Never Miss a Beat: Get a snapshot of the issues affecting the IT industry straight to your inbox.
How AI Impacts Sustainability Opportunities and Risks
AI can be used to drive sustainability initiatives, yet the technology itself has an environmental cost. How can we strike a balance?
9 Min Read
Frank Harms via Alamy Stock Photo
AI is touted as an effective tool for corporations to use in support of their environmental, social, and governance (ESG) strategies. As the question of environmental sustainability becomes more pressing, that capability holds significant promise in identifying patterns and potential solutions for complex problems, like carbon dioxide emissions.
But as AI adoption increases, it cannot be ignored that the technology is a part of the very environmental sustainability challenges it can be used to address. Training large language models requires vast amounts of energy and fuels the insatiable demand for data centers.
What are the biggest opportunities to apply AI to sustainability challenges? How can enterprise leaders weigh those opportunities against the sustainability risks inherent to such powerful, in-demand technology?
AI for Sustainability
Today’s enterprises generate such vast amounts of data it is no simple matter to leverage that data to unlock answers to key questions about business operations and sustainability.
“Right now, lack of insights from data is a big sustainability challenge that C-suite executives … are facing,” Christina Shim, chief sustainability officer at IT services and consulting company IBM, tells InformationWeek.
Just four in 10 organizations reported having the ability to automatically source sustainability data from systems such as ERPs, CRMs, and enterprise asset management systems, according to the IBM report Beyond Checking the Box published in February 2024.
Capturing data is the lowest-hanging fruit in the realm of AI for sustainability. “This ability to leverage technology and AI to surface the insights across all that data and manage and aggregate it really helps to parse out how we can lower emissions and energy costs,” says Shim.
But simply finding and reporting data on energy consumption and other environmental impacts is not enough.
“Sometimes sustainability professionals spend their whole year doing things just around data collection and reporting and then once they've finally put out the report, it starts again,” says Josh Prigge, founder and CEO of North Star Carbon Management.
Data must be translated into action. North Star Carbon Management, a carbon accounting and management software company, focuses on helping enterprises find ways to drive operational change.
“That's where we have incorporated AI into our platform … that critical step of action.” Prigge explains. “So, looking at all the different emission sources the company might have and using the AI to really dive into what are the specific strategies or plans or actions that a company can take … for actually reducing emissions.”
Greg Smithies, partner and co-lead of the climate technology investment team at venture capital firm Fifth Wall, offers the example of a cement manufacturer that needs to run its kiln at a certain temperature. AI can help that manufacturer discover temperature fluctuations and actually maintain the optimal temperature for improved yields and energy usage.
“That is a use case where AI can optimize industrial processes on … a millisecond-to-millisecond basis,” he explains.
Enterprise leaders can also put AI to work to discover new, more sustainable ways of doing business in the future. Looking at the cement manufacturer example, the industrial processes it needs to create its product emit CO2. What if AI could search all available research and build a model for an alternative material that results in lower emissions?
“It is those sorts of true novel materials discoveries or new recipes for things or new processes for doing things where I see the biggest potential savings from a sustainability side from AI, but those are also the things that'll take the longest,” says Smithies.
AI and Sustainability Strain
While AI can be applied to sustainability challenges, there are also questions around the sustainability of AI itself given technology’s impact on the environment.
“We know that many companies are already dealing with the ramifications of increased energy usage and water usage as they're building out their AI models,” says Shim.
Research from Goldman Sachs projects that AI will drive 160% growth in data center power demand by 2030.
AI is not the only strain on electrical grids. “So, we've got electric vehicles coming online. We've got gas furnaces being changed to heat pumps. We've got gas stoves being changed for induction cooktops,” Smithies points out.
Global demand for electricity is expected to grow by approximately 4% this year, the biggest annual leap in growth since 2007, according to a report from the International Energy Agency (IEA).
In addition to being power-hungry, AI models compete for another valuable resource: water. The technology could be using 6.6 billion m³ by 2027, according to Forbes, a staggering number considering significant concern regarding water scarcity.
And then there is the carbon footprint of AI to consider. In the race to build bigger and better AI models, hyperscalers’ emissions are soaring. Despite pledges to reduce CO2 emissions, companies like Google and Microsoft are seeing increases, NPR reports. Google’s greenhouse gas emissions climbed 48% since 2019, according to its 2024 Environmental Report. The culprit? In part, it’s the energy demand from data centers.
“As we further integrate AI into our products, reducing emissions may be challenging due to increasing energy demands from the greater intensity of AI compute, and the emissions associated with the expected increases in our technical infrastructure investment,” according to the report.
And Google is hardly alone in integrating more and more AI into its products that will be used by enterprises and individual consumers around the world.
Measuring the Environmental Impact of AI
How do enterprise teams measure the environmental impact of their AI usage? If an organization uses and/or trains models with a single hyperscaler, like Amazon or Google, the answer to that question may be a little easier, but not easy, to answer. These companies track their energy consumption and carbon emissions.
“The problem is we don't necessarily know the fidelity to which they're doing it. We know it's not a 100%,” Smithies points out.
Enterprises working with a single hyperscaler might at least access some publicly available data on environmental impact, like the data in Google’s Environmental Report. But many companies work with multiple hyperscalers that may not pull back the curtain that far, as well as smaller data center providers that may not even gather that kind of data let alone share it.
“The larger a corporation you are the more likely you are to have your workloads in multiple … different suppliers’ data centers and then this rapidly turns into a nightmare,” says Smithies.
Enterprise leaders can use spend as a starting point. “What are the dollars spent on the various cloud-based services, like AI companies or data storage?” asks Prigge. “[Use] average emission factors across that industry to quantify emissions based on spend.”
Drilling down into the data that enterprises amass is a challenge. Emissions and energy use are important measures of AI’s sustainability, but so is water use and land use.
“If it's a material impact for the company, it should be a part of their formal sustainability program or strategy and that includes measuring your performance, reporting, setting goals, and targets,” says Prigge.
AI can help be a part of the solution to its own sustainability challenges by sifting through that data and finding the answers to questions about its resource usage and potential solutions to drive those numbers down.
While measurement may take time to achieve, especially when it comes to precision, it is an important undertaking. “As greenwashing becomes more of an issue, people are concerned … these efforts need to be transparent and authentic,” says Prigge.
How to Strike a Balance
AI is a relatively young industry in the midst of a major hype cycle, much like the internet before it.
“I think that same logic probably holds here for AI. We're going to see a massive overbuild. We're going to see some kind of a pull back,” says Smithies. “It doesn't necessarily mean that a full bubble is going to collapse, but there will be a pull back and then we'll be able to the amortize the overbuild of infrastructure over the next decade.”
As the AI market goes through its growing pains, chips are likely to become more efficient and use cases for the technology will become more targeted.
But predicting the timeline for that potential future or simply waiting for it to happen is not the answer for enterprises that want to manage opportunities and risks around AI and sustainability now. Rather than getting caught up in “paralysis by analysis,” enterprise leaders can take action today that will help to actually build a more sustainable future for AI.
With AI having both positive and negative impacts on the environment, enterprise leaders who wield it with targeted purpose are more likely to guide their organizations to sustainable outcomes. Throwing AI at every possible use case and seeing what sticks is more likely to tip the scales toward a net negative environmental impact.
Instead, exploring smaller, purpose-built models can help enterprises reap the benefits of AI without using more energy and resources than necessary.
“Bigger is not always better. I think that it's really important that businesses consider the right size of the foundation models to accomplish what they need,” says Shim.
IBM, for one, offers Granite: purpose-built, open-source AI models.
Refining models for specific use can make them more efficient from an energy consumption perspective. Enterprise teams may explore caching results from common queries and apply retrieval-augmented generation (RAG) to reduce cost and improve the efficiency of an AI model.
If AI adoption is inevitable, and it certainly seems so, enterprises do have some control over what this means for their energy consumption. Shim argues for the adoption of a hybrid cloud strategy that takes into account energy sources and processing time.
“A hybrid cloud approach can really give organizations the flexibility to locate their processing close to their data and/or close to a clean energy source,” she says.
Shim also advocates for embracing an open-source approach in the AI world. “It means like there's a lot more eyes on the code. There's more minds on the problem. There's more hands on the solution. And that kind of transparent collaboration can have a really big impact on how we think about on the sustainability of AI,” she elaborates.
How ever enterprises opt to adopt AI, keeping sustainability a part of the conversation will require a multidisciplinary approach. C-suite executives and different teams within an enterprise will need to collaborate to discover the most valuable AI use cases, to understand how the models behind those uses cases have an impact on sustainability, and what can be done to reduce that impact.
Embracing AI governance can help enterprise teams tackle the challenging issues that come with the rapid adoption of AI: ethics, social impact, and of course, sustainability.
“If organizations and leaders are able to take these very tactical and tangible steps [in] how they think about their AI strategy, it will make a huge difference,” says Shim. “We will hopefully not be in the place where we're wondering … what we could we have done back in 2024?”
About the Author
You May Also Like
More Insights
