuels that harm the whole world,” said Jean Su, Energy Justice Director at the Center and co-author of the report. “This report shows how the U.S. is about to set off an explosion of dirty data center emissions, entrenching more fossil fuels when we need their rapid phaseout. We need meaningful guardrails at every level to ward off this huge threat to our air, water and climate, and guard against energy price spikes for consumers.”
Recommendations
Data center construction is subject to normal local, tribal, state and federal permitting of construction projects. But because their construction presents such environmental and social costs, the Center suggests additional scrutiny for the permitting, including a public interest test and needs assessment to determine and apprise benefits and harms. Data center construction should be denied when harms outweigh the benefits.
Further recommendations include the following:
Considerations should encompass environmental costs, including electricity use, carbon and air pollution, and water use, as well as the costs of public health. These should be balanced against economic costs and benefits. If a center is approved, it should be monitored independently, with regulators regularly assessing a data center’s compliance with its permit’s original estimated projections.
Data centers need to be responsibly sited, built in places where there are abundant renewable resources, as well as places where the data center’s water use will not adversely affect the drinking water availability to local residents or adversely affect the environment.
Data centers should generate power through on-site generation of renewable electricity, such as solar and wind. Batteries can extend the use of renewable power.
Data centers should also be responsible for any upgrades needed to local power distribution.
Some states have already gotten the ball rolling on some of these recommendations. In Wisconsin, Democratic lawmakers are proposing policy frameworks for data centers to protect workers and impose sustainability requirements on tech companies that want to build data centers in the Badger State. The proposed legislation would also make the data centers responsible for upgrades to the local power grid.
Electric companies would be required to monitor and submit quarterly reports to Wisconsin’s Public Service Commission on the energy use of the state’s data centers. The data center operators would also pay fees that would go to the development of renewable energy.
Energy regulators in Kansas have approved a plan to shield rate payers from increased costs driven by data centers, while regulators in Michigan are on track to develop similar measures.
The UN recently inaugurated two bodies to address, globally, the problems presented by AI, yet neither group explicitly addresses the harms to the climate or environment from AI.
If you’re interested in what’s going on in your state or neighborhood, you can check out DataCenterMap, which has been tracking data centers and their locations since 2007. I used it to find data centers close to me and found 14 centers within a 20 mile radius of my home in San Diego. I am unsurprised, as San Diego is a high tech hub. And I’m concerned. Water is an especially precious resource in this very parched part of the world. Are these centers using up my drinking water? I don’t know and am presently stymied as to how to find out.
Do you use AI? Is it worth the potential environmental costs? Do you live close to data centers? Please click the “Leave a comment” button and share your thoughts.
A previous Green Dispatch on artificial intelligence:
AI and sustainability
A couple weeks ago, as I was planning the subject of this week’s Green Dispatch, I was originally going to update last year’s sustainable gift giving guide. But once I reviewed last year’s guide, I realized I had nothing new to offer that I didn’t already cover. So if you’re interested, please visit or revisit last year’s guide for sustainable holiday gifts:
Write a song with AI?
I'm afraid there's a rather considerable mishandling of the figures here, Lloyd.
Firstly, the analysis conflates data centres wholesale with artificial intelligence infrastructure, which strikes me as a rather unwise assumption. Data centres were flourishing quite robustly for the better part of a decade before AI came into its own, weren't they? One might reasonably conclude that we'd be witnessing this expansion of facilities regardless of AI's emergence.
Furthermore, a good many data centre operators are making efforts to power their installations through low-carbon or renewable sources. Naturally, there remains considerable work to be done on that front.
What the calculations appear to overlook entirely is that this expansion is, in fact, displacing power consumption elsewhere. As data centres consolidate into larger, more efficient installations, smaller and private facilities are being decommissioned. Yet curiously, only the incremental power usage appears in the ledger—the corresponding savings, whilst admittedly devilishly difficult to quantify, are conspicuously absent.
The analysis also fails to account for the rather dramatic improvements in efficiency. The cost per prompt—both in hardware and power consumption—is dropping precipitously. The latest generation of processors operates at something in the region of two hundred times the efficiency of its predecessor, and that predecessor is merely three to four years old.
Finally, there's the rather inconvenient fact that AI itself can be employed to enhance power efficiency and reduce wastage across numerous sectors, potentially offsetting—partially if not wholly—its own impact.
So no, I rather think AI isn't going to be the end of civilization. Not through data centres, at any rate.
Of course they will… doesn’t anybody know they are energy hogs! We are driving this with our use of every possible screen we can find every minute of every day. Stop the screen addiction. Get a life,