The artificial intelligence boom has driven stock prices of big tech companies to new highs, but it has come at the expense of the industry's climate efforts.
Google acknowledged on Tuesday that AI technology poses a threat to its environmental goals after revealing that its data centers, a key part of its AI infrastructure, have increased greenhouse gas emissions by 48% since 2019. The company said “significant uncertainties” around meeting its net-zero emissions goal of eliminating its total carbon dioxide emissions by 2030 include “uncertainties about the future environmental impacts of AI, which are complex and difficult to predict.”
So can technology reduce the environmental costs of AI, or will the benefits of staying ahead be so great that the industry will forge ahead ignoring it?
Why is AI a threat to tech companies’ environmental goals?
Data centers are a core component of training and operating AI models such as Google's Gemini and OpenAI's GPT-4. They house advanced computing equipment, or servers, that process the vast amounts of data that underpin AI systems. Data centers require significant amounts of electricity to run, which generates CO2 depending on the energy source, as well as “embedded” CO2 from the costs of manufacturing and transporting the equipment required.
According to the International Energy Agency, data centers' total electricity consumption could double to 1,000 TWh (terawatt-hours) between 2022 and 2026, equivalent to Japan's energy demand. Meanwhile, research firm Semianalysis estimates that, due to the introduction of AI, data centers will use 4.5% of global energy production by 2030. Water usage is also significant, with one study estimating that AI could account for up to 6.6 billion cubic meters of water use by 2027, equivalent to nearly two-thirds of England's annual consumption.
What do experts say about the environmental impact?
A recent UK government-backed report on AI safety said the carbon intensity of the energy sources used by tech companies was a “key variable” in calculating the environmental cost of the technology, but added that a “significant proportion” of training AI models still relies on energy powered by fossil fuels.
Tech companies are buying up huge amounts of renewable energy contracts to meet their environmental goals — Amazon, for example, is the world's largest corporate renewable energy buyer — but some experts argue that a lack of clean energy will force other energy users to turn to fossil fuels.
“Not only is energy consumption increasing, but Google is also struggling with the increasing demand for sustainable energy sources,” said Alex de Vries, founder of Digiconomist, a website that tracks the environmental impact of new technologies.
Will there be enough renewable energy?
Governments around the world are planning to triple the world's renewable energy resources by the end of the decade to reduce fossil fuel consumption in line with climate goals. But this ambitious pledge, agreed at last year's COP28 climate change conference, is already being called into question, with experts worried that surging energy demand from AI data centers could make it even more unattainable.
Global energy watchdog the IEA has warned that even though global renewable energy capacity is set to grow at the fastest pace in two decades in 2023, current government plans could only see the world's renewable energy double by 2030.
The answer to AI’s energy needs may be for technology companies to invest heavily in building new renewable energy projects to meet growing electricity demands.
How quickly can new renewable energy projects be built?
Renewable energy projects such as onshore wind and solar farms can be built relatively quickly, taking less than six months to develop. But slow planning rules in many developed countries and a global impasse to connect new projects to the power grid can mean the process can take years. Offshore wind and hydroelectric schemes face similar challenges, in addition to taking two to five years to build.
This has raised concerns about whether renewable energy can keep up with the expansion of AI. According to the Wall Street Journal, big tech companies already use one-third of U.S. nuclear plants to provide low-carbon electricity for their data centers. But without investing in new sources, these deals would divert low-carbon electricity from other users, increasing fossil fuel consumption to meet overall demand.
Will AI's power demands keep growing forever?
According to normal laws of supply and demand, more power used by AI would drive up energy costs, forcing the industry to make savings, but the industry's unique characteristics mean the world's largest companies are ignoring skyrocketing electricity prices, potentially wasting billions of dollars as a result.
The largest and most expensive data centers in the AI space are used to train “state-of-the-art” AI systems, such as GPT-4o and Claude 3.5, which are more powerful and performant than any other systems. Leaders in the field have changed over the years, but OpenAI is generally near the top, competing for position with Anthropic, the developer of Claude, and Google's Gemini.
Already, the race to be “cutting edge” is perceived as “winner takes all,” with little to stop customers from jumping on the latest leader. This means that if a company spends $100 million training a new AI system, its competitors must decide whether to spend even more themselves or drop out of the race altogether.
To make matters worse, the race to develop so-called “AGI” — AI systems that can do everything humans can — means it might be worth spending hundreds of billions of dollars on a single training run if doing so leads to companies monopolizing technologies that could, as OpenAI puts it, “improve humanity.”
Won't AI companies learn to use less electricity?
Every month brings new breakthroughs in AI technology that enable companies to do more with fewer resources: in March 2022, for example, DeepMind's Chinchilla project showed researchers how to train cutting-edge AI models with significantly less computing power by changing the ratio between the amount of training data and the size of the resulting model.
But the same AI systems didn't consume less power; instead, the same amount of power was used to create better AI systems. In economics, this phenomenon is known as “Jevons' Paradox”, named after the economist who pointed out that James Watt's improvements to the steam engine greatly reduced coal use, but instead greatly increased the amount of fossil fuels burned in Britain. Watt's invention caused the price of steam power to plummet, leading to the discovery of new uses for it that would not have been worthwhile when electricity was expensive.
