The Effects of AI on Your Carbon Footprint (and How to Curb It)

Maybe you’ve seen the recent headlines—recent ones include “AI’s Energy Demands Are Out Of Control” (Wired), “AI brings soaring emissions for Google and Microsoft” (NPR), “AI emissions are fueling a new doomerism. This time it’s climate change” (Fortune), and “Artificial Intelligence Can Make Companies Greener, but It Also Guzzles Energy” (Wall Street Journal). Given the abundance of such press, it certainly seems as if the growing prevalence of AI across all platforms has sparked some climate controversy.

Because that prevalence will likely only continue to expand, it may be the right time to reflect on this technology’s role in the fight against—and its contribution to—global warming. As experienced assessors and ESG experts, we’re offering our two cents.

In this article, we’ll dive deeper into the energy consumption of AI, how it plays into greenhouse gas (GHG) emissions, and which companies are doing what in their attempts to curb those emissions—as well as how you can too.

The Dilemma That is AI’s Energy Consumption

Amidst all this publicity regarding climate concerns and AI, the IT and software industries are pushing back with claims that it’s a powerful tool in the fight against global warming, and they’re right, to an extent. AI does have the capability to:

• Evaluate vast data sets in climate research;
• Provide greater understanding of extreme weather climate risks; and
• Model scenarios and calculate indirect emissions—all of which would help experts better understand climate change.

However, the reality is that, even when using it for such a noble and productive purpose (or even when not), the energy demands of AI requests are much different than those of less advanced technology.

Take AI image generation, for instance. When you generate an image using AI, you’re engaging intensive computational hardware running the highly complex generative models that are being continuously trained in order to provide accurate images in real time. Nothing about the large datasets, iterative process, and the necessary infrastructure to support those—which is generally housed at a data center—is efficient, so in fact, a lot of energy is being consumed (despite how quickly the image may appear on your screen).

But that’s not the only way AI demonstrates how much hungrier it is for energy (in comparison to traditional data center traffic and data storage).

Even at the development stage, AI requires a considerable energy investment. In 2019, researchers at the University of Massachusetts at Amherst estimated the energy cost of training a single large language model at slightly more than 600,000 pounds of carbon dioxide-equivalent emissions. For those who don’t measure these things daily, that’s the same as 125 round-trip flights between Beijing and New York.

AI’s Effect on GHG Emissions and Global Warming

If this doesn’t yet capture the scale of the growth of global data center energy demand, maybe this will—according to the International Energy Agency (IEA), in 2020 data centers and data transmission networks were responsible for 1% of energy-related GHG emissions. In the U.S., data centers use about 1.8% of the nation’s total energy consumption.

While these percentages may not seem that alarming, remember that data centers are reliant on—and contribute to—a wider industry category of Information and Communications Technology (ICT), and the World Bank estimates that ICT generates between 1.4-4% of global GHG emissions.

That’s still not a lot, you may be thinking, especially when compared to the emissions of similarly broad industries such as transportation that are also getting their share of climate flack in the press. But our measurements now can only be based on the current state—AI use will continue to grow, but when you also factor in overall global economic growth, continued digital penetration, expansion of cloud storage, uptake in IoT, and the increasing prevalence of crypto and related blockchain technologies, the collective effect on climate change could be exponentially worse, even as efficiencies are factored in.

So, those 1% and 1.4.-4% numbers for AI energy consumption that didn’t seem concerning moments ago now won’t last, as the International Energy Agency reports that data centers consumed 200TWh (terawatt hours) of energy in 2022. By 2026, according to 8billiontrees.com, energy consumption in data centers will reach 1,000TWh, and by 2030 the total ICT emissions, including connected devices, will reach 2,500TWh.

The current reporting from ICT tech giants supports these projections—despite its public commitments to climate, Google has disclosed its increased greenhouse gas emissions and that AI’s energy consumption at Google data centers puts its climate commitments at significant risk. Microsoft has reported much of the same—despite making its “moonshot” pledge in 2020 to achieve zero emissions (or lower), the corporation has instead been seeing its overall emissions rise, and in May 2024, president, Brad Smith was plain Bloomberg regarding their climate goals, saying, “That was before the explosion in artificial intelligence.”

What Prominent Organizations are Doing About Their AI Energy Consumption

In light of all this, many organizations are attempting to better their carbon management, and that means finding helpful “levers,” or tools and mechanisms they can use to reduce emissions and achieve their science-based carbon reduction commitments. Examples of levers include:

• Purchasing renewable energy or carbon capture;
• Product recycling and circularity;
• Investment in sustainable technologies; or even
• Assuming policy commitments and adopting leadership positions where appropriate.

More specifically, here are what some larger names in ICT are doing to combat the GHG emissions driven by their AI use:

• Google remains committed to their aforementioned goal it set for itself in 2020 to rely on carbon-free electricity by 2030, but given that their research sees AI as having the potential to help mitigate 5% to 10% of global greenhouse gas emissions by then, they also remain bullish on the importance of allowing AI to scale as there are indications that suggest an energy reduction potential of 100 times for AI training and 1,000 times less for AI use.

• OpenAI founder Sam Altman, among other investors, invested $20 million in Exowatt, a company that uses solar power to help meet the needs of OpenAI data centers.

• Salesforce announced in April 2024 that it is lobbying for new regulations to compel companies to report AI emissions data and efficiency standards as part of its Sustainable AI Policy Priorities.

• In 2023, AWS started to transition away from diesel used in its back generators in Europe and the US—sites in Ireland, Sweden, and Oregon were the first to adopt hydrotreated vegetable oil (HVO) which has 90% of the carbon impact of diesel.

How Schellman Can Help Mitigate Your Organization’s AI Energy Consumption and Emissions

Though the larger enablers of AI appear to be highly cognizant of how AI’s growth can outpace the momentum towards their achieving carbon neutrality at current projections and are leading the way toward more environmentally conscious solutions, they’re not doing it alone.

As we consider addressing the carbon impact of ICT and the software industry to be a natural extension of our commitment to a secure and sustainable future, part of Schellman’s work as a leading provider of cybersecurity, privacy, and AI attestation and certification services involves assessments for different organizations regarding their commitments to responsible AI products and services—environmentally and otherwise.

Climate change is a global responsibility, and if you’re seeking to do your part by gaining an understanding of how you comply with emerging ESG and climate disclosure regulations, or how you can meet the sustainability expectations of investors and clients, we can help you too, whether you’re subject to the requirements of the EU CSRD, or you’d like to know how prepared your organization is against established ESG frameworks such as TCFD, SBTi and ISSB.

There’s a lot to be gained from an investment in ESG, and when you leverage our deep, technical, and normative expertise along with our legacy of industry insight, we can support your organization by applying the highest levels of ESG and sustainability scrutiny and impartiality required to instill confidence in your investors, clients and regulators regarding your environmental decision-making, target-setting, assurance, and certification of all your sustainability commitments.

Getting Started with Combating AI-Driven GHG Emissions

AI energy consumption has everyone worried, be it the media or large ICT corporations seeing their ambitious climate goals threatened by the increasing use of artificial intelligence. And until AI progresses and becomes more efficient, it’ll be everyone’s responsibility to do what they can to curb the emissions driven by it and preserve our planet.

There are many different roads to take when attempting to shift your organization toward sustainability, and we’ve written in more detail about some:

• ESG Reporting - A Comprehensive Guide for Businesses
• Greenhouse Gas Calculations vs. Greenhouse Gas Verification
• What are the New SEC Climate Disclosure Rules?

At the same time, your stakeholders aren’t just looking for environmentally cautious organizations, so you may want to combine your ESG AI efforts—including perhaps a comprehensive environmental management system (ISO 14001)—with third-party certification for:

• Responsible AI (ISO 42001);
• Quality management (ISO 9001);
• Business continuity (ISO 22301); and
• Information security (ISO 27001).

To learn more about our ESG and related capabilities, schedule a one-on-one session with Schellman’s AI and ESG directors Danny Manimbo and Tom Andresen-Gosselin.