Artificial intelligence continues to capture the attention of consumers and companies alike—but not all its airtime has been positive.
People have started to hone in on the potential negatives associated with AI’s expansion, both from the human perspective and the environmental. Fears about job loss have come to bear at some major companies, and research about AI’s potential to harm the environment has started to come forward, as well.
Some specific types of AI—for instance, generative AI—are more directly linked to environmental degradation than simple, legacy AI systems that companies have long leveraged. But the proliferation of generative AI paired with its potential for environmental damage entering the public consciousness simultaneously has left many wondering about whether AI’s benefits are meaningful enough to justify its continued training and usage at scale.
Related Stories
Much of that concern comes from how much water usage and energy consumption the continued training of AI models could require. Morgan Stanley research estimates that the need for these resources will only become greater.
“We now expect AI data centers to drive annual water consumption for cooling and electricity generation to approximately 1,068 billion liters by 2028 (our base case)—an increase of 11 times from 2024 estimates,” company analysts wrote in its On The Markets newsletter. “Our forecast is based on Morgan Stanley & Co. Research’s expectation that generative AI power will expand 8.5 times from 2024 by 2028.”
The report also noted that “more than half of the world’s top data center hubs are in areas already facing medium basin physical risk, or medium-level vulnerability, to threats from drought, flooding and declining water quality.”
Simultaneously, electricity consumption associated with data centers is on the rise. The International Energy Agency (IEA) released research in April that shows that by 2030 the amount of electricity demand from data centers will double to approximately 945 terawatt-hours (TWh).
The IEA said that amount of electricity is “slightly more than the entire electricity of Japan today,” and further noted that AI is the root cause of the increase, since data centers’ rapid growth is heavily attributed to that kind of project. The U.S. alone is expected to account for about half of that demand.
Despite the potential environmental consequences associated with some rapidly growing subsects of AI, experts said that on the aggregate they remain interested in the ways that AI could aid sustainability outcomes for companies across the globe.
Tallat Hussain, partner in law firm Reed Smith’s energy and natural resources group, considers herself an optimist about the technology’s future potential in many industries, fashion and apparel included.
Because the industry’s value chain is convoluted and involves a number of suppliers and transportation partners at any given moment, AI has room to create efficiencies in fashion’s production, supply chain, logistics, circularity and governance efforts, among other areas of the business.
“Where AI can contribute positively to that, I think, relatively outweighs the carbon footprint of it,” Hussain explained.
Meera Atreya, director of decarbonization science and EMEA advisory at Carbon Direct, said to get to that point, companies building and deploying AI will have to be thoughtful about their approaches.
“The same analytical power that drives hyper-personalized marketing can also be applied to optimize manufacturing, reduce waste and improve demand forecasting when deployed responsibly,” Atreya said. “AI tools are already being used to identify novel, lower-impact materials, enable textile recycling and streamline supply chains. These applications, if prioritized, could help fashion address its overproduction problem rather than exacerbate it.”
But she warned that exacerbating overproduction could be a very real side effect of rapidly evolving generative AI systems if not approached thoughtfully and with care for the environment.
Atreya particularly stressed fast fashion’s potential to use AI systems for financial gain while ignoring the environmental side effects, noting that because generative AI can “conjure endless variations of garments and aspirational looks,” companies risk creating even more rapid consumer demand cycles, resulting in higher waste and overproduction. Still, she said, if companies can avoid falling into that trap, AI has the power to positively impact the industry.
“The sustainability outcome of AI in fashion depends less on the technology itself and more on the choices brands make in deploying it. If acceleration of consumption becomes the default, the industry risks magnifying its biggest environmental liabilities. But if the sector channels AI’s capabilities into slowing production, designing responsibly and scaling circular solutions, it could become a powerful enabler of sustainability,” Atreya said.
Saskia Van Gendt, chief sustainability officer at supply chain management company Blue Yonder, concurred with Hussain’s assessment that AI has the potential to do more good than harm when it comes to sustainability—but noted that most companies are still figuring out how to make that go. She said fashion and apparel purveyors are already starting to reap the benefits of AI on their supply chains, material selections and at-large production.
For Van Gendt, it’s important that companies think about AI from the perspective of a “balance sheet.” That is to say, they need to acknowledge the water and energy use associated with AI systems while matching the technology to pressing sustainability challenges. Part of that equation is understanding the need for less energy- and water-intensive solutions; generative AI requires more resources than legacy models, for instance.
Still, Van Gendt said she knows generative AI holds real promise for companies looking to move the needle on their sustainability strategies and said that, in many cases, it could serve as a stepping stone toward cutting carbon emissions in the fashion, apparel and retail industries.
“There definitely will be the need for generative AI solutions to solve some of the more complex sustainability problems, but there’s also a lot of optimization that can come from the predictive analytics [models] and the smaller language models, and not always having large language models, which tends to be where a lot of the energy use is—in the development and the training of generative AI models that are based on large language models.”
Hussain said that, as companies continue to learn more about their AI usage, they’re likely to rally for more efficient data centers, powered by clean energy. With AI’s rapid proliferation and the climbing demand for data centers to support that growth, Hussain expects to see related energy and water reduction efforts on the rise—even if the aggregate demand for resources is increasing.
“We’re finding ways very quickly to reduce the water usage, to reduce the carbon footprints, to retrofit them with more renewable power,” she said.
But experts agreed that a thorough understanding around how AI usage impacts their carbon emissions and water footprints may still be far off for many leaders and organizations.
“I don’t think that a lot of companies are considering the sustainability ramifications of using AI, whether in the fashion sector or otherwise,” Hussain said.
She further noted that while leaders might be warned of the carbon impacts associated with specific use cases for AI—like, for instance, the fact that Goldman Sachs researchers reported that a single ChatGPT query can demand 10 times the amount of electricity as a simple Google search—they often find it hard to contextualize that.
“AI, in so many ways, isn’t tangible, so you don’t think of the tangible cost of it, whereas the cost of electricity for a light is tangible; the environmental impact of the car you drive or how it is powered is tangible,” Hussain said.
Van Gendt said she has started to see companies show an interest in the emissions related to their technology activities, but said it can be difficult to report on that because of the distance between the end user and the company creating and operating the technology.
“Just like it’s hard to get emissions associated with growing cotton if you’re buying dyed cotton and then making a T-shirt from it, the same kind of transition happens when it comes to technology,” Van Gendt explained. “If you’re buying third-party software that you don’t have [a] direct relationship, then it either becomes surveying or asking your direct supplier if they can provide [emissions information].”
She expects tracking down emissions data associated with technology to ease over time.
Hussain said companies are not yet considering in a robust way the ways AI usage, particularly coming from third-party technology providers, could influence their Scope 3 emissions. From her perspective, it’s difficult for brands and retailers to measure that today. That’s partly because Scope 3 reporting is largely optional in many jurisdictions, but also because many AI systems are considered black boxes—that is to say, most users don’t understand how they were trained or how they arrive at conclusions presented to users.
“The more we understand the role of AI in a sector—say, the fashion sector—the better we’re able to determine what the metrics are for measuring Scope 3 emissions,” Hussain said. “You can’t just randomly start reporting on it. It’s very hard to separate it out…It can be done, but it requires significant data management in order to do it.”
This article appeared in Sourcing Journal’s Sustainability Report. To download the full report, click here.
