AI Has An Electricity Problem | Ep270: Arshad Mansoor

Michael Liebreich
The example of Texas, I know well, right? It has a 77 gigawatt grid with 410 gigawatts in the connection queue and people behaving like the 410 gigawatts is the actual demand, not a 10th of it, not 40 or 20 or 10, which I think is more likely, but 410 gigawatts of demand. You've got to have some view as to whether that's real demand or not.

Arshad Mansoor
Since I went to school in Texas, let me give you from the horse's mouth. In the US one thing that has been done right, we have cleaned up that queue a whole lot more than the UK is just starting. So the queue has been cleaned up and the easy way to clean up the queue, if you're requesting a 500 megawatt connection, put $5 million down payment. Just that one thing cleans up the queue. We haven't done that in the UK.

ML
Hello, I'm Michael Liebreich and this is Cleaning Up. The world's power system is seeing an unprecedented growth in demand from all sorts of sources. Of course, the ones we talk about often here on the show, which would be the electrification of transport and heating and industrial heat, but also from air conditioning and the infamous data centres. My guest today is an expert on all of the implications of that demand growth. He's got a team of 1,500 overall worldwide, of which 1,000 are engineers and technologists. And he provides research to the whole industry in the US, but also around the world. Please welcome Arshad Mansoor, president and CEO of the Electric Power Research Institute, that's EPRI, to Cleaning Up. Arshad, thank you so much for coming in and spending a bit of time with us here today in London.

AM
It is my pleasure. Thank you for inviting me.

ML
And tell me, why are you in London? Normally you're based in the US.

AM
It's a great city and it was climate week in London last week.

ML
Indeed it was. It was a busy time. And of course, it was also an enormous heat wave, which you have now experienced.

AM
I have experience and I also have experienced you guys need more air conditioning.

ML
I'm very lucky or prescient because I actually installed air conditioning in my house a couple of years ago. And my goodness, were we glad to have it. The kids couldn't go to school effectively.

AM
That's a big difference that I found out that in the US, if you live in the southeast or 80% of the US, you have air conditioning.

ML
Now let's start with a little description of EPRI, the Electric Power Research Institute, which you've led now for six years, something like that. Just for your sort of background, our audience, they're all incredibly smart. They're all engaged in climate and energy and so on but they come from different walks of life and they won't all, tragic though you may find it, have heard of EPRI.

AM
If you're in the power systems world, you've heard of EPRI, but if you are not. So what we do is we shape the future of energy and we do that through engineering technology across more than 500 power companies and now NVIDIA and Microsoft and Google, because we bring the truth about power to the people in power with our research and we're in 45 countries, and we engage predominantly with power companies, but with regulators, with markets, with independent system operators, and that's what our engineers do. We make technologies real for power companies to adopt and do that in the right way.

ML
And in my little intro, I said that you have about 1,500 people overall worldwide and about 1,000 of them are engineers and technologists. Are those technologists doing desk research or the engineers and technologists, are they doing desk research or have you got labs? What exactly does work look like?

AM
Huge labs. So if you go to Lenox, Massachusetts, you'll see a 40 acre lab where we have transmission lines, distribution lines, Charlotte, North Carolina lab, Knoxville lab. But you know what our biggest lab is? Our biggest lab is all the power plants and transmission and distribution lines that you see when you go out, because it's those people working in power companies, they're engineers, they actually work with us hand in hand. So we are solving real life problems that they face. Not a lot of desk research at all.

ML
Right. So when you talk about your labs in the various locations, but then you switch to say the biggest lab is real life.

AM
Real life. The grid lab.

ML
Right, and what about around the world? Because you've got activities, you're building an ecosystem here in the UK, but where do you have physical presence? Where do you have people?

AM
So 22 countries, I would say Korea, Taiwan in the Gulf region, a number of countries in the US, in Latin America, in Mexico, in Canada. So in 22 countries, we have EPRI staff. Our largest technical presence outside of the US is actually in Dublin, so EPRI Dublin is a European entity and Dublin is a great place to hire people, the UC Dublin, the college over there. So it's interesting to see that in the world of grid, you can be anywhere in the world, you can be in any country, but the physics is the same.

ML
Now, you are originally from Bangladesh. Do you have a presence in Asia? Because there's Bangladesh, enormous growth in demand for electricity, right across South and Southeast Asia, India, of course, as well, and China. Are you present in any of those countries?

AM
I am always present in Bangladesh, that's where I grew up. EPRI is working in India, working in Singapore, and we would want to go to not just Bangladesh, but Sub-Saharan Africa. But the vehicle to do that would be through World Bank and ADB and other organisations that are working in those countries.

ML
Okay. And so I think there's two topics. You've got this enormous amount of output, you research everything, right? Let's take that as read. But there's two areas that I want to dive in on, and those would be AI, artificial intelligence, and nuclear, because it feels to me there's a combination, a unique combination here, which is two of the most important questions facing the energy system, the world by the way, but also the energy system, are around how much demand from AI is real and how will it be met? And then the second question is, what role is nuclear going to play, and is it going to be affordable, and does it spill over from currently a lot of talk about nuclear for AI, but does it spill over into being an actual climate solution? So AI and nuclear, Arshad, which order do you want to take them in?

AM
I want to start with AI, because over the last three years, I would say 90% of my time has been focused on understanding AI and understanding the power needed for AI.

ML
In fact, you and I, our paths have crossed a number of times, but they really crossed in the Eurelectric Power Summit. Eurelectric is one of the Leadership Circle that supports this, I'm going to say podcast, but also YouTube channel. And we were both talking about AI there. And I think we had a very different approach to it though, just to be clear and let the audience in on a secret, which is I was saying, be careful, because we've seen these kind of moral panic about data centre and IT power demand, we saw it when the Internet came along, we saw it around crypto, we've seen it a number of times and so be careful. And you stood up and said, no, no, no, you have to understand we're creating an artificial brain and nobody can understand the sheer volume of electricity demand that is going to come from that. So we were quite polar opposites in our...

AM
So what would you say now?

ML
I wonder whether I should ask you your view first, because I am not much different from where I was then, which was essentially that in the long term, I suspect we're probably a lot closer aligned, but that's no fun to talk about alignment. I think in the next, let's call it certainly in the next five years, the amount of AI data centre capacity that's actually going to be built is a fraction of the grid connection queues. And I think it's a very small fraction. So I'd be interested to hear, you have the numbers on what is the grid connection queue so let's start there and then see if we can come to a landing point on what we think will be built.

AM
So I think almost three years ago, I was exposed to this world of and I don't call it AI because it has been there for 30 years. You can call it generative AI, but everybody knows it now, ChatGPT, Cloaude and whatever those technologies are. And I was exposed to the we were at a data centre developers meeting with hyperscalers and others and that's the first time I heard about a gigawatt scale data centre. And that's the time we dove into it at EPRI. Is it real? How many gigawatts do you need? So in the global north, that's what's happening and not in the global south. Global south needs electricity for production, food, industrialisation, air conditioning and EV. In the global north in the US, I would say it's kind of two years ahead of maybe the other global north. It's real, it's unprecedented. It's 3x to 4x demand growth than the US has seen in the last 30, 40 years. Right now, we anticipated around 100 gigawatt of new demand. And our report that came out maybe six months ago is already obsolete.

ML
So when you say 100 gigawatts of new demand, let's put that in context because our audience again, they might not know what a gigawatt sort of looks like, feels like, and how it compares. So let's start with the US, what is peak demand on the US grid? It's around 450 gigawatts?

AM
So one fourth. I'll give you another good example of the UK, the UK's average demand is 40 gigawatts. So think of two and a half UK will be added to the US in the next six years.

ML
Two and a half UK's.

AM
Two and a half UK in the next five to six years.

ML
Five to six years, okay. So you are putting a flag in the ground that there'll be 100 gigawatts of new data centres in five to six years. In other words, what's that, that's 2032.

AM
And two, because we have other audiences, we don't call it data centres anymore. Data centres is what processes your email and everything. These are compute resources crammed with GPUs that everybody knows NVIDIA now. So it's a little bit different than a data centre, it's basically a compute resource that is getting into a gigawatt scale. One facility could be 1000 megawatt, even steel mills were, even our steel mills were not that big. So that's where we are right now. And it's happening in the UK now, in Spain, in Singapore, in Korea, in Australia, in India. So it's going there, so there's quite a few lessons learned in the US that we can bring to do it the right way.

ML
Okay. So let's, let's keep giving some triangulation points, but by the way, my number that I published at the end of 2024 was 30 gigawatts in the US by 2030.

AM
Change it.

ML
Well, I'm not sure that I am going to change. In fact, I want to do the opposite. Maybe we should split the difference. Are you prepared to bet, are you a betting man?

AM
In the next five to six years, let's give it, because we have shortage on capacity, labour - six years, 100 gigawatt installed capacity.

ML
Okay. And so I don't want to force you to, well I don't know, what should we bet?

AM
A good Indian restaurant in London.

ML
That is excellent.

AM
You'll take me.

ML
So if there isn't a hundred gigawatts of additional AI compute by 2032, you'll either be here or fly here and you'll take me to the best Indian restaurant in London.

AM
A hundred percent.

ML
And what do I have to do if you're right?

AM
Same thing.

ML
I have to fly over to, let's shake on that because that's, that's a deal that neither of us can lose right because it'll be a pleasure.

AM
It's a win-win.

ML
It's a win-win. But my number was 30 gigawatts by 2030. The reason, and in fact, there was Jigar Shah recently, who's a good friend of both of ours, so my figure of 30 gigawatts was from 2024, back in 2024 but this year he put a thread out on Twitter and he's followed up with various publications saying that the current, well his figure was 34 gigawatts added by 2030. And the reason he said was there's all these constraints, one after the other constraints, which start with the constraints on the grid and then there's a constraint on transformers and electrical equipment generators and so on. There's a constraint on GPUs, there's a constraint on high bandwidth memory, there's a constraint on skills, just how many plumbers and electricians. And he pointed out that the total gigawatts of AI compute that were added last year, do you know the figure? I'm sure you know the figure.

AM
A low number.

ML
Two and a half gigawatts.

AM
Yeah, because we just started. ChatGPT came in 2022, December, November.

ML
But what Jigar did was he said, okay supposing you get three gigawatts in 2026 and five in 2027 and seven. And so he plotted it out with very rapid growth in all of the supply chains, in all of the supply chains and grid connections and everything. And he came to 34 gigawatts by 2030.

AM
I'll add two more years. But go to 2032, I'll go to 2032.

ML
But if you're adding 10 gigawatts of compute in 2030, and you need to get another 66 gigawatts by 2032, you have to go from 10 to an average of over 30 for 2031 and 2032. So it's just an improbable step change function in delivery, isn't it?

AM
There are improbable things we're seeing in the US now, in terms of installed capacity. We're seeing changes in regulations, we're seeing changes in markets, we're seeing changes on who pays for it in a way that I haven't seen in 30 years that I've been in the US, so I wouldn't call anything impossible in this new world. When there is a need, it will happen. And maybe Jigar could join our bet.

ML
Let me tell you, I love that sort of talk. I mean, I'd love to, anything involving Jigar is always fun, he's been on the show twice as well. But I love that sort of talk. Do you know why? It makes me nostalgic. It makes me nostalgic for the days of the dot-com bubble, when people said, you don't understand, there's only one way which is up, and everything is changed by the Internet and therefore, any valuation is acceptable, and any use of a balance sheet. These hyperscalers are committing enormous proportions of their balance sheet to building data centres, an enormous proportion, frankly, on GPUs that have got a five-year life. And so for me, the chance of there not being a correction between now and 2032, the end of our bet, is as close to zero as I can imagine.

AM
So you have to talk about corrections. Is it a stock market correction? Absolutely, it's going to happen. But if you look at the Internet and the impact that Internet has had, right now, it's ubiquitous. We all don't even think about the Internet. I would consider AI, the generative AI, as orders of more impactful than the Internet is in our life today. So it's a very different technology that's coming in. The Internet was a means of communication. This is, I call it humans, you know, we always say knowledge is power.

ML
Humans in a box.

AM
No, knowledge is power. Now power is creating knowledge.

ML
I agree with that. And as I say, I think if we go long, you know, if we were to have a conversation about what the world might look like in 2040 or 2050, I don't think we'd be disagreeing much. The thing is that if you go back to the railways, go back even to the canals in the UK and then to the railway and then to the electrical system and then to the telephone system and the Internet, there's always been an overinvestment, an overexcitement, and there's always been the correction. And in the end, it leaves infrastructure. So AI is the infrastructure of knowledge I agree. But I'd say the chance of it being some kind of smooth projection, because you talk about a stock market correction, what happens when it's a bond market correction that feels much more threatening?

AM
So it's OK to question when will we start seeing revenue from AI to justify the trillions of dollars that we're investing and what's the timing? Perfect question. And would it in the five, six years, the revenue materialise? Great question. What I'm saying is you're seeing a technology unlike the canal, unlike railroad, unlike cars, unlike electricity, they were working on a 40 to 60 year clock, the infrastructure build out. We're working on a 10 to 15 year clock. I would say a 10 year clock because the advancement in AI compared to any other technology is orders of magnitude more. And the application of AI is actually orders of magnitude more than canal, railroad, other things. So it's not just a different technology, the deployment of the technology is at a very different speed than anything that society has seen.

ML
Let me just say, I kind of hope you're right and I'm wrong, because the hyperscalers, it depends how you define them, the five, six, seven, eight companies, they are betting their entire balance sheet on this. Because what's interesting is when you look at whatever the number is, two trillion, three trillion, the numbers are just so imaginary that they're betting on these data centres. And it's the data centre, it's the building plus the relatively short lived assets, the chips that sit in it and all of the communications software and hardware and so on that sits in it, these hyperscalers at the moment, their market capitalisation aggregate is something like $15 or $20 trillion, depends on the day. And they're betting $2 trillion, which feels like an acceptable bet but if you look at their balance sheet, not their market capitalisation, but their balance sheet, they are betting the whole balance sheet on this stuff. And so if it goes wrong, it wipes out really the entire tentpole that's holding up the US economy. This is not a trivial thing if there's a correction.

AM
So I don't think AI investment is trivial. I don't absolutely think that what we are going through is a change that society has never seen. I would bet, and it's a safer bet, that I'll take all the advanced, you mentioned railroad, you mentioned cars, you mentioned canal, printing, I'll take the printing press and I'll combine all four of them. And I would say generative AI will have a much more profound impact than all four together.

ML
I think we're getting theoretical. Maybe a conversation for another...

AM
We're not getting theoretical in the investment that we are making compared to the investment that we have made in the past, even inflation adjusted.

ML
But let me, the reason I think it's theoretical, because without the Gutenberg Bible, we wouldn't have had the printing press. So we wouldn't have had, you know, the Reformation so neither of us would be here. You know, I don't know where I would be, but we wouldn't be because it would be an alternate universe. But it's very difficult to describe the impacts of a very complex system like society, it will have a huge impact. I'm not arguing that. The question is only whether the current, and I think this is where if we bring it back to the power system, the real question or the question that I think we should explore is, you're planning an investment or you're planning the system or you're a minister, we have a lot of, a bunch of ministers will be listening to this, I can assure you. And you've got to decide to what extent, what is the demand signal on the grid from these data centres? In other words, the UK has a connection queue of 50 gigawatts in a system that's only got a 40 gigawatt average demand. And now the question is, do you assume that that 50 gigawatts is going to happen or is it not going to happen? Is it going to be five gigawatts that happens? Is it going to be two gigawatts that happens, is it going to be 25 gigawatts? This is an absolutely existential question for those who are investing and planning our grid, right?

AM
Well, I said that's not the right question. I think that's an interesting question, whether it's five gigawatt, 10 gigawatt, I think it's around five gigawatt. I don't think that's the right question though. It's gigawatts, it's going to be more than anything that the UK has seen. What the right question is, fundamentally, we have built a grid in the north where 50% of average utilisation, it's a hotel that is a 50% utilisation factor. In spring break or your bank holiday, the hotel is booked 100%. And the other time I have 50% rooms in a hotel. That's what a grid is. I have a lot of capacity that I can use if I bring flexibility to the data centres. That is the fundamental change we will have to do to power, whether it's 5, 10, 15, 20 gigawatt, which will increase the utilisation of existing generation, existing capacity. Because you got a 50% occupied hotel and you're building new hotels. We have to build new hotels, don't get me wrong, but get the utilisation of existing hotel rooms higher.

ML
I just want to, I do want to, I want to come to this question of flexibility. I absolutely want to go there. But I want to finish off on the demand function, right? Because if you look at the UK, five gigawatts is a 12% growth, something like that. And if you say, even if you just say, well, it's by 2032, so six years 12% growth, it's 2% per year. Then at some level you can say, listen, we used to grow the grid, right? We used to grow the grid in the 1950s, the 1960s, the 1970s, even the 1980s. We grew at 3%, 4%, 5%, 7% around the world, US, Europe, everybody grew the grid, right? Of course, Bangladesh, India, China, they're still growing the grid. Well, China is probably stopping now, but it's not an abnormal rate of growth historically. If it's 2% growth per year, you sort of feel everybody should just stop squealing and just get on with it and build it.

Whereas if it is 50 gigawatts in a 40 gigawatt system, I've just come back from Australia, again they have some 30, 40 gigawatts of connection requests or projects being discussed and so on. If it's doubling the grid in size, I mean the example of Texas, I know well, right? It has a 77 gigawatt grid with 410 gigawatts in the connection queue and people behaving like the 410 gigawatts is the actual demand, not a 10th of it, not 40 or 20 or 10, which I think is more likely, but 410 gigawatts of demand. You've got to have some view as to whether that's real demand or not.

AM
Since I went to school in Texas, let me give you from the horse's mouth. In the US, one thing that has been done right, we have cleaned up that queue a whole lot more, the UK is just starting. The queue has been cleaned up and the easy way to clean up the queue, if you're requesting a 500 megawatt connection, put $5 million down payment. Just that one thing cleans up the queue. We haven't done that in UK because you have to ask the, is there a rule, if you're not asking other people for a $5 million interconnection, But if you look at what's happening in Ohio, what's happening in almost every part of the US now, the queue is cleaning up the total amount of connection that we're seeing across the US, I would say 60, 80, 100 gigawatt in 2032 is a very realistic number. But cleaning up the queue is step one.

ML
Can I ask the metric on that? How much do people have to put down per megawatt or per gigawatt of connection?

AM
Depends. So the US is a collection of 50 states.

ML
But what's the average? What do you see?

AM
I would say for a 500 megawatt data centre, a $2 to $5 million connection request. If a 500 megawatt data centre, how much do you think it costs?

ML
$30 billion.

AM
So if you're going to spend $30 billion, do you think $5 million is a lot to ask for?

ML
No, but that's where I'm going. Because if you are going to sell or IPO your portfolio of developments and somebody comes along and says, you're about to sell 20%, 40% of your pipeline and make $8 billion or whatever the numbers are. There's lots of examples, right? Do you think that people are not going to put the 5 million down?

AM
The real players will put it, the tier one players, they will put it.

ML
Can I tell you when we've seen this before in my space, solar silicon in 2000, you probably remember the silicon bottleneck. And there were these solar companies, a lot of them European, but around the world, there was also this company called Solyndra, you'll remember very well in the US. But the Solyndra was a solution to the silicon bottleneck because it didn't use silicon. But the silicon bottleneck, there were all these companies that wanted to IPO. And the analysts, the Wall Street and the London Stock Exchange, all the analysts who knew very little, vanishingly little about what they were talking about, they all asked, have you got silicon to deliver your business plan? And so these companies went and they bought, they entered contracts to buy silicon for $400 and $500 per kilo, a commodity that used to cost $20. And it cost, and of course, immediately after the bubble, it cost $20 again. But the point is people, rational people, every single one of them rational, did really irrational things collectively. So I just, I really challenge, because you've said 100 gigawatts by 2032, I think a more realistic number is probably 50, but that's what we're going to have dinner about. But that is minuscule compared to even now, a cleaned up connection queue, isn't it?

AM
I think we're getting over-fixated on numbers. So I'll tell you where to skate, where the puck is going. And nobody's talking about where the puck is going. So yes, we'll be building large data centres. You know where the puck is going, where we have to skate to? You will have a compute resource in your house. You will have a compute resource in the districts. Distributed data centres in vacant office spaces, in warehouses, in your house, is coming in the next 12 months. And now distribution, which, you know, that's the last line of power company that comes to your house, they will see power need for AI coming to the distribution grid. There are companies now coming up with, there's a company called Span. It's an emerging company in the US, they make circuit breaker panels. In your house, you have circuit breakers. So it's a smart circuit breaker panel. Now they have added, partnering with NVIDIA, a GPU resource. So it's not a data centre, it's compute resource. Compute resource will be in your house, in a distribution transformer, on the distribution system. That's where we are going. In addition to the large macro one gigawatt data centres.

ML
I agree with that. I think it's incredibly exciting. I've just come back from Australia where they are already the world champions, they are putting solar on roofs. Which they do by the way, at half the price of a solar, the same roof in the US which I know you've done research on that, but we're not going to go there today. But they've also now added in the last year, 500,000 batteries in homes, which in a country which is only a tenth of the population of the US, imagine suddenly having 5 million homes with batteries. And now you could start putting compute into those homes. It starts to be really interesting, but I just come back to, you say well we shouldn't get distracted by the numbers, which we're both engineers it's a kind of funny statement. But it matters because if we're talking about planning a grid and having to, for instance build pylons, then you need to know whether the additional demand by 2032 from these big data centres, the really centralised ones, whether that's going to be another 50 gigawatts, 100 gigawatts, 250 gigawatts, or whatever the connection queue is now in the US, which must be I'm sure it's still 500 gigawatts. And it does matter, does it not?

AM
It doesn't to me. So this is where I fundamentally disagree because I can go to GE, Siemens, Ansaldo, Hitachi, I can go to the five big gas turbine manufacturers, and every quarter they're doing their earnings report. I can look at their backlog, I can feel how much gas turbines is coming. I can do the same on solar, I can do the same on transformers. Hitachi is building a new plant in the US. So there's a much better way to get to the numbers than stated queue requirement or connection request that is coming to NESO or the system operator in the UK or the system operator in US. I don't get confused with, is it 50, 100, 200, 500,000? I know it's going to be 3-4X more than what we have done in the last 20 years. I know 50 years ago we used to do that, but there's a thing called muscle memory. So our muscle memory in the global north has been declining to flat growth. So we got to exercise the same muscles we did in 1965, where we were growing three, four, 5% in the UK and US and Germany and every place. So that's where I focus on. Are we going to do the same thing that we did in 1965 and 1970, where we were growing 3% to 4% or are we going to do it differently? And I think we will, and I hope we will. And if politicians are listening to this, let's do it differently than what we did in 1970.

ML
So if there's one message I think we agree on is.

AM
Don't get confused with the queue.

ML
If you think the queue is the demand function, you are wrong.

AM
I would 100% say that. And this is actually a good thing to know for all the investors and everybody else. Queue is important, but focus on that this is going to be 3x, 4x growth than what we have seen in the last 20, 30 years.

ML
Cleaning Up is proud to be supported by its Leadership Circle. The members are Actis, Alcazar Energy, Arup, Copenhagen Infrastructure Partners, Cygnum Capital, Davidson Kempner, EcoPragma Capital, EDP, Eurelectric, the Gilardini Foundation, KKR, Mitsubishi Heavy Industries, National Grid, Octopus Energy, Quadrature Climate Foundation, Schneider Electric, SDCL and Wärtsilä. For more information on the Leadership Circle, please visit cleaningup.live. To keep up with all that's going on in the Cleaning Up universe, make sure you subscribe to our newsletter. Written and edited by my longtime New Energy Finance and BloombergNEF colleague, Angus McCrone, it comes out every second Monday. Angus provides the latest on the episodes we're recording, the events we're hosting, stories we're watching and what Bryony Worthington and I are up to. To sign up for the Cleaning Up newsletter, visit cleaningup.live.

So now let's talk about that flexibility that you were saying. Now, you were talking about the data centre turning up at the hotel in the off-season and saying, can I have a bed, can I have some power? And of course, they'll be welcomed. The problem is, though, that data centres want to operate 24-7 flat when they've got a full capacity, when they've got a job to do, when they're actually doing a training run or if they're asked an inference question, they're going to want to work flat out. So doesn't everybody want the high-quality bit of power? Nobody wants the peaks, nobody wants the surplus wind 10% of the time when it's very windy or at midday when there's too much solar on the system. Nobody's interested in that. What everybody wants, the thing that people value, is the dispatchable and in a sense, the bottom of the stack.

AM
So I would say the data centre always wants 100% 24-7 power. That was then, this is now. And this now means power is becoming the bottleneck for growth of AI. So necessity is a model of invention. We are seeing across the world now, there are a number of ways you can do that. The easiest way to do that is every data centre has backup generators. If you've got a 500-megawatt data centre, you've got 500-megawatt of backup generators. So we are working with companies like Caterpillar, Wärtsilä, because you can make those backup generators a little bit more emissions-controlled, and you can put renewable fuel, HVO, hydrogenated vegetable oil, I call it chicken fat from McDonald's, and you can put 80 gallons of tank. So for 80 gallons of tank or 80 hours worth of tank, for 80 hours, I can flex.

ML
I visited, I did an episode, we'll put a link in the show notes, to the big data centre being built in Sines in Portugal, Start Campus. I did an episode with the CEO Rob Dunn, and that's what they're doing. HVO, big tanks, they're on backup.

AM
It's going to happen everywhere you need, there's a new phrase called speed to power, which means it's not just I'm growing 2%, I really need that compute research to be done in two years. Speed to power is not going to happen, you're 100% right, in the last 15 years, data centres, just give me 24-7 reliable power, don't ask me to do anything. That is completely changed. For your audience, I'd say, go and Google data centre flexibility, DC flex, and the word mosaic, M-O-S-A-I-C. Because when you Google flex mosaic, you will see one of the biggest challenges, we don't speak the same language between data centre people and utility people. When I ask you to flex, how many times will I ask you to flex? Ten times a year? How many hours do you have to get out of the grid? 4 hours, 6 hours, 50 hours? So there are different types of flex. We created a classification, class A, B, C, D, E, which is how many times you have to get out of the hotel. We have now more than 60 organisations, including Google, including system operators, including utilities that have embraced, let's standardise the language of flexibility before you talk about how technically you will do it. I feel that is the biggest change that we will see in the next 12 to 24 months, that we will bring flex to power so that 100 gigawatt doesn't need 100 gigawatt of infrastructure investment. Maybe it needs 50 gigawatt when you're right and I'm right too.

ML
Can I push back on that?

AM
Sure, absolutely.

ML
I'm enjoying pushing back, as you can tell, it's one of those days. And the reason is, we did an episode with Varun Sivaram, whom I'm sure you know very well.

AM
I know him very well.

ML
Exactly. And also Steve Smith from National Grid, another of our Leadership Circle. And Varun has got Emerald AI which is about flexing data centres at the level of the compute, so within the actual algorithm.

AM
So they engaged with us about two years ago. We brought them and we engaged with them in a project in Arizona, in a project with National Grid. So we are intimately familiar with Varun. That is one way of doing flex, which is I can flex my compute. There's another way to do flex, which is if I want to train our inference, I'll go to another data centre. There's a third way to do flex, which is I'm going to run hydrogenated vegetable oil on a backup generator. There's a fourth way to do flex, I'm going to create a microgrid that is connected. In Arizona, we are building a data centre with Caterpillar Wärtsilä units. They're configured as a microgrid that can isolate from the grid and just power the data centre. There are at least six, seven, eight ways to do flex, we have never done flex.

ML
Let me finish my pushback, which is of those six or seven ways of doing it, the ones that I challenge are the ones that actually delay the compute. The reason is that, I mean, if you've got, I don't know, you didn't mention, but I'm sure one of the six or seven is also batteries, because they're getting-

AM
Oh, I forgot, yeah, batteries, definitely.

ML
And with the sodium chemistries, they get so cheap and so on, fine. But the thing that I challenge is when you start to slow down the compute, because fundamentally, for a gigawatt data centre, you're spending your $60 billion and then if you're slowing it down, you're basically treating it like a battery. You're basically treating it as a grid resource. And there may be bits within the algorithms, within the compute, there may be things that are not urgent, but there's always another task waiting to be run on your $60 billion asset that I think you want to get started doing. I don't think you want to be slowing down the compute.

AM
I agree and disagree. So I think I agree on critical functions, you don't want to slow down compute. What I disagree is, if I can get a data centre done in three years, but I have to slow down a bit, otherwise I will need five years, I have to optimise whether I'm going to take my asset and slow down a bit five times a year, or I'm going to wait five years to get the data centre. I may choose slowing down a bit five times a year and get that done in three years. Because speed to connection, a UK grid, so we are now working with the UK grid on a project to see how much load you can connect with the different amount of flexibility that the load brings. So if you can connect now two gigawatt with no inflexible 24/7, you may be able to at the same time connect six gigawatt if it's bringing just six hours of flexibility in a year. And that speed to connection is valued by hyperscalers as much as how much they can run their asset 24/7.

ML
Let's talk about public acceptance, because that's the elephant in the room. We're two engineers, we're talking about this, and I've talked to Varun and I talked to Steve Smith from National Grid, and we have these very technocratic and very informed conversations, it's great, or my visit down talking to Rob Dunn in Sines. But the public can vomit on the best technocratic plans because they've not been consulted, they don't like one aspect of it, like air quality for two or three years before a grid connection is in place, those sorts of things. So where is the US on public acceptance? It feels to me like the trend is going in the wrong direction for the hyperscalers, is that fair?

AM
If you, this is a long I'm trying to short it. If you think this was the one biggest challenge in the next five years, it's not a bubble of stock market crashing, it's not, we don't have enough gas turbines or transformers, it is the societal public acceptance, that's going to be the number one bottleneck. So engineers can also learn from history and every transformative technology, I'll give you an example of electricity. And I'm in the UK now, in 1880 to 1910, the Ottoman Empire banned electricity. During the same 30 years, the British Empire created the UK Electricity Board and started electrifying Great Britain. Germany, during the same time, created a company that we know as Siemens. The Ottoman Empire was never able to catch up because of the 30-year delay. Societies that have stopped transformative technology because of opposition that is from reasonable people, but they didn't deal with the opposition the right way, have not prospered.

We will see railroad in the UK, Manchester and Liverpool, that's where it started, 60 years of opposition. Medical articles written in Lancet, that's a British journal, by reputable doctors that travelling 25 miles an hour or higher causes brain damage. And this, just read history as engineers, we can read history. You look at automobiles in the US, what we called cars? Devil's wagon, that's what rural people called cars. There were decades of opposition on railroad, electricity, cars. The opposition that you're seeing today is in its infancy. But don't take that opposition lightly, it's not coming from unreasonable people. We have to do it the right way.

ML
But there's a few things mixed in there, right? So the opposition needs to be taken very, very seriously. And I agree with that but some of those examples are absurd examples where people worried about brain damage from going too fast. Or they were worried about the impacts of electricity, presumably.

AM
I wouldn't call it absurd, doctors were not absurd at that time. It was evidence at that time.

ML
But at that time. But in retrospect, wrong. Let's just call it wrong so it's not as disrespectful. But here we're talking about air quality, water quality, electricity prices, job destruction. These are real things that in retrospect, you may come back and say, oh well, it was worth it. But you can't say that these are not real concerns, for those local communities, they're damaging concerns.

AM
But you know what the real one, so you mentioned several things, air quality, electricity price, land use, and job destruction. I wouldn't weigh them equally. I would say the fear of AI is the number one thing that is driving everything. In the US we have graduation ceremonies, college graduates. Just look at the press, in the recent college graduation, you have a commencement speaker who talks about, you know, to the new graduates. There were more boos heard when the word AI was mentioned. By college graduates who are using Perplexity and ChatGPT all the time, so I think fear of AI is starting to drive a lot of the community engagement subjection that you see.

ML
So let me address that because I put that under job losses, job loss, fear of job loss. And so I think that on that one, I think I have to be careful not to get lots of hate mail in the comments, but if young people think that we can put this technology back in the box and not use it because it will preserve their job as a junior bookkeeper or as a junior lawyer, that's not good.

AM
Then you will become the Ottoman Empire.

ML
You'll become, exactly. I think we agree on that. But there's also, I was thinking, so that's the job loss piece but what about things like local inflation, right? Taking all the electricians, all the plumbers, all the construction workers, all the transformers, all of the generators, the cable and so on, and using it for AI and driving an enormous bulge in inflation into every business locally, which then kills jobs in a very different way. So that's kind of what I was thinking about job losses. So there are these, and air quality is a real proper concern. And underpinning it is also a concern about governance. The fact that the people who benefit from AI are, how can I put it, they're not living in Washington, but they're extremely politically connected and all guardrails seem to have been ignored. There's some real concerns.

AM
Can I disagree on that?

ML
Do you think there are guardrails?

AM
I would say the biggest impact on AI, on education and healthcare, will come to the people in countries like Bangladesh. So think about education, of all the things we have done in education, a country in Bangladesh, Nigeria, I would say that's the biggest impact.

ML
I'm still in Virginia at this point. I'm still in Virginia where the electricity prices are soaring, the quality of electricity, as you know as an engineer, is deteriorating. The capacity market payments have up 800% and so on. I'm still, so if we stick with where those data centres in the US are gonna be built, I agree that there's enormous positives from AI around the world, right? But at this point, it's a public acceptance question. Because I'll tell you, the people in Virginia don't care whether somebody in Bangladesh is getting better medical service as a result.

AM
So if you're going to focus on Virginia, I would say we need to move out of Virginia because there is a lot of data centres that are already coming in Virginia. Bad news travels fast, good news doesn't travel. So I'll talk about good news. The school teacher in Louisiana is getting a bonus structure that is equivalent to the school teachers in New York because the revenue that is being generated by the data centre that's going in Louisiana is paying for the school teachers. I'll give you another example, and this is happening because data centres or AI compute resources are going to Kansas, going to Georgia, going to, so Virginia is the wrong example because it already has a lot of data centres like Dublin. I think this is what doing it right means.

Doing it right means if you're going to be investing $30 billion in a local community, make sure that the property tax rates are going down for everybody in the local community. Make sure the schools are being built in the local community. Make sure the teachers in the local community are getting the same salary that a teacher in New York is getting. That's starting to happen. We have 13 case studies now that we'll be publishing, case studies of where local communities got benefit. I want that to be 300 case studies in the next 12 months of every local community that is benefiting. And you know the biggest benefit? By far the biggest benefit, a rising tide will float every boat, which means in a declining demand growth when Germany and UK and US were investing hundreds of billion dollars on generation assets, your demand is not growing, but you're investing billions of dollars on generation and grid. Who's paying for it?

ML
So I've written about this. I wrote a piece called Make Electricity Cheap Again, saying we need demand because it grows the denominator.

AM
A rising tide will float every boat.

ML
Now, Anthropic in February this year, made a pledge that they will not drive up local power prices. The Trump administration then published something called the Ratepayer Protection Pledge in March, which was a voluntary agreement by the hyperscalers on the same lines. But what would you say to the Secretary of Energy in the UK, despite the fact we're on track to have a new prime minister as we record this, he has offered discounted power. So in Scotland, in order to attract data centres, he hasn't said to the hyperscalers as you pointed out, all they care about is speed. So instead of saying to them, we will get you first compute, time to power, we will drive this through, do whatever it takes, and you'll get first compute in 12 months or 15 months, but you'll pay for it. What he said to them is, we will remain just as bureaucratic and slow as ever, but we'll give you cheap power, which will put costs onto everybody, all other power users in the UK.

AM
So I would say, read the paper called Win Win Watts. For your audience, Google Win Win Watts, it's a paper that we published six months ago that says what doing right means. And doing right means, make sure that the community that is housing the data centre, all the ratepayers, they're not actually even paying more, they're paying less. Those hyperscalers are ready to do that. So as part of EPRI, we are not a lobbyist organisation, but we speak the truth about power to people in power. I would say, do it the right way. And right way means these hyperscalers, they want this to be done in the right way for the community. They will pay for transmission, they will pay for generation, they want speed to connection. And they want to make sure that the local community and the ratepayers...

Now to be honest, if you had zero data centres, you still would need more transformers because electricity demand has grown at a 3x speed than other forms of energy globally, air conditioning, EV. And transformer prices are all, before data centres came, the inflation after COVID, so rates will go up because your cables, your transformers, everything has gone up. The milk has gone up in the last five years, so your rates will go up. The only way you can check that rate going up is to bring in another 10 gigawatt in the UK. And the only way you want to do that is turn off the data, would they want discounted power? That's not what they want, they want connection in two years. You haven't even cleaned up a queue yet so I would clean up the queue, I would engage the data centre community, Digital Realty is here, big data centre development. I'll tell them we'll bring flexibility, you bring flexibility, we'll connect to you in three years, and we will reduce the rate for the consumers which otherwise would go up because inflation is happening. That's my two minute message to your politicians.

ML
Brilliant, and it's exactly what I would say if I had the same opportunity. Make electricity cheap again by doing that and essentially, I hate to say it, but send the bill to the hyperscalers because they can pay it.

AM
And bring speed to power.

ML
Now, I want to move on because I said that we also talk about nuclear. And there is a segue which is that as these hyperscaler AI, I don't know if I'm allowed to call them bros, but I'm going to call them bros.

AM
You can call crypto bros.

ML
Well, but there was the hyperscaler bros who decided, and they went through the kind of five stages of grief when they discovered that the electricity system would not allow them to execute their grand plans. They went through these stages of grief, sort of anger and denial and negotiation and depression and eventually they decided that when they were in the negotiation phase, they decided that nuclear was the answer. They didn't have to accept any of the constraints you and I have been talking about because nuclear, particularly small modular reactors, were going to be the get out of jail free card for them. So two questions, one is nuclear a get out of jail free card for hyperscalers and data centres? And two, does that spill over to make nuclear a meaningful climate solution or not?

AM
So just a quick thing, nuclear is the largest research portfolio. Every single nuclear power plant operator in the world that we work with engages with EPRI. So we are deeply passionate about nuclear.

ML
Your research?

AM
So for our research. So we have, of the thousand engineers and technologists, over half of them are on nuclear side so we are deeply engaged. So answer your two questions. One, there's no get out of jail free card for any one technology, you need many technologies. But the reality of advanced nuclear is in the next six years, that the bet that we are making, that's not the technology that enters the equation. So six years, that's not the technology.

ML
So 2032, 100 gigawatts, none of it, or what proportion of it?

AM
500 megawatt, one gigawatt. Because we have looked at how technology progresses all the way from hydro to solar, you need to build at least 10 commercially deployable, we call it advanced nuclear technologies, or SMR, small modular reactor. Once you build 10, which we could build in the next seven, eight, nine years, then you could build 100. The scale, so is it a get out of jail free card for the next six, seven years advanced nuclear? No. Is it an opportunity to take the trillions of dollars of investments and put some of it to accelerate to get the first 10 built? Yes. If I have already 100 nuclear plants in the US right now, is it an opportunity to take one of the plants that has been shuttered down because the demand was growing and bring it back? We're doing that right now, that's two gigawatt. Can I take existing plants in the UK and France and US and say, hey, if you can produce 800 megawatt, can I do something to produce one gigawatt? That's 200 megawatt more, can I do that in the next six years? Yes. So I feel nuclear in the next six to 10 years, and there are many forms of advanced nuclear, small modular reactors, that's the time we figure out of the 50 designs that we have, which three or four designs really shines, and we build four, five, six of them. That's only 500 megawatt. After 2035, if these are done right, built right, nuclear does become a significant opportunity 2035 and beyond for both cleaner energy, powering demand, and we don't know what's the next demand, that's humanoid. We're not even talking about humanoid here. That's the next phase of AI. And I'm excited, but what I'm not excited about is when I hear a timeline that is not grounded in reality.

ML
But let me push on your timeline because I think your first point, which was, there'll be some of it powering some data centres. You said 500 megawatts, I would even go beyond that, to be honest, because some of these designs are 200, 300 megawatts. So the Westinghouse and so on, you only need to build a couple of those, you get to 600 megawatts. And so let's call it low single gigawatts

AM
Possible.

ML
Possible, okay. But building 10 of different designs will not drive you down the cost curve. We know this from Boeing and Airbus, you have to build hundreds, they need hundreds of orders for airframes before they can get a cheap price for any of the airlines. So I think there's still a difference between 10 and 100. And I ran the numbers on how many of these small modular reactors you would need to catch up with where wind and solar are today. Not because I'm trying to set up a competition between the two, but just as a metric, as a triangulation. And if you take those 300 megawatt small modular reactors, so the Westinghouse size, you would need just to catch up with where wind and solar are globally today, you would need to build 2000 of them.

So I don't think we're talking 2035. I think we're talking 2045/2050, even to provide the 20% of current electricity demand that is currently met by wind and solar. I'm just finding it hard to see how these SMRs are going to be a meaningful contributor to climate change this side of, you can say 2040 or 2050. A few percent of electricity, 10% of electricity, fine. But we both know how insignificant that is in terms of overall energy demand and emissions and the speed with which we need to be acting.

AM
I actually don't disagree on any one of those points. I think you need to build 10 of the same type before you can build 100. And I would hope that, you know, when you talk Westinghouse, it's also GE. Right now we are building like one or two of them. I hope in the next couple of years, we will see announcements to build another six, seven, similar types, couple of different design. So my timeframe is by 2035 around that time, we would have figured out to build eight or ten of them in a couple of different designs, 2035 and beyond.

ML
Eight or ten each or eight or 10 total?

AM
Eight or ten each. In order to get a design to a point where I can estimate the real cost, you know, whether we need ten or eight. Everything, PV, wind, nuclear power plant, large ones, we had to build 10 before we could build 100. Cost curves, we could give assurance on cost curves after building 10. Now, after you build 100, don't forget to build it for another 50 years then again, you have to get to the cost curve as we are seeing. So 2045/2050, I am more optimistic about advanced nuclear playing a significant role in the energy ecosystem. Kinda your timeframe at scale in 2040, 2045. At scale means at the wind and solar scale, think how long it took wind and solar to come to the scale that it's today.

ML
No, and I do see it being at wind and solar scale. So today, wind and solar scale in the 2040s is sort of-

AM
2045

ML
2045 is the outside edge of optimistic. But I can tell you the comments are gonna fill up with people saying, you don't understand, this is $60 per megawatt hour electricity, it's cheap, it's baseload they'll call it, even though you and I know what you need is flexible rather than baseload.

AM
Some SMRs absolutely can be flexible.

ML
Of course they can, but then their economics become even worse because their output, as soon as you follow demand and you start ramping them down, then the cost per megawatt hour produced goes up. So what, in all of your research, those incredibly smart technologists, engineers, no doubt you've got lots of economic modellers, technical economic modellers, what would be the price of, let's call it the 10th of a kind and then I don't know where you want to go, hundredth of a kind or thousandth of a kind? Can you just stick the tail on the donkey?

AM
I would not because let's build 10 to get a good price projection. The price projection that has been done today, depending on the type of technology, you can get it competitive to any generation that will give you 24/7. So I would say you could make it competitive to any generation that will give you 24/7 and flexible, I would say it will be even more competitive and this is where the comments will start. See, here's the beauty of EPRI, there are people who would say nuclear is too cheap to metre, heard that? There are people who would say, sun is free, wind is free, we like to right stick in the middle. And we're saying if you do it right, you could make it even more competitive than variable generation. You could make it more competitive, nuclear that is dispatchable, done in the 50th kind that will provide what I need, which is I need 24/7. I mean, I understand that industrial load can flex but society needs, you know, we need electricity.

ML
Hang on, but let me push that. Let me come in there because that one I hear a lot and what society needs and what everybody needs is 24/7 reliable power. Well, what they need is reliable power throughout the day and night. They don't need a flat level. It still has peaks and troughs and they need that reliable power from a system. So many people, I mean, I'm talking thousands, tens of thousands of people over the last 25 years have told me that the only way to provide reliable power is a power station that works 24/7. First of all, none of them do. And second of all, that's not what you need. There are aluminium smelters that are right next to hydro power stations okay, fine but mostly it's the system that provides the 24 seven power. Not any, not any individual power station, not coal, not gas, not nuclear, none of them. So I'm just finding it hard in a system that's already 20% globally, 20% wind and solar, that's clearly gonna go to 40% wind and solar, which means that for large parts of the day and the week and the month and the year, there is zero marginal cost electricity. What is the role of something that essentially has to run 24/7 or it has horrible economics?

AM
And so I would say zero marginal cost of electricity, I know that's a phrase that has been used many, I would say it's zero marginal cost when you're not putting a cost or premium on insurance of electrons that I need when that wind and solar doesn't exist.

ML
But it's the zero marginal cost. That's what it is, right?

AM
It's zero marginal cost when the wind and solar is blowing. But if you look at a society that will get to wind and solar, you're not gonna do it with a four hour battery, you're not gonna do it with 10 hour battery, you're not gonna do it with 500 hour battery. What are you gonna use? And who's gonna pay for that?

ML
But the problem is, I agree that that's the big challenge. I mean, I work on this all the time as you do. But the train has left the station, right? We've already got 20% wind and solar globally and in many countries and in Texas and in Idaho higher. What are you gonna do?

AM
It should go even more where I have wind, where I have solar, but we have a fundamental issue that is being addressed now, which is we never, when it was five, seven, 10%, 15%, 20%, you can start not thinking about that. It goes to 25%, 30%, 40%, then the real question comes up, who pays for the insurance that I will need something for eight days once in four years?

ML
So you have that as the question, which I also have, right?

AM
I call it the debt valley, the value of debt.

ML
But I'm gonna come back at you with another question, which is in what way does nuclear solve that problem? Because that's the thing I can't see.

AM
I think what nuclear solves the problem is once I have nuclear at the cost that I need, which I can only get once I have built 10 or 15, then it becomes another portfolio in addition to wind, solar, and if you're looking at resurgence of wind and solar, I go back to your Mitsubishi Heavy Industries. Ask any one of the gas turbine manufacturers how many gas turbines they're gonna be building in the next 10 years and ask them how many new gas turbine lines they're building. So I see a resurgence of gas turbines higher than wind and solar in the next five years. So I'm looking at a different set of numbers. So I'm looking at a portfolio of gas turbines coming, wind and solar, Texas, Arizona, Pakistan, I mean, look at the number of solar in Pakistan growing, and so we always have been in a region, different things will work.

ML
And I see you in terms of the demand for turbines, partly obviously at the moment, AI, which by the way, I think they're gonna, after my correction they will all become available for other uses. And I see a lot of demand for turbines and engines for that last six days, eight days, four days, whatever. But I see that as being the solution and if necessary, working on biogas or HVO or methanol, or if we're rich enough to make it clean, we'll make it clean. I just can't see that the solution to that is to postulate an entirely different electricity system with very deep penetration of nuclear.

AM
I would say you have to give the time for nuclear because of the timing it takes. But if you project your energy system over the next 30, 40, 50 years, and when you look at that and you have existing assets that have to be replaced, you will have demand growth that's happening. The key thing on nuclear right now is, the number one thing on advanced nuclear, make sure we can show to the world that we can build it and then we can build it on time and we can build it on budget. We won't be able to do that unless we do at least seven, eight, nine of the same kind.

ML
Seven, eight, nine of the same kind. So I suspect that the numbers, as I say from the experience with airframes, I suspect the number is 20, 30, 40 of the same. And by the way, we probably want to place five, six, eight, 10 bets. So now whether we do that or not, I think I have a feeling that what I ought to do at this point, because I'm keeping an eye on the clock, what I ought to do is to say, rather than wait until 2032 for our other bet, why don't we-

AM
Take another bet?

ML
No, bring you back in a year or two and we'll revisit this conversation and we'll push it out because I have found this absolutely fascinating.

AM
So I would say, so let's talk about reality. The reality in North America, let's just take that, there are maybe two confirmed commercial builds of SMR that's happening. And that's not going to cut it. That's not going to cut it.

ML
Last week, one actually last week was announced in Idaho.

AM
Well, announcement is one thing, putting steel on the ground is another thing.

ML
I think they started construction.

AM
So I would think right now it's OPG, Ontario Power Generation is the two GE BWRX units, TVA, Tennessee Valley Authority also looking into it. It's too few, it's taking too much time. So in two years time, if I come back and if we don't have at least 10, not just announced, but projects that are going on, then we're missing the boat to make advanced nuclear. So we have means, the industry, the nuclear industry, the policymakers, everybody has a role, which is talk is cheap. Let's get confirmed orders to build the first plants and let's do 10 of them. And do that in the next two years, let's have 10 confirmed orders. Are we going to have it?

ML
Arshad, let me invite you back. You come back in 2028 and we will review whether there are 10 in the US and I think there should be 10 elsewhere in the world.

AM
Let's do 20 confirmed orders where steels are being put on the ground for advanced nuclear. If we can do that, I am very optimistic about nuclear.

ML
Okay, do we include China in that? Because China could easily be doing 10 or 15 or 20 of their own.

AM
I think you have to include China in any global discussion. And how many in China? Because right now we don't have good numbers.

ML
Let's look for, in two years time, we're going to look for 10 in the US, 10 in China and 10 in the rest of the world. And if we see that-

AM
We will be very optimistic and bullish. That's a great number because talk is cheap. Interest is, the nuclear interest, I have not seen the nuclear interest like I'm seeing globally right now. Interest is good, I need confirmed orders with steels on the ground. And if you can have 30 of them, 10 in the US, 10 rest of the world and 10 in China, I think we'll have a great discussion on nuclear in 2028.

ML
Very good. That is a great place to leave this discussion because we can frame that as a challenge to the various ministers and their teams that I know listen to Cleaning Up. So I'd like to thank you for your time here today.

AM
Thank you. Thank you, Michael, thank you for inviting me.

ML
So that was Arshad Mansoor, President and CEO of EPRI, the Electric Power Research Institute, based in the US, but operating around the world. As always, we'll put links in the show notes to the very many resources that Arshad and I mentioned during our conversation, too many to list here. I'd like to thank Oscar Boyd, our producer, Jamie Oliver, our video editor, Kendall Smith, our head of operations, the team behind the scenes at Cleaning Up, our Leadership Circle, without whom none of this would be possible, and you the audience for joining us here today. Make sure that you subscribe to our newsletter. You can find the sign up at cleaningup.live, and that way you won't miss any episodes of either the normal programming like this or our deep dives. And with that, please join us at the same time next week for another episode of Cleaning Up.

Cleaning Up is proud to be supported by its Leadership Circle. The members are Actis, Alcazar Energy, Arup, Copenhagen Infrastructure Partners, Cygnum Capital, Davidson Kempner, EcoPragma Capital, EDP, Eurelectric, the Gilardini Foundation, KKR, Mitsubishi Heavy Industries, National Grid, Octopus Energy, Quadrature Climate Foundation, Schneider Electric, SDCL and Wärtsilä. For more information on the Leadership Circle, please visit cleaningup.live. If you're enjoying this episode, please hit like, leave a comment, and also recommend it to friends, family, colleagues, and absolutely everyone. To browse our archive of around 250 past episodes and to subscribe to our free newsletter, visit cleaningup.live.