⁠Can Exponential Growth Save a Finite Planet? Ep187: Azeem Azhar

Transcript

Michael Leibreich

Can you continue with Exponential View and exponential growth on a planet that is really stressed environmentally, and from a climate perspective?

Azeem Azhar

I think the climate crisis will get addressed in large part by these technologies. There's a bigger question about this idea of infinite growth on a finite planet. I like to play the thought experiment of saying, 'Well, look, if you were in 1750 and you were presented with that question, you probably wouldn't believe we could do what we did by 1950, let alone by what we're doing by 2024.'

Hello, I'm Michael Liebreich, and this is Cleaning Up. In 1961, had you wanted to build one gigaflop of compute capacity, that would have cost you the equivalent, in today's money, of $190 billion. In 2023 that figure was one and a quarter cents. Solar power in 1975 would have cost $130 per watt to buy a solar panel. This year, the number is 31 cents. And batteries are headed the same way: In 1991, to get one kilowatt hour of lithium ion, batteries would have cost $7,500. Today, you can buy that for less than $50. My guest today has spent the last decade of his career examining how these sorts of trends work and their implications for the economy and for society. Azeem Azhar runs something called Exponential View. It's a podcast, it's a substack, it's thought leadership, and the man himself is here with me today. I must say, of all the pieces of content that land in my inbox every day, the only one that I make sure that I open and read is Exponential View. This episode will also be going out on Exponential View so it reaches a broader audience. Please welcome Azeem Azhar to Cleaning Up.

Azeem, thank you so much for joining us here on Cleaning Up.

I'm so excited to be here, because I listen to this podcast on my flights, when I'm driving around, and to actually be on the show, Michael, is a real honor.

So this is the mutual fan society, because I likewise listen to yours. Let's do the following, because this is a bit of a different episode of Cleaning Up in that you're also going to put it out on Exponential View. So I want to start with the way I normally start, and ask you who you are — the short version — and what you do? And then, who knows, maybe you'll want to reciprocate.

I will try my best. So I'm Azeem Azhar. Right now, I'm the founder of Exponential View, which is a newsletter, podcast and Substack that I set up about eight years ago that looks at exponential technology change, artificial intelligence, computing, renewables. And it does it from the perspective of having been a founder and an investor. And I spend about half my time investing in early stage, very risky but very exciting companies, and I try to bring it all together for my audience. And the reason I know you, of course, Michael, is that I was lucky to be an investor in New Energy Finance 15 years ago. So for my audience, though, please tell them who you are in a sentence.

So I'm Michael Liebreich. I was, when we first met, the founder of something called New Energy Finance. I'd like to think we were amongst the, maybe not the riskiest, but the most exciting startups. And what we were doing was unpicking what was going on in — what was at the time called — alternative energy. Of course, now 95% of all investment is going into the clean stuff. And so we met then. Since then, I built that business, sold it to Bloomberg, and I now do, like yourself, perhaps investments, but I'm also building a number of businesses. One is this media business: Cleaning Up. One is an advisory business called EcoPragma Capital, which is part of our leadership circle here. And the third is actually an HGV charging business called Pragma Charge, because I try to put my my money where my mouth is,

And the word 'pragma' shows up, because it's your heritage as an engineer, of course, which is, engineers are very pragmatic, very practical. They get things done.

Well, that's right, and there's a nuance. It's not just engineering, it is also, for instance, the advisory business is EcoPragma. Because if you say eco, it's like, everybody gets that. That's climate change. But you can be a real Ultra, and you can be very sort of almost fundamentalist about it, whereas 'pragma' is saying: I'm sorry, somebody has to be able to make it. Somebody has to buy it. The only way to scale things, and I did this with New Energy Finance. I scaled it enormously because it made money, because it was an actual product that people wanted to buy, not because some billionaire gave me some money and said, 'Michael, send out the best information you can.' That would have failed, because I would have been focused on the billionaire and not the market. So that's 'pragma.'

Well, you know, I think that is one of the keys to why we're seeing this vast expansion of investment in solar and batteries. It's because the premium has disappeared, or is disappearing, or people can see a path to it disappearing. And so you have technologies that are fundamentally, in many, many ways, better than the ones they're replacing, but there's no price premium you have to pay. And that reality, when the economics makes sense under the umbrella of the law of physics, you start to see markets expand rapidly.

That's exactly right, and the analogy that I use for that is mobile telephony. Now you go back 20-25 years, you had to pay a premium for mobility. So you had a landline that was cheap, but if you wanted to be mobile, you paid more. And there just came a point when it was like, no, you pay less for mobile. So now it's got all the attributes of the landline, plus the mobility, and it costs less. And this is what's happening with some of the solutions we're going to be talking about, solar, of course, first amongst them. But let's talk about Exponential View and actually, I want to back up. I want you to talk about not just what you do, but how did you get there? Because you've gone all in on, in a sense, the exponentiality in a number of sectors, which we'll go through. But what's your background? and what was your epiphany?

Well, my background is actually being born as a child of the microprocessor. So I was born a few years after you, in the early 70s, and the first computers showed up when I was seven or eight years old.

You had to get that in, that you are younger than me, right?

Well, I mean, age is a state of mind, so we can see at the end of this show who's really younger. But, you know, I grew up at the point at which computers were making their way into our homes. And so I had lived with the idea of Moore's Law since I was a teenager, reading about it in computer magazines and having watched the internet really follow an exponential curve in those years, from 1980s through to about 2005 when you looked at the number of people coming online. I started to really understand that you could get these exponential phases that last quite a long time in technology. So it's really been the drum beat in my career. And when I sold my last company a decade ago this year, I needed something to do, and I thought a good way of disciplining myself would be to write every week. And that writing turned into Exponential View. What I discovered after about a year was that it wasn't just silicon chips that have this characteristic of exponentiality, which is that they get cheaper by 20, 30, 50% per annum every year. There were other technologies. There were technologies like solar panels, technologies like lithium ion batteries, that were getting cheaper and cheaper and cheaper. Genome sequencing as well. And once I looked at that, I tried to make sense of it. And of course, I rediscovered, or I discovered myself for the first time, Wright's Law about learning curves, in 1936, and I saw in the data that several technologies were at a point where they were about to break through in the economic sense that you described. So I looked at those trends across a number of different technologies, and I tried to ask, what happens if they all cross this point where they are good enough and affordable enough? And so that became this idea of the 'exponential age', and focusing on exponentiality, and constantly facing the criticism that, of course, an exponential doesn't last forever. And my view is it's going to last for long enough to change the world.

Well, that resonates, because I wrote a piece which you may have seen earlier this year. First of all, I did the five villains, the five reasons why we'll never get to net zero. The five super villains, you could call them. But then I did the five superheroes, of which one is exponential growth. Homage to you, of course.

Thank you.

But I also made the point that, of course, it's not really exponential, but it is the point of penetration curves, where things look exponential, and anybody who has bet against them being exponential has lost. That's the important point, that it doesn't matter whether they are truly exponentials, does it?

No, but it's also about over what time frame do you look at this? So I've done a piece of work to look at the total amount of computing capability on the planet. And of course, I started in '72 because that's when I was born. It's easy to count the number of computers. And if you total up the number of computers times the amount of compute they have, which is measured in operations per second, or floating point operations per second, and you run it through to 2023, you see a 62% compounded annual growth rate that's over 51 years. So yes, I'm sure at some point that exponential may stop. But 51 years of 61% growth is a really, really long time. And it's enough to transform economies and transform the ways where we live, and I think we're seeing similar patterns when we look at something that is completely the reverse of a silicon chip. Let's look at a lithium ion battery, and look at the cost per kilowatt hour. $1,200 in 2011, you say, $50 in 2024. So that's a huge reduction. I'm just going to do my math, so a 24 fold reduction in 12 years. I mean, that's remarkable. And what we know from the battery manufacturers is that they think there's much further to go, right? I mean, how far? A dollar I'm not going to say that, but certainly much further than 50.

Well, I don't know. The analogy there would be perhaps solar where, you know, it went from being this thing that you could only ever do on a satellite to a thing you could only ever do on a pocket calculator, to a thing you could only ever do on, let's say, a parking meter, to a thing that you could do on a roof, which was ridiculous. And the statistics from the time that I started New Energy Finance — since you invested in this risky and exciting business at that time — one gigawatt of solar took a year to install. One gigawatt, one year. Six years later, it took a month. Six years later, it took a week, and six or seven years later, it took a day. And where we are today, you've now got the Chinese solar industry, which can produce enough solar to produce alm two terawatts of power a year. It'll take another year or two to get to two terawatts. It's over a terawatt now, but when it gets to two terawatts, what that means is that 10% of all the electricity that we use today could be replaced with one year's output of Chinese solar manufacturing, right? It's just a different world.

And they're bringing more factories on stream as well, right? That's not the full capacity of the Chinese solar industry.

It is... I mean, look, I actually think some of those will go bankrupt because we're not going to install two terawatts, because there are some rate limiters in terms of just how many people there are, how many electricians, how many people can connect things to the grids, potential trade tensions and so on. So that industry has never worked at 100% capacity utilization, nor will it. But it's very interesting. We're at the point where, frankly, in solar, either we'll be installing maybe not two terawatts, but certainly a terawatt, or a terawatt and a half, or there'll be huge bankruptcies of Chinese solar companies. And of course, you know that's not the end of the world, others will pick up the assets and keep going, but we're going to see some pretty non-linear things.

So let's go back to a couple of things you said. One, particularly for your Cleaning Up audience, they may not be as familiar with exponential processes as my audience is. That analytic where you said, 'How long did it take to put a gigawatt onstream,' and it's gone from many decades to years to months to weeks to days. That becomes common in industries that are going through an exponential transformation. You certainly saw it in terms of how quickly people got onto the internet from the 80s to the 90s and then into the 2000s. You see it with the growth of saas or cloud. And a really, really good way to figure out where you are in the curve is to look at that compression. But there are two other points I'd make. One is that on one of your shows, it was either you or your guest who said, 'the cure for high prices is high prices.' The old commodity traders adage.

That would probably have been around minerals. Maybe there was a discussion, probably about lithium.

That's right. So at the time, I mean, I wrote about lithium, and I said: there's going to be no supply problem. We're going to have a few sticky quarters. It's going to make journalists excited, and then we'll come through it. And when all of the things that you've described about the rate-limiting factors that exist within the market, which is the number of electricians, the number of widgets you need to stick a solar panel on a roof. All of those things create economic incentives for entrepreneurs to respond to. So when you see something like that, my experience would be that complementary businesses will emerge in order to take advantage of that shortage, that blockage. Then the third thing I think that we should understand is that solar behaves and exponential technologies behave very, very differently. Because at some level they are mostly highly modular. And what modularity means is that you can enter that market at many, many, many different sizes of customers. So in 1973, I think the time of The Man With The Golden Gun, the James Bond film about solar power because it was so powerful you're willing to send 007 after it, only a handful of companies in the world had computers. If you or I wanted to buy a computer, we would have needed to front half a million dollars. As soon as computers became modular through the PC, the computer that I would have bought in 1990 would have been the same computer that a big investment bank would have bought. They would have bought 1,000 of them, but it's the same machine, and that's what modularity does. And what's happened with solar PV. In particular it's transformed who can access the energy system and where that addressable market is. 30 years ago, the addressable market for energy was essentially only really big companies who could find billions of dollars and stand up a nuclear reactor or a gas plant that could enter that market. And today, homeowners, businesses, mid-scale businesses, mid-scale investors can get into this market, and it really expands the market much, much faster than ones that have been constrained by an older technology with an older configuration. And we know that's happening. We know all over the world, the energy system is changing because of that.

Now you've done some work on a problem that, or an issue that Jenny Chase raised, a couple of months ago.

Yeah.

Which is, where is all the solar going to in Pakistan, right? And you've tracked down some of it down, and that's a great example of what you're saying. Pakistan has economic growth, but grid-tied electricity is actually dropping, and the delta is people are buying these very cheap solar panels.

It's behind the meter. So, I theorized that a few years ago, from looking at what had happened to the internet, in particular. Once the internet allowed anyone to be a producer of services, people started setting up web servers, and blogs cropped up everywhere, and you didn't just rely on the Times or The Financial Times or The Guardian. And by analogy, what we started to see in Pakistan is a great example. So of course, as you might imagine, the grid is strained. There's not enough generating capacity, there's not enough transmission capacity, so a lot of brown outs and load shedding, and individual businesses started to buy solar panels. And Pakistan became the sixth largest importer of solar panels in the world. And then Jenny Chase and people at Ember Climate, which is a wonderful resource, started to look at satellite photographs and physically see these solar panels. And of course, the other data point was, as you say, the split between GDP growth and reported energy consumption from the Pakistani regulatory authorities. But it's important to note, it's not just Pakistan. And the first time I wrote about this was about four years ago, looking at South Australia, because South Australia reached a point where rooftop solar was generating 50% of the state's energy demand, and it's quite a demanding place, the Australian electrical system. And that suggested to me that there was this opportunity. And of course, South Australia is well positioned, there's high levels of insulation and so on, but it indicated that this decentralized, bottom-up approach could take place. And I would say the other place that we've seen this battle play out has been in South Africa, but not with solar, with wind around the Cape, where individual entrepreneurs and businesses were investing in wind resources and Eskom, the sort of national utility, was making it very, very difficult. And you know, it came to a court case to allow these people to effectively bring their resources into the grid, but outside of the aegis of Eskom's national pricing. The reason I bring those stories up is because, again, for listeners of Cleaning Up, you're seeing these somewhat radical technologies come into a market that is regulated against particular technology assumptions. And where I've been a historian of technology, I've seen that same pattern emerge in the computer industry relative to mainframes and in the communications industry, when you look to the internet relative to the centralized, highly-controlled telephony system. And I just bring it up as a sense of saying, if you ask that question in a particular way, you may find these results — Pakistan, South Australia, South Africa — unsurprising as a result.

Okay, so far, it's all been sort of smooth sailing. We're talking about how to spot exponentiality and what it does and I think a lot of people find it very hard... You know, humans are quite good at linear growth, quite bad at exponential growth. But it's a tradition on Cleaning Up that I push back.

Of course, yeah.

I say, wait a minute, all this sounds much too easy. And so I've got four problems with that perspective. The first is that it doesn't work everywhere and in every technology. So, you know I've spent an inordinate amount of time working on hydrogen, and I would actually put up my hand and say, I got it wrong. I was always very skeptical. When I wrote my first pieces, first big pieces on hydrogen, called Separating Hype from Hydrogen, I did the supply side and the demand side. The demand side, I said it's going to be very hard to see where this hydrogen is going to go, because it's expensive to transport, expensive to store, expensive to distribute and expensive to use, unless you're already using hydrogen, right? So I couldn't see where it was going to go. But on the supply side, I said, 'ah, my friend Azeem says these things are exponential, I agree.' By the way, I was calculating learning curves in my first job in 1985 or 86. So I also had a kind of the privilege of…

Around PVC manufacturing was it?

Funnily enough, I did fibre optic cable. I did PCB.

PCB.

I also did rock quarrying. Rock quarrying also has a learning curve. It's just that we don't double our cumulative knowledge every year and a half, or whatever we do in computer chips, or every every three or four years, like we do in solar. So, you know, I'm a complete convert to experience curves. So I said, 'hydrogen gets cheap.' And it's only been much more recently, the last couple of years that I thought, hang on a second, the electrolyzer stacks for green hydrogen get cheap, they could go to zero. It doesn't matter. They are only — and the data now that we know is that electrolysers are only 11% of the cost of green hydrogen today. The rest of it is either electricity, which has already become pretty darn cheap — clean electricity — or it's a chemicals plant. It's heat exchangers, it's tanks, it's storage stations and pipelines and so on. But even in the green hydrogen plant, even on the production side, it's a bunch of stuff that is not going to go down by five or 10 times in the next decade, decade and a half, whatever. Unless you tell me I'm wrong?

No, I mean, you're absolutely right about that. And that's one reason why there is a disconnect between how far the price of solar PV can fall, and how rapidly the prices of actual installations will fall. Because when you get to the actual installation, and you've got to put down 100,000 panels on ground that's been surveyed, even if it's very flat, it's not very flat, you know, it's not like a factory floor, lots of odd things will will show up, and somebody will drop a, crate load of panels and you'll be delayed by a day, and that'll add to cost. So when you have to look at the full system cost of these things with hydrogen. I mean, I went down a journey where I really thought that the electrolyzers would find themselves on a learning curve. One of the problems we'd had with hydrogen previously had been that each hydrogen facility was treated as a plant — an n of one — and people hadn't approached hydrogen as a problem of modularity. Can we produce products? And I tracked down some startups doing this, and they were quite compelling with it. But what really persuaded me — the two things that persuaded me — one was listening to you and hearing what you were arguing about hydrogen, the physics of hydrogen. But the second thing was that the competitive alternative technologies were getting so much better and so much faster to the point at which, in your famous hydrogen ladder, more and more things that I think... Green is bad for hydrogen, is that right?

Green is the things that could go to bio, and yellow are the things that can go to direct electrification.

Okay, right. So, essentially the space for hydrogen keeps falling. And then one of the things one sees is that effectively, if there is an ecosystem that is being built around electrification, around batteries of different types, around moving chemical processes to electric first, more and more innovation goes there. More inventors go there, more capital goes there, and those solutions get better and better. While all the things on the hydrogen stack, slightly wither and die. So I think of course,hydrogen, had it had a much, much better learning curve, might have had a bigger chance, but right now, it doesn't have enough to nurture a big economy around it. A souffle that has flattened, perhaps?

Well, yes, so that's my hashtag, hydrogen souffle.

I mean I'm more simple, I call it the bubble that burst.

You see, I don't take joy in things failing and people losing their jobs. But you know, at some point when it's predictable. But what's interesting I find in the context of exponentiality is that there are people, you know, I pointed out that, for instance, the density of hydrogen means, if you put it on a ship, that ship will carry 1/3 of the energy of an LNG ship. And then people say, 'Michael, I'm surprised you're so negative when you're the person who was so convinced solar would get cheap.' And I'm like, 'Yes, but physics.' And many people seem to think that exponentiality or the learning curve can actually overcome physics. I hear you that the focus of innovation moves to the constraint. So you then get lots of people. So solar, apparently there's this real problem that the sun doesn't shine at night.

I've heard that.

It was a surprise to me, but there you go. But now there's so much innovation around batteries, because it's the theory of constraints. You move on to the next thing and so on. But I want to keep going. Three more pushbacks. I want to do my pushbacks, right? One is that you talked about, you have to take the whole system cost, and I agree. We were just talking about solar, then the innovation around batteries, but there's also a system which, in many cases, is highly resistant. So what is the role, you know all of this is a bit utopian, because it's easy to do a few percent of wind and solar. But you know, once you get into sort of grownup bits of the energy system, and by the way, it is regulated, and regulators answer to political masters and mistresses or political leadership, and they are paid for, bought and trussed up like like Christmas turkeys by the oil and gas industry. There is a point where none of this stuff can continue to happen, right?

Well, I think that we're seeing this in the UK, and I'll give you a US example as well. But in the in the UK, dear old Greg Jackson from Octopus Energy, guest on your show, but also on the nightly news, regularly making the point that the energy system is now exponential technology, first wind and solar and batteries, and we need to make changes to the framework by which the market operates. We need to have nodal pricing, which means that we price the electricity relative to how and where it's actually produced. We need to look at this question of marginal pricing, where expensive natural gas plants set the price for most consumers. And of course, he hasn't gone into the other areas that we care about, like demand response and all these other things that actually make the electricity market work more like a market. You know, I need the electricity more. I need it more consistently. I'll pay more for it. You're willing to consume when you need. But I think the other place to look is what's happening in the US, and who knows what will really happen... But the Texas electricity market, and Texas as a state, and ERCOT more broadly, has really transformed, and it's transformed towards predominantly solar, but also a lot of wind. And Greg Abbott, who's the governor there, has been messaging the point that Texas is the leading electricity energy market in the US for this reason. And I think that what you start to see is a moment, and we saw it with the internet, compared to telecoms regulation, where the momentum of the market starts to overwhelm the objections that the politicians have. When will that happen? Will it be within four years? Will it be within two? Two it's quite hard to say, but ultimately you have the laws of physics, then you have laws of economics, and then the third line is just the law of reality. And if the electricity market is not functioning, governed as a fossil fuel market, because it's now 35 to 40% renewables, I don't believe lobbies will be able to stand in the way.

So you're confident that, in a sense, Greg Abbott or Texas... I wrote something in 2014 saying that clean energy needs less regulation, not more, and we need more price signals. More price signals work, we know this. This is why we get cheap renewables. It was basically, it was the experience curve and innovation. But then it was when we changed from feed in tariffs, state pricing, to reverse auctions of various sorts, including the CFD with the auction, and then suddenly these cheap prices got delivered. So I agree with you, that's what should happen. So Greg Jackson came on Cleaning Up at the beginning of this season, the first episode of this season, our most popular episode yet.

Second most, hopefully by the end of this episode.

Maybe second most after this one. But I think the third or fourth most popular is Yanis Varoufakis, the firebrand lefty from Greece, yes. And he came on the show and argued that the electricity system should be nationalized. There is no innovation that a national electricity system couldn't deliver better and faster, that this is a fake competition, fake innovation, fake businesses, nationalize the lot. I thought it was the most ridiculous argument possible. But let me tell you, looking at the comments on all the social media, the people were saying 'Yanis is a God,' the person who interviewed him was an asshole.

Well, I mean, that can be true and he can still be wrong. I think you're wonderful, I want to say just because you've sort of potentially referred to yourself as that. I mean, I think you understand this market better than nearly anyone. The reason you don't want to have a nationalized energy system is because this is a technology. Electricity, which was essentially driven as a commodity. Gas prices, oil prices, determined it. They were determined by whatever happened in the market and whatever the local autocrat wanted to do. We've turned it into something that's driven by knowledge, by research and development. That is what the learning curve is. And once you have something that is driven by learning curves, it is so dynamic that if you try to control it, you will ultimately control it in the fixed state of last year. In the same way that if... you know Bill Gates is reported to have said, 'oh, 640 kilobytes should be enough memory for any computer.' We now have a million times that in some computers. If the state had then come in and said, 640kb will be enough, because Bill Gates has said that and regulated that, the industry would never have taken off. And the reason it was inappropriate is because it was a learning curve industry, an exponential industry, and the energy system is becoming that. So what we need to do is make sure we have the right incentives. Price is one fantastic signal for it. But I think there's a second one that we, I think, are also well aware of, which is that we're moving from technologies or energy systems that are high upfront cost and low running costs, they become a financing problem. And I don't think that's solved exclusively by a price signal. I think that is solved by active policies that will foster the kind of financing that make it practical for businesses and home homeowners to make that transition.

Cleaning Up is brought to you by members of our new Leadership Circle: Actis, Alcazar Energy, EcoPragma Capital, EDP of Portugal, Eurelectric, Gilardini Foundation, KKR, National Grid, Octopus Energy, Quadrature Climate Foundation, SDCL and Wärtsilä. For more information on the Leadership Circle and to find out how to become a member, please visit cleaningup.live, that’s cleaningup.live If you’re enjoying Cleaning Up, please make sure you subscribe on Youtube or your favourite podcast platform, and leave us a review, that really helps other people to find us. Please recommend Cleaning Up to your friends and colleagues and sign up for our free newsletter at cleaninguppod.substack.com. That’s cleaninguppod.substack.com.

Objection three, you said, 'Oh, it's so marvellous, Pakistan is the example. Suddenly, everybody can participate in the energy system.' This is kind of small is beautiful, how marvelous. But, you know, that's actually not how it works. I mean, those may be the consumers might be, but just because you have a little mobile phone and you can now microblog or whatever on your phone, let's be absolutely clear, we've now got these companies that are worth $3 trillion each, that actually dominate that space. So what's really happening is it's just as much concentration. But instead of having Standard Oil or Exxon Mobil and Aramco, we're just going to have, I don't know, Longi and and Goldwin and a whole bunch of colossal Chinese companies dominating everything. We are not going to be any freer at the end of this than we were before.

Yeah, well, we will be freer because the Chinese companies making solar are much more like the people who made RAM chips in the computer industry. And the RAM chip... who makes them, who knows? Because it's a super commodity. They compete on price. As you say, they're going to be swayed by boom and bust cycles. Many will go out of business. We shouldn't shed a tear for their shareholders, they've got broad portfolios. But the monopolies came from the silicon chips, and they came from the fact that the chips were programmable, had operating systems over the top, and solar PV is much more like DRAM than it is like chips. So although the Chinese companies may be the biggest producer, they'll be the biggest producer of a pile-it-high, sell-it-cheap product, and they'll have to continually compete on that level of price. I thought where you might take that objection is to the fact that a few thousand kilowatts here or there just doesn't make aluminum, right? It's just not enough. But it does have, in aggregate, the problem of breaking the way the grid works. And in Pakistan, this is a real issue. Capacity payments that are required may put some of the producers out of business, and in a country like Pakistan, there's not a lot of cash to go around and to bridge that. So I think one of the hardest things as the energy system goes through this exponential transition is there are going to be these break points where ultimately someone's business model is not going to work, and we're going to have to figure out who's going to take that pain, and that may be where political resistance slows things down.

Thank you. You've come up with the fifth objection, which is, 'how do you do the really big things?' But also how do you keep the lights on? When everybody's got their very cheap system, which is fine 48 weeks a year, but on the four weeks a year that it's not fine, they turn around and say, 'What's the government doing about the fact that my lights don't go on anymore.' Because they've defected, they've gone off and done the cheap thing. Which is part of the same...

It's part of the same thing. The way that I think about this is the types of new language that you need when you go through a transition like this. Again, by analogy for Cleaning Up listeners. Before the internet, we had these things called publishers. Everyone knew what a publisher was. The internet came along, and now we don't really know what a publisher is. Is Facebook a publisher? Is it not? It's still fought over in court. You don't have the words for it. And so when we think about building an exponential, renewably driven energy system, a phrase that comes out is over-build, or over-capacity, which only makes sense in the context of a fossil-based system. Because if you were designing a system that has intermittent resources that are very, very cheap, so you could put lots of them up that you can interconnect and exchange and you can have storage systems of different capacity and cost, you would never really think about over-building. You'd think about a measure of availability and resilience. And I think that design issue is one that we will need to figure out. It might be that we have to keep a bunch of gas plants either idle as insurance and be willing to underwrite their costs for those weeks where we have cloudy skies and dunkelflaute going on. I think the other thing that's really underrated will be that by the time we get there, 60, 70, 80% of all the cars in the UK being sold will be electric vehicles, which means about 20% of those on the road will be EVs. An EV will have an 80 kilowatt hour battery. It'll power my two neighbors' homes and mine, and not really affect how far I can drive the next day when I do the school drop off. So I think that there are resources that will emerge.

So I have two points to add to that, the discussion about sort of capacity utilization and over capacity, it drives me crazy, because there is not an asset class that we manage for 100% utilization, right? Your car is used 4% of the time. Your bed is used 30% of the time. Your whole home is used 50% of the time. Your school is used 60% of the time. Your hospital, nothing... Your gas peaking plant is used 10% of the time. Your hydro plant is used 40% of the time.

Nuclear, 92%.

Some nuclear. Look at the average across the world, and it isn't. And by the way, when it doesn't get used, when it falls out in an unplanned way, which is rare, but you know, half of France is nuclear. I'm a fan of nuclear, I have to say that, for your listeners who might not know all of my background, I'm a fan. But we shouldn't pretend that it's this kind of 100% reliability thing. Why is it that only when it comes to renewables, do people say, 'Oh, the expectation is we must use it 100%. It doesn't work, it's rubbish, because it doesn't work 100% of the time.' Or whatever it could produce we must use by building — and we're back to hydrogen — we must electrolyze using surplus wind or whatever, without understanding that it is so much cheaper just to build a little bit more wind or definitely a little bit more solar, and not go through this acrobatics of trying to use 100%. The other point that I'd make is that when it comes to the aluminum smelter, or whoever it is, they need to pay for the fact that they need 24/7 dispatchable power. What's actually happening is some of these companies are lobbying and using their influence to say the whole system must provide six nines (99.9999%) — or five nines or whatever it is — of reliability, because we need it, whereas the homeowner can simply say, 'Well, do you know what I can do? A little bit of demand response.' Frankly, if once a year, once every two years, I have to tell the kids not to have long showers, or whatever it is, the homeowner will do that, right? The aluminum smelter or whatever is using their influence to try to force the whole system to meet their needs. And it's a form of subsidy.

It is a form of subsidy, and it's one of the... I mean, I would love for more people to understand that ultimately, in the energy market, the most demanding users pay the least. It's a bit like when you go on a flight, and it turns out the passengers in first class are paying a third of the price of those in economy. But there is one objection to that view, which, by the way, I'm 100% aligned with you, which is that if the aluminum smelter or the hospital or the glass manufacturer did have to pay the real price of this constant electricity, it would get reflected either in business closures or in costs going up for the consumer when they get the product.

There is a whole episode to be done on the macro-inflationary impact, sector by sector, country by country of all of this. I've got to figure out who to talk to, but I definitely want to do that one. It gets framed as deindustrialization, but you can also look at it through different lenses, like inflationary force and so on. Now, those electricity users that need 24/7 dispatchable power. We talked about aluminum smelters. This is the most contrived segue possible. AI data centers, which you've also done, and I've done an enormous amount of work on, so I've just written a very substantial piece for Bloomberg. You have worked on AI for, I mean, tell me how many years, a lot longer than me.

I mean, the first time I worked for an AI company was in 2001. We competed with Google. They won, I would add.

I was going to ask, 'and how did that go?'

Well, it's, you know, you've got to be in it to win it. And we were definitely in it.

So they did Alpha Fold, and you did...?

Well they were doing search. We were doing, back then it was sort of natural language, those types of things. So AI data centers, again, I mean, I think this is a function of the media environment really needing stories. Data centers were growing by the boatload before AI, then this new AI showed up. Companies were digitizing, they were generating more and more data, they then didn't want their data just in one location. They wanted many locations. Then they had to have it locally because of local regulations, so data centers had to grow there. And if you actually look at the incremental build of AI data centers relative to traditional data centers, I mean, I think in one Goldman Sachs report from May of 2024 it was about an additional 20% on data center build.

So far. And let's just back up, because I've done a deep dive into the history of how much electricity is used. You probably are across the numbers of flops, right?

Well I know a little bit about the electricity, but go ahead, yeah.

So what my potted history of electricity demand for data centers is: go through till 2000, when the internet came along, and people just had their data center. It was just called a server, and it sat in the corner of somebody's office, or in the IT department. And then the internet came along. Suddenly, everybody went digital crazy. But also there was a lot more video started to be produced. So a lot more data. The millennium bug didn't trouble anybody in the end, it didn't happen. But suddenly there was this new problem, which was power demand of data centers. But it never really happened, because people moved the data centers into the cloud, and they got so much bigger. And then there was this thing called Koomey's Law, after an academic called, I think, Jonathan Koomey, who said, 'well, everybody thought it was going to grow exponentially, but it was pretty much flat.' And it stayed very flat until about 2020, so 1-2% of global electricity is being used by data centers. It didn't do what the World Economic Forum said. They said crypto meant that data centres...

... would use as much as Taiwan, then Poland, then China.

Well, the World Economic Forum, and I think it was Time Magazine, said that it was going to use all the electricity in the world — crypto would use all the electricity in the world by 2020.

Yeah, that's a danger of taking a curve that's going like that and keeping on drawing it out.

Ah so this exponential stuff doesn't really work, that's what you're saying.

Well, sometimes it doesn't work, we've established that.

But in 2020, the power demand did start to go up. As you say, there were boatloads more data centers everywhere. And we started to see these new, you know, we did see 2022, we saw AlphaFold, we saw, then of course, Chat GPT.

Just to come back to physics, right? That was two years after the growth that you tracked started. And the 2020 growth would have been ordered in 2017 and 2018 by the businesses, which is, you know, five years before anyone knew Chat GPT was going to be a thing. So that's part of the observation that I wanted to make, really, which is that the curve bends well before this new AI, new-fangled technology shows up.

So what we've got then is, without doubt, being turbo charged by generative AI, which is extremely power hungry. And by the way, I'm going to argue the other side as well, which is that by the electrifying heating, electrifying vehicles, that will absorb vastly more power than any of any of this AI stuff. But it will be different because it's distributed in the economy and some of these AI data centers have to be huge and centralized in order to train these models.

And I mean, the AI companies have put in orders for 500 megawatts, 700 megawatt data centers, and they're talking about gigawatt and five gigawatt ones in a few years time, which will be centralized, which will need to draw that kind of power or produce that kind of power because of on site nuclear reactors or solar, is there sort of back of a napkin sketch.

So on site nuclear, that's right, because they have all sort of headed off to say, well, the only solution here is nuclear, and, you know, I'm just sort of smiling. I don't know how to pronounce the word gnomically...

Gnomically.

Like the Mona Lisa — I'm just smiling quietly to myself, because they will discover exactly what everybody else has discovered, which is: you can build new nuclear, but it's very expensive and it takes a long time. And by the way, if you co-locate it when it does fall out for whatever reason, you've got a real problem now. You've got no grid connection, no backup.

Absolutely well, they'll try and think about the backup. I mean, they're smart people. One of the things I would say is that the prize is so large that, you know, Microsoft and Google have both talked about spending more than $100 billion to develop super state of the art models. Not the ones they have today, but the AI models of a couple of years time, or beyond. And that is a big, big prize. And they're in a game theoretic standoff where every boss believes the cost of not getting this right is existential, and these are companies with access to hundreds of billions of dollars. So when a nuclear plant is expensive, and even if it is a Hinkley Point level of cost, it's kind of affordable to a big tech company if they really think the price is there.

Absolutely, and they will do some of that. I just think with my gnomic — now I know how to pronounce it — smile, that they will also find that other things are cheaper.

I suspect they will. Yeah. I mean, if you're thinking grid connections, solar and batteries.

Absolutely. And so there's a data center being built in Portugal, just south of Lisbon, and it's going to be a lot of solar, a lot of wind, and then it's going to be a lot of batteries, and then it's going to be some form of longer duration outage backup, which will probably just be fossil, used for a small amount of time. But you know what it could be? We're back with our third bite at hydrogen. It could be, because if it's used only a little bit, it doesn't matter how expensive it is. And actually, that's one of the things about these companies or data centers, which is fun, which is these bulk buyers or bulk-needers of gigawatts of dispatchable clean electricity are suddenly not industrial companies working on low-margin commodity products. It's suddenly the richest companies in the world. So it's a fantastic source of demand to test geothermal and long duration backup and nuclear SMRs or whatever. It's kind of exciting.

And these companies are the few companies in the world who know how to work with innovation and know how to work with scale. So when you go into the data centers at Meta, which is what Facebook used to be, and Alphabet, Google, they have really done their own hardware design. Of course, they're buying Nvidia chips,but lots of the other technology that lives within there has been designed themselves to their own specification, their own architecture. And so I can't think of companies that are better positioned to be innovative while also deploying large amounts of what is effectively risk capital, than this particular sector. That doesn't mean they'll get there, but I do think it's a really great opportunity for anyone who is building innovations in the energy system to be able to find partners willing to fund this.

So I am a bit more skeptical. They do build. And the complexity of a modern AI data center, because they're not putting in single CPU chips. They're putting in these GPUs. A single rack could be 130 kilowatts — single rack — and you've got to then get rid of all that heat. So you've got liquid cooling to the chip. Then you've got n plus 2, n plus three power redundancies on the power supply. I say they're as complicated as aircraft carriers.

They're remarkable. I would recommend people go and watch the YouTube video inside the xAI data center that they're currently building — the Elon Musk one. And they look at these racks, and they show how complicated it is: they show the cabling, which is beautifully done. And then in the basement, all the manifolds from the cooling aggregate into this large pipe, and in the basement are huge pipes, probably five feet in diameter, that are moving the water around the cooling system. I mean, it's remarkable.

We'll put a link to that video in the show notes. Do they also show that the data center has got, I think it's 14, two and a half megawatt natural gas generators. It's now got a 150 megawatt grid connection that has just been approved, but it was started with natural gas generators. The reason I'm skeptical about their abilities is that, I think that if they step out into the world of developing small modular reactors, then I think they're going to find that this is, I don't want to say it's an order of magnitude, it's just very different. The things that go wrong when you're building a nuclear power station, or any power station, are just very different. Because the data center world has come largely out of real estate. Sure, we build a big shed, and then we put some stuff in it, and then we put in a cooling system.

Well, I mean, the big real estate owners I've spoken to in the US have been talking about data centers for a while. They're not reading this in the Financial Times, the Wall Street Journal, as a big surprise. They've been having the conversations for a little bit, which is why I think we see the numbers trend up so quickly.

They've been talking about it, they've been trying, they've been doing it, but they've been doing it as a real estate development with a 25 megawatt, 50 megawatt connection, right? A gigawatt... You know, how many of the people have they hired out of aluminum smelting or out of the glass industry, or out of the big utilities, the transmission networks and so on.

Look, there are some time horizons that don't add up. So on the one hand, you have the pronouncements by the big tech companies and the hundreds of megawatts or gigawatts they want per data center within three or four years. Then you have the reality of needing to build new technologies, which will take 10 years — not four — even in this accelerated time. Then you have the issue of, can you actually get the grid connection? Can you get the permission to do it? Can you get the land and can that be done in three or four years? So I suspect that that timeline will not be as close as three or four years. It will get pushed out in a couple of years. That's just the reality of it. But I did want to bring in a couple of numbers, because Morgan Stanley, actually, just before we got into this discussion, sent out a note, and there was some interesting data about their forecasts. So they believe that globally, there'd be $140 billion invested in data centers this year. That excludes the chips. The chips are another 140 billion — or the GPUs and the sort of networking chips that go with them. But by 2027 that would rise to $250 and $260 billion, so effectively doubling over a three-year period. And it would result in a $200 billion incremental investment in the power system. Now, the reason I raise that with you is because I think the global renewable investment is about to cross the $2 trillion a year mark. So it is meaningful, but it's also only 10% right? I mean, it's not 50%.

No, exactly we are seeing... so the things that I tracked in 2004, it was basically renewable energy, because there were no electric vehicles, right? It was $36 billion...

Yeah, but there was probably a hydrogen car back in 2004, selling better than they are today.

Yes, but there were still so few of them they didn't actually sort of register. But it was $36 billion. And now, if you add all of the clean technologies, you get to, I think, $1.4 trillion. And you add the grid on top, and you'll get to your figure of $2 trillion. And now what you're saying is it's $200 billion more. I'd say the interesting thing is, a lot more of it will be in the US for all sorts of reasons. Two thirds of that could be in the US simply because you can't export the latest Blackwell NVIDIA GPUs to China, and Europe is mired in its own stagnation, or its own issues.

We have lovely cathedrals.

It would be nice to have a cathedral of AI. So I think I agree with you that it's going to happen, and it's not super scary. It's just really interesting, because it's such concentrated demand with such rich companies. I think they'll get delayed, not by a couple of years. I think that the things that they think they're going to be able to deliver by 2030 are probably 2035.

You mean, from a power standpoint?

Because energy, yeah, well, and that's an interesting question. What is the constraint to rapid rollout of AI services? Is it a lack of energy? Is it lack of... What is the constraint? I think power is going to be a constraining issue.

So when we look at this, there are two different things going on within AI. One is the training of these big frontier models that are enormously energy intensive the first time. That's just hard to get around, right? The second is what happens when they actually get deployed and distributed? And after the first big, expensive model gets trained, we're able to do optimizations for the models that actually get run. And in fact, that is orders of magnitude (less power intensive). And so GPT-4, which is the best model we currently have, when GPT4 class models are actually run for businesses, they're using inference.

They're actually used to answer questions rather than just trained.

They're millions of times less power intensive than when they're being trained. But even when you train a new GPT-4 class model today, you can probably do it for 1/1000 of the energy because of optimizations. So I just want to just finish up that point, which is that we've not broken all of the use cases of GPT-4 yet. So when I think about this sort of hypothetical experiment, imagine you just can't train GPT-5 or 6 for energy reasons, power reasons or anything else, I still think you have this enormous demand to build AI data centers to run GPT-4 level applications for businesses and for consumers.

But those will be the inference data centres, which will look much more like today. They'll be 100-150 megawatts. They won't be the 1 gigawatt, 5 gigawatts. Okay, now let me come back to something very interesting. So if we agree there's going to be some delays and some speed bumps, and there's also a piece written called AI's... first it was a $200 billion question, by David Cahn of Sequoia Capital. It was fantastic, we'll put it in the show notes. And then it became a $500 billion question, the more they invest. And the question is, the nature of the question is, who's going to pay enough for the AI services to justify the investment? And where is it going? What I want to challenge you with is, haven't we seen this behavior where there's an arms race, the companies think it's existential, they have to win. And we've seen it in railways. We've seen it in electric light. We've seen it with the internet. We saw it, and you and I spoke about it with crypto. Do you remember there was Terra and Luna? That's right. And I was laughing at the chat that you were very excited about Mr. Kwon.

Well, I mean, he was very exciting.

He's now, excitingly, a fugitive from the law. But aren't we going to see, not this sort of smooth pathway to 2030 or 2035, of enormous additions to AI infrastructure, power use and model capability and so on? Aren't we going to see a correction at some point along the way?

I mean, we are bound to see a correction. The question is, what type will it be? With the railways, as you brought up, 1846 was the peak or 1845 of these stock capital, market capitalization before the collapse. Some shares didn't get back to their par level for 70 years.

Many never did. Many disappeared.

I mean, essentially, if you look at the UK stock market, it took decades to recover.

The NASDAQ took 20 years, I think, to recover from the dotcom bubble.

And look at where it is now. And if you think about where Amazon was after the dotcom bubble collapsed, it's gone up certainly 100 times, if not more. I think a few of the things that are different between railways and AI are that some of this is infrastructure, right? The absolute frontier models are infrastructure. But the deployed models, the ones that I use many, many times a day, are not infrastructure. And the rate with which companies are actually using them is growing extremely, extremely fast. And there was some data that was published in the Financial Times, which showed that the time it took an AI enterprise company — so that's an AI company that sells to businesses to get to a certain revenue, which I believe was $10 million per year — the run rate was three times faster than the best in class cloud based software companies from five years ago. And that's real companies in the real world buying these services from flashy startups. So I think that because there is money flowing into the system, it feels quite different to crypto, certainly to the dotcom where the mantra was ‘build it and they will come’. It was a Kevin Costner movie at the time, Field of Dreams. And it feels like there's real use. Now that doesn't mean there isn't exuberance in private markets. From the startups I've been investing in, I've seen some hair raising valuations for companies without products and that isn't to say that there aren't public company stocks that aren't overvalued. However, what I would say is that if I look at the future demand for chips, and you thought — you have to remember — only 300 million people are using Chat GPT today. And my perspective is that that type of technology will be embedded in every phone for the 6 billion of us who have phones. That growth rate still looks like it's going to continue for a few more years.

I agree with all of that. I guess if you agree that $600 billion of revenue needs to flow into AI to justify the dream, that turns into, and I did the numbers, either $600 per person for a billion people, or $6,000 per person for people like you and me, who use it a lot. And I just think it takes probably a decade to get there, to identify and isolate the budgets and to build the kind of connective tissue around how you use it and so on. So I do think that the chance of a correction is greater than 50%.

But I would say to add something else, which is, of course, a lot of these companies are expected to fail, because they're going to be bought up in a consolidation, right?

Not if they're Meta or Amazon or Google.

No, not the Meta's but the startups pursuing online translation using GPT-4, right? That is where you'd expect it.

I'm just conscious of time, and I'm also conscious of one thing, which is something that I said. I said that there were these five pushbacks, and we only did four. There is a last one I'd like to close with, which is Herman Daly, economist at the World Bank, environmental economist. He said there's no such thing as infinite growth on a finite planet. And you know, I'm on the record as disagreeing with him, and the reason is that the nature of growth is not just adding value to materials, it's actually knowledge and dematerialized services and so on. But the things that we've been talking about — solar panels, batteries, AI, data centers — those are physical things. So I guess I would challenge you: can you continue with Exponential View, exponential growth, on a planet that is really stressed environmentally and from a climate perspective?

You know, I think the climate crisis will be addressed in large part by these technologies. It'll get addressed by being able to provide electricity in all sorts of places so they can better adapt to the new environmental realities through air conditioning, powered by solar panels, through battery storage, through desalination, all technologies that are going to improve over the coming years. But there's a bigger question about this idea of infinite growth on a finite planet. I mean, infinity is a really big concept. Maybe we don't have to get that far, but I like to play the thought experiment of saying, 'Well, look, if you were in 1750 and you were presented with that question, you probably wouldn't believe we could do what we did by 1950 let alone what we're doing by 2024.' And there's a lot that we don't know. We know about the laws of physics, and there's a lot that we can learn, and we can research. And maybe one key limitation in his claim is the idea of where the resources come from. I mean, I've met companies now who are starting to plan on building data centers in space for many reasons, so that you might choose to go off and do that. You can power them 24 hours a day. You don't have the same heating problems, you have got high bandwidth laser and connections back to Earth. And so I wouldn't doubt the capacities of our science over the hundreds of years and bound them by the limitations of the science of today.

There's a company called Lumen Orbit I've looked at. They want to do data centers in space. And I looked at that. Their numbers for the cost of this don't stand up to the very slightest of scrutiny. The idea of building 40 megawatt-type power stations in space is completely absurd. Now, in hundreds of years, I'm not going to say no, but I'm kind of worried about 20 years and 30 years when we're going to see this climate pinch point. And I worry that these technologies, the first people who will use them, could easily be using them to do more fracking, or to exploit frozen methane in the seas, or to build all sorts of things that are going to be extremely negative.

Or to build even faster commerce than Temu and Shein, disposable instant commerce for millions of people.

Or even a very, very good scenario where they enable wealth for billions of people. Something I want to have happen, but it happens at a speed where those people then, the emerging lower-middle class, they all buy vehicles because with AI manufacturing, the vehicle became so cheap, and robotics became so cheap that they can all drive cars.

And yes, but you know, we can imagine that dystopian system, because what it does is it requires thinking along one axis alone. And we get to set some of these values, and we get to articulate them. And I think, you know when I think of, my little trifecta of people who've influenced me around how I think about climate change, and they're three, very different people. It's Elon Musk, it's Greta Thunberg, and it's Michael Liebreich, my little tripod.

It's a pretty good crowd, that would be a good dinner party.

And, well, of course, you've got Jim Hansen as well, and so on and so forth. I mean there are lots of names, but I think when we look at what starts to matter and what didn't matter 20 years ago. Look at lithium-ion battery recycling. Look at the extent to which there is confidence that we will get to 25, 30, 50% of all the lithium batteries recycled reasonably soon.

We're already at 90%, there just aren't any lithium batteries. The recycling we've built is 25% capacity utilization because the batteries last too long.

No that's right, they last too long, yeah. But I think the thing to think about is that we haven't for years scratched questions like, how do we scale out heat pumps? How do we make desalination work far, far better? What do we do to enable circularity within industrial processes and the consumer economy? And I think that we have those choices ahead of us. I mean, if everybody in Sub Saharan Africa behaved like an American in the 1970s with their muscle cars and expanding their houses and buying AC, then we've got a big problem.

AC is not, because AC pairs beautifully with solar well, so it's the least of the problem, although it requires minerals, but that's it.

Except that when people think about AC today, they think about it being powered by coal plants, right, in India somewhere, right?

Yeah, I love this because I think there's been three points where you've been more optimistic than me. You think that regulatory capture pushback will lose to these fundamental pressures. You think that the data centers will get built a bit quicker than me. And you think that we will generally use these technologies for the good rather than for the bad. And I think, where I am on that last one is that that's a choice that we get to make. That is a choice, and it will not be made by Elon Musk or by the CEOs of the tech titans. It will be made by society as a whole. And I suppose this week, it's a little harder to be optimistic than last week. But overall, I think it's just a choice that we have to be prepared to fight for, and to make sure that those technologies are used for the good. Because we could be in a land of pain otherwise.

Well, that's why more people need to listen to our two podcasts. Michael, I think that's the ultimate message there.

That is a very good place to end up. And I couldn't agree more. The world will be set to rights if only everybody who's listened this far tells all of their friends and all of their colleagues and families to listen as well, because I think it's been an absolutely tremendous conversation. Thank you so much.

Thank you so much, Michael, appreciate it.

So that was Azeem Azhar, founder, CEO and thought leader behind Exponential View. As always, we'll put links in the show notes to resources that were mentioned during our conversation. So first of all that's Exponential View, the substack run by Azeem, and he's offered a year's free subscription to anyone listening to this podcast or watching the YouTube video who wants to sign up. We'll put details in the show notes of how to use that offer. We'll also put in links to the conversation I had with Jenny Chase on solar, which was episode 173 of Cleaning Up, and the conversation that I had with Hans Eric Mellen on battery recycling, and that was episode 165. We'll put in links to the two pieces that I wrote on hydrogen way back when, which was Separating Hype from Hydrogen, and also the audiolog versions, which is audioblogs three and four on Cleaning Up. And finally, two more, we'll put the video that shows behind the scenes at the X-AI data center, mentioned by Azeen, and also the article written by David Khan of Sequoia Capital on AI's $600 billion question. And with that, let me just thank the production team and say that I hope we see you this time next week for another episode of Cleaning Up.

⁠Can Exponential Growth Save a Finite Planet? Ep187: Azeem Azhar

Michael Liebreich

Azeem Azhar

Listen On

Featured Episodes

Recent Episodes