EY helps clients create long-term value for all stakeholders. Enabled by data and technology, our services and solutions provide trust through assurance and help clients transform, grow and operate.
At EY, our purpose is building a better working world. The insights and services we provide help to create long-term value for clients, people and society, and to build trust in the capital markets.
In this episode of Sustainability Matters, host David Rae, EY Global Lead for Sustainability Technology and Innovation, explores the impact of AI on climate change. As AI companies build massive data centers worldwide, a critical question emerges: Will the skyrocketing resource footprint of AI push global grids past their limits, or will AI become a vital tool to help accelerate the energy transition?
David is joined by James Grabert from the UNFCCC (United Nations Framework Convention on Climate Change) and Michael Lepech from Stanford University to break down the duality of this digital megatrend. Together, they look at the physical reality behind "the cloud," tracking how AI is driving a massive surge in electricity demand while also leaving an extensive water footprint through data center cooling and the lifecycle of chip manufacturing.
However, they also identify areas of opportunity. The episode offers a hopeful look at "decision intelligence." The guests explain how AI is already delivering measurable climate benefits — from transforming early warning weather models in the Global South to balancing power grids and optimizing renewable energy.
Key takeaways:
AI is driving a massive surge in data center electricity demand and water consumption, creating local grid bottlenecks.
However, AI can deliver clear benefits for climate adaptation and mitigation by optimizing renewable energy grids and transforming early warning weather systems in the Global South.
To protect the planet, organizations should avoid narrow KPIs and utilize lifecycle-based tools to analyze AI against total systems impact.
For your convenience, full text transcript of this podcast is also available.
Michael Lepech
We want AI to serve our most basic human aspirations. We want it to do good things for us. If it can provide real-time updates to hazardous situations and save people's lives, I think we would all agree that that's what we want it to do.
James Grabert
Used well, AI can help us navigate a narrow path through our transition. Used poorly, it becomes just another source of pressure on a system already close to its limits.
David Rae
Welcome to the EY Sustainability Matters podcast. I’m David Rae, Global Head of Technology, AI and Innovation for EY Climate Change and Sustainability Services, and your host for this episode.
This is the second in our special series exploring the duality of AI and sustainability. In our first episode, we looked at nature. Today, we tackle the "light and the shadow" of climate. Right now, we are at the intersection of two tectonic movements: the race to a net-zero, nature-positive future and the rapid diffusion of AI into every aspect of our lives.
In this episode, we are asking: Is AI a climate solution, a climate risk — or actually a bit of both? To help us navigate this, I spoke with two distinguished experts. First is James Grabert — he has been working at United Nations Climate Change (UNCC) for more than 20 years. Currently, he heads the Mitigation Division there, leading the work from market-based approaches to climate change mitigation. I also spoke with Michael Lepech, a faculty member at Stanford University. Michael is a leading voice at the intersection of civil engineering, sustainability and innovation. And his work focuses on reimagining the built environment to be carbon-conscious.
David
So, let’s go ahead with the discussion. When we talk about AI, we often use ephemeral terms like "the cloud," almost as if they’re weightless. But the physical reality is much heavier. We are seeing a massive dislocation, where AI-driven demand is colliding with stresses on our power and water systems. James has been tracking this growth curve, and the numbers are staggering.
James
Indeed, this concern is growing.
What we're seeing now, the AI models with hundreds of billions, perhaps even trillions, of parameters and platforms processing billions of prompts every day. And that has translated into a very sharp rise in data center electricity demand. And in some regions, data centers already account for a quarter of the household-level electricity use.
For example, data centers consumed around 415 terawatt hours in 2024, which is about 1.5% of the world’s electricity consumption already. So, that may sound modest until you look at the growth curve. They say data center electricity demand is projected to more than double, to about 945 terawatt hours by 2030, which is significantly driven by AI.
David
But there are gaps in the conversation. What some miss is that it isn't just about the meter running at the data center; it's about the entire lifecycle of the infrastructure. Michael argues that we are often oversimplifying how we view this consumption.
Michael
Yes, running the data center is energy-intensive. But as we shift to lower- and lower-emission energy sources, that piece becomes bigger and bigger. And that's not just what it takes to construct the facility. That's everything that it takes to produce the chips and the semiconductors and all of the IT infrastructure. And that is truly substantial.
I think we've all been on our favorite LLM, and we asked it one query and we didn't stop there. As humans, we're naturally inquisitive, and so we keep querying. And that's a perfect example of the rebound effect, where you don't ask it one question. Once you ask it one question, you continue to ask it questions. And that will only increase demand on these systems. And I don't think we've fully appreciated what effect that's going to have.
David
And it’s not just electricity. There is another constraint that matters just as much, and it’s easy to overlook until it becomes local and immediate: water.
A few things about the water consumption: Without a doubt, and it's wonderful to see it getting more attention, AI and data centers globally consume tremendous amounts of fresh water, a couple of percent.
But even that number is difficult to measure because one of the first questions for the direct on-site water use, right, is, is this water actually consumed and not returned to the watershed, evaporated off as part of cooling? Or is it direct cooling and therefore doesn't leave the watershed? And so, the way we think about what does water quote “use” even mean can be confusing to folks.
And then offsite, you know, we have electricity generation. Well, that electricity generation, where does that cooling come from? There is a big water footprint there as well. And then, of course, there is water consumed in the manufacturing of chips and servers and construction materials. And we very rarely, if ever, count that. But it can be substantial.
And that's on the sort of the downstream side of the water, so to speak. On the upstream side, are you using potable water? Are you using reclaimed or recycled water? Are you using groundwater? If you're going to use seawater or brackish, we need to talk about desalinization, special treatments, and all of that adds additional energy and costs.
This brings us to the critical issue of equity. As James points out, it’s not just about how much water and power we use, but where that demand lands, and who bears the costs when systems get pushed to the limit.
James
The AI infrastructure is highly concentrated geographically. A small number of regions now carry a large share of the demand, creating local grid bottlenecks, higher electricity prices and honestly, a public backlash when upgrade costs fall on households. So, and this isn't limited to advanced economies. Data center growth is accelerating in emerging markets as well, where grids are often constrained and very fossil-heavy, and where millions of people still lack reliable electricity. So, allocating scarce power to AI rather than basic service raises real development and political risk. The key point is that this energy is no longer a background issue for AI. It is becoming a binding constraint that affects costs, growth, timelines and social cohesion.
David
This mismatch in timelines — the rate of AI growth versus the speed of our grid expansion — is forcing us to look at every option in the power production playbook. And that includes nuclear.
Michael
The mismatch in the timelines is concerning, if not alarming. The rate of growth of the demand, by some estimates, is going to 1,000 terawatt hours in the next few years annually, and we will not be able to match that. And so therefore, we're truly going as far and as deep in the energy production, power production playbook as we possibly can, right? Which leads us, of course, to nuclear and some other sources.
Small modular reactors, the technology is there. We know how to do this, right? And this is Gen 4, right? This is not Chernobyl. This is not the kind of nuclear accidents that folks get really concerned about. But the permitting, the regulatory compliance — there are most places in the world where, number one, that's not allowed. And also, as we think about neighborhoods and — do you want to live next to a small nuclear reactor?
David
I asked James if nuclear can actually bridge that gap between supply and demand, or if it risks becoming a distraction.
James
Well, nuclear can play an important role, but it's not a stand-alone solution. The case for nuclear today is stronger than it has been in recent times. It provides stable, carbon-neutral power at scale, which explains renewed interest from major technology companies and investors. But there are real constraints. New nuclear capacity typically takes 10 to 15 years to come online. While again, AI-driven electricity demand is growing now — it's happening now.
Nuclear can help as a longer-term anchor in some countries alongside renewable energy grid expansion and demand flexibility. But the near-term risk is that nuclear later becomes an excuse for more fossil fuels now. And the real question won't be nuclear: yes or no. It will be “What is the credible, economical and equitable pathway to serve growing electricity demand without compromising our climate goals?”
David
This is the duality we must manage. And if we are to ensure AI serves the planet, we have to bridge the transparency gap and get a clearer view of its real footprint. As James outlines, you can’t manage what you can’t measure.
James
A central problem of the industry is the lack of transparency. The key details, like which data center processes a request or how much energy it uses, or the carbon intensity of the power — these are largely not known or are known only to the companies behind the models. And of course, that makes it difficult to pin down the impact of the sector.
At a minimum, policymakers need facility-level energy data from data centers, including the sources of electricity used. Annual averages are no longer sufficient. Time-of-day data matters because it determines whether AI workloads are sometimes being met with clean power or with fossil generation.
The evidence is increasingly saying that AI climate impacts won't be decided by the technology alone, but how we power it, where we cite it and what we choose to use it for. So, from a climate governance perspective, governments should treat AI like a new class of heavy infrastructure, because the scale is now comparable to that of other major industrial loads.
David: Because if we get that governance right, the opportunity side of the evidence starts to show up in our climate resilience.
This is where the conversation becomes more hopeful. So, looking at the evidence, I mean, there are two clear areas: early warning systems and the actual energy system operations itself. So, on early warning systems, AI-based weather models are delivering major gains already, often matching or outperforming traditional forecasting methods at a much lower cost.
And that is transformative for adaptation, especially in the Global South. So, better flood, cyclone and heat wave warnings — they save lives and significantly reduce economic losses, particularly in regions that have historically lacked that forecasting capacity. And again, in energy systems, AI is already delivering measurable operational benefits. It improves renewable generation forecasting, demand prediction, grid balancing, dynamic line ratings and predictive maintenance, among other things. So, it is accelerating the energy transition.
What we're also seeing is that AI is strengthening climate monitoring, from tracking deforestation to detecting methane leaks and accelerating clean technology innovation, including batteries, hydrogen and carbon capture. So, a lot is happening.
David
Michael sees this as a path toward "Sustainability Intelligence," but he does warn that we can't use these benefits as a rationalization for waste.
Michael
Yes, we're going to consume resources — water — but if we use it to do water-saving activities by creating new intelligence and new optimization of industrial systems, there is a pathway there to thinking about sustainability.
But in many regards, it's the same argument as, “I'm going to produce a vehicle and drive it around with an internal combustion engine, but as long as I'm doing good things with that car, it's okay.” And in some regards, it can be a rationalization of why we're doing it. Yes, we absolutely can see a pathway there. And I think we should be thinking about the use of, in particular, the largest foundational models to solve those toughest problems.
David
Which brings us to a bigger question: not just what AI can do, but what it should do, and how consciously we choose to use it.
Michael
I don't need the entire internet to help me answer pretty mundane questions like when do the Pistons play tonight from a basketball standpoint, right? I don't need an entire large language model to answer that question. And so, how we use it, what we use it for, I think, is going to be a very important discussion that will balance the resource consumption questions as well.
You know, I often ask my students and other groups, do you say thank you to a large language model? And almost everyone does because we anthropomorphize this tool. It talks like us, it seemingly thinks like us. And yet, that is a tremendous waste of resources and energy and water to respond to that unnecessary query.
David
And there is another important point here: The companies building and running these LLMs are also the ones paying the power bills, which gives them a massive incentive to reduce the energy intensity across these systems. That could mean introducing lightweight front-end triage intelligence that can determine whether a user simply wants a quick factual answer — like what time the Pistons are playing tonight — without triggering a full large language model query every time.
In conclusion of our conversation today, I think one thing is becoming clear: AI for climate must be validated against total systems impact — not just narrow-focused, siloed KPIs.
At EY, we are bringing that systems view through our tools, like the Green Tech Optimizer, a tool that is designed to help measure, optimize and report the footprint of technology solutions. Using these tools, we help organizations design and deploy AI with an efficiency-first mindset and outcomes that are aligned with climate goals.
We have a real opportunity to design AI to work for climate rather than against it. But that only happens if we build it with clear guardrails and measure impact across the whole system — not just the outputs. Thanks to James Grabert and Michael Lepech for joining us. In the next AI special, we’ll be discussing the "social effect" of AI — equity, the workforce and the future of our communities.
Tell us your thoughts and your challenges. Your voice matters. Thank you for listening to the Sustainability Matters podcast. If you enjoyed this episode, please check out previous episodes on ey.com or wherever you get your podcasts.
We'd love for you to subscribe, and ratings, reviews and comments are also very welcome. Visit ey.com, where you'll find a wide range of related and interesting articles to help put these bigger topics in the context of your business priorities. And we look forward to welcoming you to the next episode of Sustainability Matters.
My name is David Rae. You can find me on LinkedIn, and feel free to connect with me there. Thanks so much for listening.
