How rising data-center demand is reshaping energy systems | Zurich Talks - Risk to Resilience

AI-driven data centers are pushing energy demand to new extremes – raising urgent questions for risk managers, policymakers, and business leaders alike. How can energy systems adapt without increasing fragility and emissions? Hear from experts including Google, the IEA, a scientist and Zurich Resilience Solutions.

Transcript

[Title: 0:02

Meet prepared] Meet the future prepared] Meet innovation prepared] Meet climate risk prepared] Meet unexpected prepared] Meet disruption prepared] Zurich Talks - Risk to Resilience, May 4 - Zurich, Explore how climate risk is shaping resilience decision-making globally. ]

[Title: Zurich Logo. Disclaimer: 0:14

This video has been prepared by Zurich Insurance Group Ltd and the opinions expressed therein are those of Zurich Insurance Group Ltd as of the date of writing and are subject to change without notice. This video has been produced solely for informational purposes. All information contained in this video have been compiled and obtained from sources believed to be reliable and credible but no representation or warranty, express or implied, is made by Zurich Insurance Group Ltd or any of its subsidiaries (the ‘Group’) as to their accuracy or completeness. This video is not intended to be legal, underwriting, financial, investment or any other type of professional advice. The Group disclaims any and all liability whatsoever resulting from the use of or reliance upon this publication. Certain statements in this publication are forward-looking statements, including, but not limited to, statements that are predictions or indicate future events, trends, plans, developments or objectives. Undue reliance should not be placed on such statements because, by their nature, they are subject to known and unknown risks and uncertainties and can be affected by numerous unforeseeable factors. The subject matter of this video is also not tied to any specific insurance product nor will it ensure coverage under any insurance policy.

Pauline Lowe, Director Public Affairs, Zurich Insurance Group: 0:16

Hello everybody and welcome. This is the first episode of Zurich's Risk to Resilience, a series that follows Zurich's sustainability engagements across the world, tracking the key climate moments that we observe. Our aim is to bring Zurich leaders and external experts together to explore the anticipation of climate risks and how we build resilience. Now, without further ado, today our first episode is on planning power for the AI economy. AI is driving one of the fastest surges in electricity demand in decades and data centers are scaling at unprecedented speed. And the question no longer is, will that demand continue to grow, but rather will energy systems be able to absorb that growth without increasing emissions and systemic risk? So to take us through some of these big

Lindsey Hall, (Moderator) Global Head of Thought Leadership, S&P Global: 1:02

Thank you so much Pauline. Thank you Zurich Insurance for the invitation to be here today. My name is Lindsey Hall and I am the head of the sustainability research lab at S&P Global where I also host the All Things Sustainable podcast. You've just heard our theme music which if you're a listener you might recognize and if you are not a listener you can subscribe and listen anywhere you get your podcasts. And I’m really excited that we are the official podcast of this inaugural Climate Week Zurich. I am also really thrilled to be sitting down with an amazing panel to talk through some really important topics today for this conversation. In my role and in my podcast interviews, one of the main questions that I hear, one of

the main trends that we cover is: 1:46

How do you balance the immediate needs like energy security and availability. And rising demand from AI with the longer-term scientific realities of climate change and nature loss. So that's what we're going to be talking about with our excellent panelists today, and I'm going to invite them to come on up and join me, please. So we will be hearing from Dr. Enass Abo-Hamed, who is a scientist and entrepreneur with H2GO, Adam Elman, who's the Sustainability Director at Google, your very own Emmanuel Mtika, who is Global Head of Sustainable Energy at Zurich Resilience Solutions, and Cecilia Tam, who's Head of the Energy Unit at the IEA. Thank you all for sitting down with me today and really looking forward to this conversation. Cecilia, if we could start with you for some scene setting, can you offer some high level perspective on how the IEA is thinking about the energy system? What are you seeing in terms of trends and demand.

Cecilia Tam, Head of Energy Unit, International Energy Agency: 2:48

Thanks very much, Lindsey. So at the IEA, we've recently started doing quite a lot of analysis on the impacts of AI on energy and energy on AI. And this started last year with a major report that we released on some of the findings that developments in technology and AI in particular was going to have and is having on the energy sector. And just recently, we provided an update because it is an extremely fast-moving area. Technology innovation is really driving the space, both in terms of what is happening amongst AI firms, but also how companies, the energy sector, are really responding. So just to share with you some key numbers in terms of what we're seeing. Energy demand last year from data centers rose about 20% year-on-year. But if we look at AI-focused data centers, that percentage actually increases to 50%. And what's quite important to highlight is that traditional data centers and AI data centers operate very differently. And they're really creating new challenges for the market. So, a traditional data center consumes power reasonably stably, whereas what we've seen for AI data centers is that fluctuations in power consumption in just a second can vary up to about 50%. And this is really calling for more investments into stabilizing the grid and for frequency regulation. And one of the things we did in the major report that we released last year, we really try to unpack this question of how rapidly or how much electricity will data centers AI lead to? And it's a very difficult question to answer. But our base case scenario really sees that energy from data centers over the next five years, so to 2030, could roughly double over that time period. If we look at energy from AI-specific data centers, we're anticipating that under a base case scenario that could triple by the end of this decade. Now, if we look at what this means for spending for energy related to AI usage, we anticipate that over that period, we would need to increase investments in the energy sector primarily for power for somewhere between half a trillion and one trillion dollars. So that sounds like quite a big number out to 2030. But if you look at over that same period of time, how much the world will spend on energy overall, that's roughly 18 trillion dollars. We've see some really interesting trends over the last year in terms of tech companies trying to secure power and low carbon power sources with big increases and developments in. Renewed interest in nuclear power, both in expanding or extending lifetime of existing operators and also for developing and investing in small modular reactors. And then more recently we've also seen a big jump in the demand and orders for gas turbines. Really led by AI companies looking to source more on-site power demand, including from gas-fired power plants.

Lindsey Hall: 6:38

Thank you for setting the scene so effectively, and yes, there were a lot of really compelling numbers in there about tripling of demand and huge investment number that's needed. Let me jump to you, Adam, just to bring in the Google perspective here. How do you respond to what you're hearing from this IEA perspective? How does this gel with how you're thinking about building out to meet the growing demand from AI and energy.

Adam Elman, Sustainability Director, Google: 7:05

Thank you and good morning and thank you Zurich for hosting this important conversation. Look, I think that the numbers for the IEA are hugely important and we actually were a partner with the IEA in helping to make sure the right numbers are out there. Certainly as you've heard sort of the power demand for AI infrastructure and data centers is growing as you know the opportunity for you know industry and for you know science and health and education sort of grows. I think what's really interesting though is you've had, whilst it's growing, data centers around the world, all data centers, which AI is a fraction, is about, I think it's one and a half to 2% of energy demand. And whilst it is growing, and it's certainly growing, I think is only sort of a tenth of the growth. So you've still got a growth coming from industry, heating, cooling, transportation. So we're certainly growing but in perspective of entire industries that are looking to electrify, which clearly is a huge part of sort of decarbonization. Look, from a Google perspective, we've been focused on this for a long time and the energy supply side is really important, but the first thing we think about is efficiency. Obviously, the cleanest energy you use is the energy you don't use and we're very privileged. We're a full-stack player, so we can look at every part of our operation, right down to the hardware, the infrastructure itself. We develop our own processors, our own TPUs, Tensor Processing Units. Those are the processes that are powering AI. Our latest processor, which is Ironwood, is 30 times more power efficient than our first one in 2018. So there's a lot we can do in terms of the infrastructure. In fact, you know, Google data centers are some of the most energy efficient in the world. To put it in perspective, compared to five years ago, we have almost six times as much compute power for the same amount of energy. So we're really good at driving efficiency. By the way, it's a huge cost for us. It's good for business as well as being good for the planet. We also focus on the models themselves. So. If you've heard of sort of Gemini, which is our sort of core foundational model, that continues to get more and more efficient. Over a 12 month period, when we look at sort of a median text prompt, it's 33 times more power efficient and 44 times more energy efficient, so more carbon efficient. So the infrastructure's getting more efficient, the model's getting efficient, and then of course we need to look at the energy that's powering our data centers. transition.

Lindsey Hall: 9:24

Thank you. Can I dig in a little bit more just to understand how is Google balancing its sustainability goals, what those goals look like with the growing demand? Like how are you making sure that the sustainability goals are not lost in the shuffle?

Adam Elman: 9:41

Yeah, and look, this is not easy. You know, we're growing very fast. And again, you know, the grid, you know I don't need to tell this audience, the grid wasn't designed for you know the current needs. But the investments we're making are helping us stay ahead of this. You know in 2024, eight gigawatts worth of new clean energy capacity, which is significantly more than we actually need. So we're actually trying not just to keep up, we're getting ahead of our demands. Again in 2024. Our electricity use grew 27%, but our data center emissions reduced 12%. So this is not easy by any means, but it's through innovation, it's your investment, it's through partnership with the energy sector, with other players, right across the piece.

Lindsey Hall: 10:26

Thank you. You talked about the importance of sort of building something right from the start. And I think that's Emmanuel, maybe where we could bring you in. I mean, how are you thinking about the need to design something in the right way, right from the start?

Emmanuel Mtika, Global Head Sustainable Energy, Zurich Resilience Solutions: 10:41

So in the last five years, we provided risk transfer solution coverage to over 500 projects with a total insured value of about $350 billion in the U.S. alone. And in the last year alone, we've provided coverage to about 200 projects, that is in the projects. At a total insured value of$150 billion. So really, the insurance sector as a whole has stepped out to provide that coverage. But I sit alongside the underwriters that do that. I sit where I look at upstream, right? Looking at, when I look all this, it's how do we scale? We scale by really understanding the risk, right. So understanding what fails, how it fails, and how much it costs actually to recover when things fail, right? So this means that when we have that clarity on risk, that we provide the mitigations, the adaptations very early on at design stage, right. So it means that from very early design, we understand the system. How the asset itself and the surrounding system, how robust it is to withstand shocks, right? So it could be a nut-cut exposure, you know, could be extreme weather events. All those aspects could be systemic issues there that the asset, itself, and the system around it must be able to absorb those shocks. Then you're looking at, you need to design it in such a way that you have redundancy within the system that you eliminate all those single point of failures. And if you're looking at power supply, for example, that you are able to look at multiple supply points for your power, for cooling the same thing. So understanding all that through the design phase, think about how that will play out in the construction phase, in the commissioning phase into operation, right? So you understand all that, then we'll put in place there. You need to have the resourcefulness, right? You need to be able to have the skilled manpower labor force that understands how data centers work, that have got the experience then to really track potential issues, anticipate those issues, perhaps aided by AI to anticipate all the issues and put in place remote condition monitoring, right, condition-based maintenance. But help to anticipate issues much earlier and putting the right mitigations in there. So being able to do that in an early stage is key for us. So we support our customers from very early stage that we can put those mitigations.

Lindsey Hall: 13:47

So starting off with that really early stage system perspective. Enass, I'd like to turn to you because I'd like to understand from an energy system perspective, where do you see this rapid data center build out, putting the most strain on grids and low carbon power systems?

Dr. Enass Abo-Hamed, Scientist-entrepreneur, H2G0 founder: 14:07

Thanks for having me first, and thank you for bringing me last to the conversation, because I'd really like to build on what's already been said here. If you want to look at the numbers, and these are stolen numbers from IEA reports, but I was told that if you confess to stealing data and say it was the source, that's fine. I mean, in 2024, and that's not contradictory to any numbers that's been shared already, the global electricity consumption that was derived explicitly by usage of data centers was around 415 terawatt hours, do you approve? And that's around the consumption of Italy as a country. Now, by 2030, which is by the end of this decade, it's going to double or exceed the double by a little bit to around 950 terawatt hours. Mind you, the end-of-the-decade is four years, three and a half years away. So there's a bit of a contradiction between the growth of that demand and our physical capability to keep up with that growth because we are driven by the applications and what AI pace is bringing to our world. So these are not infrastructure curves. These are rocket trajectories. And I think we really should pay attention to the opportunities that could come out of this, as well as feeling that uncertainties here might kind of clash with how we think about insuring things. We could also redesign things that are insurable and guarantee opportunities for further growth. Now where does the strain get stuck? I would say three places have already been mentioned. First is the capacity, the clash between physical limitations and how fast AI moves. Connection queues would take about a decade. We're talking about by the end of the decade of how we're going to double our demand. That clash is significant. The second one is the concentration. If we're spreading the concentration across national grid if we have a problem. The problem is spread across. But if we concentrate where data centers and the power around them could actually exist, we might be able to solve solutions in local AI zones, and that is an opportunity to look at. And the third point, it's been mentioned as flexibility, but because I wanted to highlight the problem, I would call it inflexibility. Data centers run flat, 24/7. That's the exact opposite of what renewable grid demand would need. But if we use these three points as data points that allow us to redesign a system from a system perspective, inflexibility becomes flexibility. And with technology and innovation, and we've already solved that with peak shaving. Storage on site and demand response solutions. So that kind of like, that was a catalyst to create a solution which we have partially solved. And the solution is also like, the adoption of the solution, is growing. Second point in terms of concentration, concentration could become co-location where integrated energy zones where heat networks, grids, and industrial clusters, which actually share the same backbone. And that means that we could look at the waste heat created in a... In mass as an opportunity to actually turn waste into value. And that's not something that we've done much of so far. And I think this is a massive opportunity to be looking at where we can take waste heat and turn it into value across. I mean, if we start from London, London has the ability only from waste heat that is recoverable coming from data centers to heat about half a million homes without additional investments. This is infrastructure that exists, demand that exists, waste heat that exists. What's missing is the policy that enables connecting the dots together to put together a functional solution that could be used.

Lindsey Hall: 19:27

I love this focus on the opportunity side of things. Emmanuel, did you have something you wanted to add here?

Emmanuel Mtika: 19:32

Yeah, so I was going to say from an engineering point of view, the initial phase has been about optimizing the operation of the data center as an asset, almost standalone, right? And now I think we need to move more to be able to think about how we harness the worst hit. We need to think in a system kind of concept, right? So, thinking not just the data center, thinking about the entire electricity grid on which actually the data center that depends and the communities where it's located. When you do that, you start seeing the ability to connect those dots, right? So you start saying could be waste heat, could be renewables. So the demand signal that you're getting from the data center is good for renewables, but actually when you bring renewables on the grid because of their nature, you bring another layer of risk, so you make the risk more complex than it was before. But if you take that as a design decision from day one, you can fit that in, you can ensure that the waste heat is put in, you can insure that data centers are designed with a flexible load management system.

Lindsey Hall: 20:50

What do you think, Adam, do you agree with that?

Adam Elman: 20:52

No, absolutely. I mean, I think that the sort of heat story is a really important one. I'm just back from Finland, actually, a few weeks ago. So our data center in Hamina there, we've partnered with Hamina Energy, the local operator. We give them the heat from our data centers for free. It will cover about 80% of the needs for about 2000 homes there. Because of the work we've done on clean energy, that's 97% carbon-free heat. So it's a great story. We're building our data centers heat recovery ready. Our latest data center in Germany will be doing the same. But actually this is where there's a sort of bigger story which is there is only so much we as a data center operator hyper-scaler can do. You need the right infrastructure, the right policy framework. So we can build a data centers that's heat recovery ready, but if there isn't an off-taker and the infrastructure there, then it's not getting used. So again, this is why. Partnering with others, having the right policy frameworks in place is so important. So we want to be good citizens, we want be good neighbors, and we're doing as much as we can do, but in many instances, there's only so much we can do and we need that bigger system change.

Lindsey Hall: 22:00

You talked about policy as did and us and maybe Cecilia I can bring you back in here because the IEA spends a lot of time talking to regulators and policymakers. What concerns do you hear from them? What conversations are you having?

Cecilia Tam: 22:12

I think there's a variety of things. One is really about the outlook for overall energy demand growth and how that's going to be met in a secure and sustainable way. As I mentioned in my first intervention, the exact growth is hard to forecast because the technology and innovation is evolving so rapidly. So a lot of electricity regulators, countries, are concerned about overbuilding if the power demand does not materialize. Today we talk a lot about the physical bottlenecks that are emerging from the very rapid growth today in electricity consumption and people are building a lot of power both on site and to meet this growing demand. But we also need to be aware that there are a number of other physical bottleneck that are in the system. In terms of being able to source and supply quickly the electrical equipment and two really come to mind. One, you know, the delivery times for transformers right now vary between two and three years or more. Gas turbines delivery right now is about five years and as already mentioned about the long connection queues. Which can vary between five to 10 years. And policymakers and regulators are starting to take action to address these bottlenecks, including looking at reforms to connection procedures and wait times. There's also a lot of push back around public sentiment because of all of these uncertainties and knowns of whether this big boost in electricity demand coming from data centers is causing electricity prices to spike. And obviously under current situations, affordability is top on the mind of agendas of governments.

Lindsey Hall: 24:21

Thank you. You talked about some of the community implications for data center build out. So Adam, maybe I can turn to you to ask, what are you hearing in terms of community push back? How are you thinking about working with communities?

Adam Elman: 24:35

Yeah, I mean, it's an absolutely important factor in terms of, you know, where we're growing. I mean obviously data centers come with lots of benefits. You've got the sort of, the workforce benefit, the sort local economic benefits in terms of taxes and property and all those sorts of things. And you know you've got a mix around the world. There are certain places in the world that are very welcoming, there are the others that have The concerns, that's both from policy makers, that's from local communities. We have to think about everything from nature and biodiversity to water. So some of our data centers are air-cooled, others, where it's appropriate, are water-cool'd. Our data center in Belgium actually uses canal water, and actually we return it back to the canal cleaner than we intake it. But then there's all the other things we do in terms of you know the things you'd expect us to do in terms of volunteering, local investments, you know in Denmark we've partnered with the local technical college and so they are training up the data center operators of the future.

Lindsey Hall: 25:41

This is near and dear to my heart. I grew up in Northern Virginia, which is now data center alley. Emmanuel, please go ahead.

Emmanuel Mtika: 25:48

I was going to say, in terms of training and working with a lot of communities, because remember, if you go back 10, 15 years ago in the wind industry, it used to be difficult to find the right skill set of people. In some countries, it was a little bit easier because people were moving from the oil and gas, for example, to going to offshore wind. But the data center is the same now. What we are seeing, actually... We've looked at about 25 major outages of data center in the last 10 years and 80 percent of those were caused by electrical faults, power cooling issues and some of those actually some of the worst ones were caused human factors so in some cases in commissioning phase we had a case in Ireland in commission phase because of the inexperience of the personnel there. No fault of the design or the technology, but actually as you put the different systems and subsystems together, under time pressure and with an experienced workforce, we had a major delay and quite significant costs. So it's important actually as we build out data centers that watch out for those compressed time scales, but actually we've got the right people in place. We're training, we're working with the local community to train the right people. Get them in early so they actually learn and contribute positively to the growth.

Lindsey Hall: 27:20

Thank you. I'd like to end this conversation on a forward looking note and ask all of our panelists, maybe Enass, I can start with you, but as we look ahead, what do you think is the single power system planning or investment decision that we can make today that will determine whether an AI driven economy ultimately strengthens long-term resilience in the coming decades?

Dr. Enass Abo-Hamed: 27:43

If I had to pick one, I'd say that if we just reframed data centers and the world around them that will power them and make them work from a real estate world to a facilitator or a national energy infrastructure framework. I think we could be moving from creating problems to creating solutions and facilitating transition to a healthier energy system that we could live with. What that means is that from a single frame we could bring change to those who apply for permits if they bring solutions with them, including on-site storage, taking care of how they're going to put renewable energy in the ground. You do like a natural filtration to the projects that won't work in terms of their negative Impact on climate. And that is good business to your words. In addition to that, if we were able to kind of spot big problems like heat reuse, what makes communities not only involved in terms of we've created job at the construction phase, but long term, how are we able to lower the heating prices for them and how Are we? Going to lower the prices of energy for them. That is like a gain for the community. So, you know, you'll have less pushback and more engagement. So taking a problem and turning it around into a solution. I really like that. Okay, thank you. How about...

Lindsey Hall: 29:42

Same question to you, Adam.

Adam Elman: 29:45

Two very quick answers. I think one we've just touched on a lot, which is collaboration. Doesn't matter how big you are or where you sit in the system, this is going to require a collaboration from industry to policymakers, to communities, to investors, to startups. How do we drive that collaboration? On the other side of the fence, I'm the Google guy, so I think AI is a key part of the solution. It's not the only part of it, but it's a key point of it. I think within that, data is really, really, key. There are huge amounts of hidden data sets. That need to be out there so that innovators, AI companies, grids and operators and what have you can better plan for the future.

Lindsey Hall: 30:23

Thank you. Emmanuel, what do you think?

Emmanuel Mtika: 30:25

So we need to be deliberate from design stage that we have resilience by design embedded in how we build out data centers. That means we ensure that the data center itself, the system supporting the data center is robust. We have the redundancy in the system that we we have the real time sensors that can give us the data predictive analytics that can help us to anticipate where the issues are going to be and put in the measures in place, and lastly that we were able to recover quickly, that we have the right people in place. Trained people, you're doing a good thing in making sure that you've got a lot of people trained there, but we do more of that. If we do that, we invade it from the start, it means that we will secure insurability from our point of view. If you secure insurability, you unlock bankability, and that's how you scale data centers.

Lindsey Hall: 31:23

That's a great point. And last but not least, Cecilia.

Cecilia Tam: 31:27

It's hard to go last, because... Adam took mine, but I will come up with something different. I think picking up on what Emmanuel mentioned around systems planning. We often think of systems planning in siloed sectors, and I think what we need to start doing is more effectively planning as a broad system overall, not thinking only just about energy system planning or system planning for the IT or AI industry, but looking at how these interconnected parts of the economy can be more effectively planned, how we can optimize for costs and investment and for siting. Because if we can foresee a lot of these bottlenecks, we can address these problems before they occur.

Lindsey Hall: 32:16

Fantastic. Okay, so some really good forward-looking solutions that all of our panelists have offered. Thank you so much for this conversation. Thank you to our audience. And we hope you've enjoyed this very first episode of Zurich Talk.

[Title: 32:33

Meet prepared] Meet the future prepared] Meet innovation prepared] Meet climate risk prepared] Meet unexpected prepared] Meet disruption prepared] Zurich Talks - Risk to Resilience, May 4 - Zurich, Explore how climate risk is shaping resilience decision-making globally. ]

[Title: Zurich Logo. Disclaimer: 32:45

This video has been prepared by Zurich Insurance Group Ltd and the opinions expressed therein are those of Zurich Insurance Group Ltd as of the date of writing and are subject to change without notice. This video has been produced solely for informational purposes. All information contained in this video have been compiled and obtained from sources believed to be reliable and credible but no representation or warranty, express or implied, is made by Zurich Insurance Group Ltd or any of its subsidiaries (the ‘Group’) as to their accuracy or completeness. This video is not intended to be legal, underwriting, financial, investment or any other type of professional advice. The Group disclaims any and all liability whatsoever resulting from the use of or reliance upon this publication. Certain statements in this publication are forward-looking statements, including, but not limited to, statements that are predictions or indicate future events, trends, plans, developments or objectives. Undue reliance should not be placed on such statements because, by their nature, they are subject to known and unknown risks and uncertainties and can be affected by numerous unforeseeable factors. The subject matter of this video is also not tied to any specific insurance product nor will it ensure coverage under any insurance policy.