Cloud Computing or Coal Computing?

AI might solve climate change someday, but right now it’s an energy-consuming monster…

Dec 12, 2024

Google’s Green House Gas (GHG) emissions in 2023 were almost 50% higher than in 2019. Despite tech company claims of “clean energy,” some data centers use coal-burning power.

Seattleites spend an inordinate amount of time trying to conserve energy and recycle. A birthday luncheon for a friend illustrates our preoccupation, and while some of it may seem performative, it comes from a sincere drive to save our planet. As I’ve delved deeply into all things AI, for the first time I questioned whether our conservation efforts could possibly make a difference.

I was sweating despite the cold, after rushing to find parking on the congested streets of Capitol Hill, downloading yet another damn parking app, and locating the restaurant intentionally hidden from view.

“It’s called Stew?” I had asked Jeannette when I received the invite. “And that’s a good thing?”

Upon entering the restaurant, I joined a table of impeccably dressed and groomed women. They were already ordering. Arden was interrogating the server. Are the lettuces organically grown? How many food miles on the tomatoes?

“I don’t know,” the server stammered, “I can check.” Arden scowled.

“They have a cow out back you can milk for your cream, Arden,” Jeannette teased.

“Very funny. You know I’m lactose intolerant.”

The conversation drifted to orderless food waste bins for the kitchen and skin care products “with a conscience.”

“I hate to burst your bubble, ladies, but energy needs of Big Tech are bigger than ever. And despite what they would like you to believe, they are currently relying on a high percentage of carbon-based fuel.”

Silence…everyone just stared at me. “Hey, how bout them Seahawks?”

A closer look at the energy requirements of AI reveals alarming power demands.

We've all heard about “The Cloud.” It sounds fluffy and delightful, however, enormous, power-guzzling data centers, spread across deserts, prairies, and farmlands globally, is the reality. How big? The Citadel in Reno, Nevada, spans 7.2 million square feet, or approximately 165 acres. Data centers are the real “cloud,” housing the physical servers and hardware required to store data, run applications, and deliver services.

While conventional computing taxes the grid, AI’s demands are exponential. Large Language Models (LLMs) like ChatGPT require vast amounts of data for training and inference, and the computational power needed for image-generating models like DALL-E is even more resource-intensive. Think twice before asking DALL-E to make you a picture of a cat smoking a cigar.

One ChatGPT query takes nearly ten times as much energy as a typical Google search.
In 2019, researchers at UMass, Amherst, studied the energy consumption required to train large AI models. The CO2 emissions were equivalent to the entire lifetime emissions of five gas-powered cars.
There are more than 8,000 data centers worldwide and counting.
Data centers could consume as much as 17 percent of all U.S. electricity by 2030.
Not all power used for data centers is clean. Depending on location, it is often a mix of fossil fuels and clean energy sources, including coal, natural gas, solar, and wind.

The Washington Post reported in October 2024 that residents of a low-income, largely minority neighborhood in North Omaha were shocked to learn a coal-burning plant scheduled to be decommissioned would remain operational. The power-hungry demands of Meta and Google data centers in the area outweighed the community’s need for clean air while many suffer from asthma and other pollution-related ailments. Rural parts of West Virginia have experienced the same, where coal plants fuel the world’s internet hub in Northern Virginia, 35 miles away.

But we can take comfort from Meta’s website. “Operations of our data centers and offices have already reached net zero emissions and our data centers’ electricity use is matched with 100% renewable energy through our strategy of adding renewables to the local grids, which also helps decarbonize the electricity system.”

"Net zero" means the total balance after accounting for emissions produced and emissions removed or offset. Offsetting involves supporting external projects to neutralize emissions; while matching ensures renewable energy generation corresponds to the amount of energy used. Companies can purchase renewable energy credits (RECs) to match their electricity consumption with clean energy production, ensuring renewables balance consumption.

Paying a solar farm operator additional money to produce electricity they would have already generated doesn’t in any way reverse the emissions from a natural gas or coal power plant. Nor do offset credits have to be in the same community where the data centers are polluting.

The concept of balancing carbon emissions emerged from early climate negotiations in the 1990s, but it wasn’t until “net zero” became a key goal of the 2015 Paris Agreement that the term became popular as corporations signed on. NASDAQ provides a carbon credit market for corporations.

Big tech is recognizing they need to feed the beast. Microsoft made news by teaming with Constellation Energy to bring back the Three Mile Island energy plant, the site of the worst nuclear accident in U.S. history. Meta, Amazon, and Alphabet are also investing in companies developing nuclear reactor technologies.

Nuclear power is back in favor due to advanced technology and key benefits, including low carbon emissions and remarkable energy density. Uranium can produce 1,400 times more energy than the same amount of natural gas. But the storage of radioactive waste, and rare meltdowns like Chernobyl (1986) and Fukushima (2011), make nuclear potentially devastating to the atmosphere and living organisms in the vicinity. Who wants a nuclear reactor in their backyard? Let’s see a show of hands.

Tech’s carbon footprint will get worse before it gets better. Right now, they are balancing the immediate and massive energy needs generated by AI with currently available and prospective (nuclear) power sources. Let’s hope those short-term gains don’t jeopardize the real innovations that might be coming.

As they say, necessity is the mother of invention and AI is an excellent lab partner. AI's ability to process and analyze data at unprecedented speeds, learn from patterns, and adapt to new information is transforming climate science and driving innovation. The good news is big tech is effective at solving problems, and climate change is one they are highly motivated to address.

Subscribe to find out how AI might eventually solve climate change and other fascinating ways AI will affect us all for better and worse. Please share Getting Real About AI with friends. Were you already familiar with the concept of Net Zero? Let me know by commenting below.

Celeste Garcia Getting Real About AI

Discussion about this post