Michelle ThompsonArtificial intelligence (AI) has an insatiable appetite for energy, especially as companies strive to enhance its size, intelligence, and complexity. This poses a significant challenge for an industry that promotes itself as a tool to combat climate change, yet contributes to a substantial carbon footprint.
Sam Altman, the leader of OpenAI, the organization behind ChatGPT, advocates for nuclear fusion as a definitive solution to this complex dilemma. Altman has heavily invested in fusion technology and in recent interviews has highlighted its potential to meet the substantial power demands of next-generation AI.
Altman emphasized the necessity of fusion alongside the expansion of other renewable energy sources in a January interview. When asked about solving AI’s energy demands in March by podcaster and computer scientist Lex Fridman, Altman once again underscored the importance of fusion.
However, mastering and commercializing nuclear fusion, a process akin to what powers stars like the sun, is likely decades away. Some experts view Altman’s focus on a future energy breakthrough as indicative of the AI industry’s failure to address the immediate challenge of meeting AI’s escalating energy needs. This inclination toward “wishful thinking” echoes broader sentiments in climate action, according to Alex de Vries, a data scientist and researcher at Vrije Universiteit Amsterdam.
OpenAI declined to respond to specific inquiries from CNN, only referencing Altman’s previous statements.
Nuclear fusion holds great appeal for the AI industry due to its potential to release vast amounts of energy without emitting carbon or leaving behind long-lived nuclear waste. However, recreating the conditions necessary for fusion, akin to those found in the center of the sun, presents monumental technological hurdles. Aneeqa Khan, a research fellow in nuclear fusion at the University of Manchester, suggests that fusion may not be viable for addressing the immediate climate crisis. Instead, she advocates for utilizing existing low-carbon technologies like fission and renewables.
Fission, the process currently used to produce nuclear energy, remains a significant part of the energy landscape.
On January 22, 2024, a portion of JT-60SA, an expansive experimental nuclear fusion reactor situated at the Naka Fusion Institute in Naka city, Ibaraki Prefecture, Japan, was photographed.
The challenge lies in securing adequate renewable energy to fulfill the escalating needs of AI in the short term, rather than resorting to fossil fuels, which contribute to global warming. This challenge is exacerbated as the global transition to electrify various aspects such as cars and heating systems increases the demand for clean energy.
According to a recent analysis by the International Energy Agency (IEA), electricity consumption stemming from data centers, cryptocurrencies, and AI could potentially double within the next two years. In 2022, this sector already accounted for approximately 2% of the world’s electricity demand, as per the IEA.
The analysis forecasts an exponential growth in AI demand, expected to surge at least tenfold between 2023 and 2026. Besides the energy required for manufacturing chips and hardware, AI necessitates significant computing power for training models—where vast datasets are utilized—and subsequently for responding to user queries.
As AI technology progresses, companies are racing to incorporate it into applications and online searches, leading to heightened computing power demands. A recent report by de Vries on AI’s energy footprint estimated that an AI-powered online search could consume at least ten times more energy than a standard search.
De Vries notes that the prevailing dynamic in AI favors larger, more energy-intensive models, presenting a fundamental challenge to sustainability. This tendency toward “bigger is better” exacerbates the issue.
In the United States, the situation is particularly critical, with energy demand surging for the first time in approximately 15 years, according to Michael Khoo, the climate disinformation program director at Friends of the Earth. Khoo, co-author of a report on AI and climate, states that the country is facing an energy shortage.
A significant factor driving this demand is the proliferation of data centers, with electricity consumption expected to triple by 2030, equivalent to powering around 40 million US homes, as outlined in an analysis by the Boston Consulting Group.
Khoo emphasizes the necessity of making difficult decisions regarding energy allocation, whether prioritizing thousands of homes or a data center powering next-generation AI. He asserts that energy distribution cannot solely favor the wealthiest individuals.
On March 30, 2023, high school students at Pascal School in Nicosia, Cyprus, are seen collaborating on “Alnstein,” a robot energized by ChatGPT.
For many AI companies, concerns regarding their energy consumption overlook two significant points. Firstly, AI itself can contribute to addressing the climate crisis.
According to a spokesperson for Microsoft, which collaborates with OpenAI, AI holds the potential to advance sustainability solutions significantly. The technology is already being utilized for various environmental purposes, such as weather prediction, pollution tracking, deforestation mapping, and ice melt monitoring. A recent report by the Boston Consulting Group, commissioned by Google, suggests that AI could help reduce up to 10% of planet-heating pollution.
Additionally, AI could play a role in advancing nuclear fusion. In February, scientists at Princeton announced a breakthrough in using AI to forecast potential instabilities in nuclear fusion reactions, marking progress toward commercialization.
AI companies assert that they are also striving to enhance efficiency. Google claims that its data centers are 1.5 times more efficient than typical enterprise data centers. Microsoft states that it is investing in research to evaluate the energy consumption and carbon impact of AI while working on methods to improve the efficiency of large systems, both in training and application.
Although there has been a significant increase in AI efficiency, Alex de Vries, a data scientist, warns that this doesn’t necessarily translate to a decrease in AI’s electricity demand. History suggests that efficiency gains often lead to increased demand, as seen in the case of cryptocurrency mining.
In the United States, there is some political momentum to scrutinize the environmental consequences of AI more closely. In February, Senator Ed Markey introduced legislation aimed at compelling AI companies to disclose their environmental impacts, including the surging electricity demand of data centers.
However, few anticipate bipartisan support for the bill to become law. Meanwhile, the trend toward developing increasingly complex and energy-intensive AI is perceived as inevitable. Companies are engaged in an “arms race” to produce more advanced AI models, resulting in higher electricity consumption.
Michael Khoo, a climate advocate, emphasizes the need to question how companies are actively addressing climate change. He urges scrutiny of whether their actions lead to reduced energy intensity or merely serve as a distraction.
