Artificial intelligence is rapidly reshaping how we live, work, and communicate — but as excitement around its potential grows, so do concerns about its environmental footprint. At the same time, global carbon emissions continue to rise, raising important questions about where AI fits into the broader climate conversation. In this post, we explore whether the growing use of AI is part of the problem, part of the solution — or perhaps just an easy target.
Global Emissions Are Rising — And So Are the Stakes
Despite global pledges to curb emissions, the numbers tell a different story. In 2024, global carbon emissions hit a staggering 41 gigatonnes of CO₂ per year — a 14% increase since 2020. This upward trend continues even as governments and industries promise decarbonization.
Meanwhile, Earth is heating up. For the first time on record, global average temperatures in 2024 approached the critical +1.5°C threshold — a limit scientists warn is essential to avoid the most catastrophic impacts of climate change. But based on our current emissions trajectory, we may be heading toward a much hotter future: between 2.7°C and 4.4°C of warming by the end of the century [IPCC 2021, multi-model mean Delta T].
The energy hungry elephant in the room?
AI has captured imaginations, we’ve written and deployed AI algorithm techniques in our climate analyses, and now, AI is starting to make an impact on our power grids, too. While it’s difficult to pin down exact numbers, recent studies estimate that the energy consumption of AI-related data centers accounts for 2.5% to 3.7% of global emissions — outpacing even the aviation sector.
As the AI revolution accelerates, so does its carbon footprint. Training large language models, running continuous cloud services, and powering vast data infrastructures all come with environmental costs that are increasingly hard to ignore. In Ireland, for example, more energy is consumed to power data centres than urban homes.
NVIDIA’S H100 GPU. Powerful, power-hungry and powering our AI queries. [Image credit: NVIDIA]
Is AI a Climate Problem — or a Hidden Solution?
At first glance, the intersection of AI and the climate crisis can look bleak. With headlines warning of energy-hungry data centers and the growing carbon footprint of generative models, it’s easy to cast AI as another accelerant in our global emissions spiral.
But that’s only part of the story.
A deeper — and often overlooked — question is: what is AI actually being used for?
As of late 2024, nearly 40% of U.S. adults aged 18–64 report using generative AI tools either for work or recreation. Interestingly, a third of users (33.7%) engage with AI outside of work — often in ways far removed from its originally envisioned purpose of boosting societal efficiency. Think more “turn me into an action figure” than “optimize a wind farm.”
Reframing AI’s Carbon Footprint
In reflecting on this over the past few days — and diving into some recent research — one point stood out: AI’s current carbon emissions need to be weighed against the counterfactual. In other words, what would global emissions look like if AI had never been developed or deployed?
It’s a provocative question. And the answer, surprisingly, might be that emissions would be even higher.
Across industries, AI is delivering real-world efficiency gains that directly reduce fuel use and CO₂ output:
General Electric reports that AI has slashed fuel consumption by 15%, leading to 20% lower emissions at some facilities.
In Germany, Siemens credits AI optimization for 10% improvements in energy efficiency in gas turbines and industrial operations.
In the UK, Google’s DeepMind has refined electricity demand forecasts, helping avoid fossil-fuel-based backup power altogether.
UPS saves a massive 10 million gallons of fuel per year through AI-powered route optimization — avoiding more than 100,000 metric tonnes of CO₂ annually.
These aren’t hypotheticals. These are measurable, sovereign-level savings that don’t often show up in the growing wave of AI skepticism.
AI algorithms optimises gas turbine fuel valves – in real time – leading to more efficient combustion and less waste / emissions.
Use Matters: Memes vs. Megawatts
So what’s the catch? It may all come down to use case.
AI use for entertainment — think avatars, memes, and novelty chatbots — doesn’t offer meaningful CO₂ offsets. It’s pure consumption. But industrial and infrastructure-scale AI? That’s where the climate dividends are real — and potentially massive.
The question then becomes: Do the carbon savings from these large-scale applications outweigh the emissions of recreational use? It’s hard to say for sure, especially given the lack of detailed breakdowns by sector. But initial research suggests that not all AI is created equal — power efficiency varies wildly between model types and applications.
A detailed analysis partitioning emissions by recreational vs. vocational AI use would be hugely valuable. Even better: quantifying which sectors generate the most carbon savings from AI and which are adding to the load.
China’s DeepSeek claims extraordinarily low power consumption. It’s driven by NVIDIA H800 chips since NVIDIA H100s – more powerful but more energy hungry – are prohibited for export. [ Image credit: NVIDIA ]
Got Data?
If you’ve come across any great studies in this space — especially ones that dig into the real-world emissions trade-offs of AI — drop them in the comments. We’d love to see them.
– Craig Wallace, 14/4/2025
At EarthSystemData, we use advanced, self-training AI algorithms to process large, multi-parameter climate and weather datasets. Our goal? To help clients diagnose and quantify industry-specific risks linked to climate variability.
In this context, AI functions as a powerful data reduction tool — clustering similar weather conditions into smaller, more meaningful subsets. These refined datasets are far more effective when correlating environmental patterns with business-critical metrics like water availability, crop yields, energy demand, and more.
Curious about the technical side? Dive deeper into K-means clustering here and here and further high-dimensional analysis techniques we use for large-scale assessments.
If you’re looking to understand — and reduce — your organisation’s exposure to growing climate volatility, we can help. Get in touch today.