According to a new paper, two integral factors influence why some clean energy technologies get cheaper much more rapidly than others: design complexity and need for customization. In this episode, authors Abhishek Malhotra and Tobias Schmidt discuss their findings and the implications.
Text transcript:
David Roberts
In 2021, a group of Scholars at Oxford University published a paper that made big waves in the energy world. It argued that key clean energy technologies — wind, solar, batteries, and electrolyzers — are on learning curves which guarantee that, if they are deployed at the scale required to reach zero carbon, they will get extremely cheap.
This is, as they say, big if true. In September, I had one of the lead authors, Doyne Farmer, on Volts to discuss the paper in-depth. He made a convincing case for the paper’s thesis, but when I asked him why these technologies were on learning curves and others weren't, he could only speculate.
That's the question that's been on my mind ever since. Why are some clean energy technologies getting rapidly cheaper while others aren't? What is it about particular technologies that make them amenable to learning curves?
I cast that question to the academic gods, and lo, they returned with a paper, and that paper is what we’re here to discuss today. It’s called “Accelerating Low-Carbon Innovation,” by Abhishek Malhotra of the Indian Institute of Technology and Tobias Schmidt of the Swiss Federal Institute of Technology.
It sets out to chart technologies against two basic axes: design complexity and need for customization. That creates a schema that can help illuminate why some technologies developed quicker than others.
I don't want to say much more than that, since I have my Malhotra and Schmidt here with me to help explain.
Gentlemen, welcome to Volts. Thank you for coming.
Abhishek Malhotra
Thanks a lot, David. It's great to be here.
Tobias Schmidt
Thanks for having us.
David Roberts
Abhishek, let's start with you. You're the lead author on this paper. Tell us just a little bit about what drew your interest to this question and what you thought the literature on learning curves was lacking.
Abhishek Malhotra
Thanks, David. So, this question is something that preoccupied me throughout my PhD. And in fact, this is the last paper that I wrote as part of my PhD, so it's kind of something that I was digging into throughout the three or four years I was working on the question. I think my starting point, personally, and this might be different for Toby, but for me, it started out with technologies for rural electrification in developing countries. So, I was looking at decentralized solutions to use renewable energy sources like Solar PV and batteries to provide electricity where the grid does not exist in off-grid systems.
And you can do this in different ways, right? You can have really small, modular systems which have a solar panel and LED, and basically it's a lantern. And these are things that work. These are things that are mass-produced consumer products and are pretty common in Sub Saharan Africa and parts of South and Southeast Asia. At the other end of the spectrum, you have mini-grids. So it's like the grid, but not connected to the grid. It's a small, self-contained system which has its own power generation. Maybe if it's intermittent energy, some sort of energy storage, and you connect all the households and that kind of powers the entire thing.
Although, in theory, they are great and end up being the low-cost, least-cost solution in a lot of contexts. We found that they don't diffuse as easily as these smaller, simpler, standardized modular systems. And that kind of got us thinking, regarding whether there are any systematic differences across technologies that make some technologies easier in terms of the diffusion, in terms of their progress down the learning curve, and whether there are others that are inherently much more difficult because of the nature of the technology itself. And that's kind of an idea that we built on, and that's what led to the ideas that we present in the paper.
David Roberts
And so, nothing before in the learning curve literature had really answered this question to your satisfaction. I know there's been some speculation, there's learning by doing. There's a lot of stories about why technologies do or don't develop quickly. What is your sort of contribution here?
Abhishek Malhotra
So I think one big contribution is to assimilate bits and pieces that already existed in the literature. So, there are papers that talk about how complex technologies, technologies that have a lot of components which interact with each other in non-simple ways, which are very difficult to model from first principles. I'm thinking of nuclear plants here, right? A lot of things you only figure out once you actually put them up. Some things go right, some things go wrong, and then you incorporate those learnings in the next generation.
There is evidence in the literature that such technologies are inherently much more difficult to innovate in and improve slower, than something that's simple and standardized and not as complex, something like Solar PV. So you have that one axis that we talk about: "complex" versus "simple" technologies.
David Roberts
Right.
Abhishek Malhotra
Something that did not receive as much attention in the literature is the distinction between relatively standardized technologies versus technologies that are inherently much more customized to specific contexts. So again, if you want to contrast Solar PV as being something that's very standardized, a solar panel in India looks pretty similar to a solar panel in Switzerland, or a solar panel where you're sitting in the US. Things are very different for something like biomass.
David Roberts
Right.
Abhishek Malhotra
Where, although the basic principle of burning biomass is pretty much the same everywhere, in practice, the design of the plant really depends on the feedstock, on what you're burning to produce the energy, which you then use to produce steam and then to turn a turbine.
So we kind of look at the existing literature, we assimilate it in a way that makes sense to us and hopefully to others. And it's a heuristic that hopefully people find useful in thinking about technologies in a systematic way.
David Roberts
There's a slight disadvantage here at the podcast being a listening medium, since you have a really great graph in the paper. So, if listeners can imagine one axis, as he mentioned, is "design complexity", the other is the "need for customization". So you have two axes, and then that creates, sort of, a nine square grid where you are characterizing by increasing levels of both. And that provides a nice, really nice visual way of sort of laying out where technologies fall on this grid. So, maybe let's start with degree of complexity. Toby, maybe you want to jump in here and explain what do we mean by that access and what do we mean sort of what are the divisions as you increase in complexity?
Tobias Schmidt
So this, as Abhishek said, has been something that — by the way, also Doyne Farmer has been working on... So, the role of complexity, design complexity, in technological innovation and how it can slow innovation... What it means is, as Abhishek said, how many components are needed to build a functioning piece of technology? And that's very important. How integral are they? So how much interrelation is there? And are those parts, are those components communicating in an easy way with each other? Or is it a non-simple way?
And I'll give you an example: So, for instance, a laptop or a smartphone — of course — also has quite a few components, right? From that end, it would be rather complex. But it's not complex because the way those components interact is relatively simple and it's very clear. So you can always think "Is the technology complex?" by asking this one question: If I change one component, like replace it by something upgraded, for instance, will I have to change a lot of other components? And do I know how to change it? If I took out the battery of my phone and replace it with a bigger one, I wouldn't have to do anything, right? Because it's the same voltage and then everything else will work.
Whereas, if I take a wind turbine and replace the blades by longer blades, then I would run into problems, right? Because, at some point, this turbine will fall apart.
David Roberts
So, complexity is not just more components, it's also more components that are more integrally tight to one another.
Abhishek Malhotra
Exactly.
David Roberts
So in your graph, the three levels are: simple, design-intensive and complex. So what's the sort of classic "simple technology"?
Tobias Schmidt
So "simple technology", as Abishek said before, is — from the design perspective — is Solar PV, like a cell or a module. There's not a lot of component and it's very clear how they interact. Another example would be a light-emitting diet, right? An LED lamp. A more ... like a design-intensive product would be, for instance, a car or a wind turbine. And then a really complex technology, also in the literature is also called "complex product systems", that's for instance a nuclear power plant or a combined cycle gas plant and so on. So where there's a lot of non-linear interactions between subcomponents.
David Roberts
And it might be, I guess, intuitive why low complexity devices get cheaper more quickly. But do we have a more, sort of, specific answer to why that is? Or, I guess, a more complex answer to why that is?
Tobias Schmidt
It might seem to be an easy question, but it's not necessary. We thought about this quite a bit, and actually, in a project that we're currently working on, Abhishek and I, together with Lynn Kaack at Hertie School in Berlin, who's a Machine Learning expert, we really wanted to understand this a bit better. And we also said "Hey, there's other forms of complexity", especially we should not only consider design complexity, we should also consider complexity in the manufacturing.
If we think about Solar PV, why it got so cheap, what's the innovation? It's not in the design. There's not a lot of potential in the design, right? Because it's so simple. But most of the innovation actually happened on the manufacturing side. So we were able to slice those ingots much thinner and thereby lose much less material. We were able to handle those wafers that you have sliced this ingot in, much, much better, much faster, lose much less material, and so on. Those were actually the number one cost drivers, besides economies of scale in affordable takes. So, what we find in this new research, that is not published yet, but what I can give away, is that it looks like a certain level of complexity in the manufacturing process is needed in order to see high learning rates. And that makes total sense.
David Roberts
Yeah, that's just because there's just more points at which you can improve things, right?
Tobias Schmidt
Exactly. If you think about a very simple technology that also has no complexity in the manufacturing process, let's say a stone axe, you won't see dramatic cost declines in stone axes as long as they're produced, probably even if they're produced with a robot, because it's just like, I don't know, five steps that you need to produce this thing, right. And how much can you optimize there, how much can you innovate there? But for a solar module, Solar PV module, you have hundreds of manufacturing steps, so you have many, many more opportunities to innovate and reduce the cost.
David Roberts
So you need device simplicity and manufacturing complexity together.
Tobias Schmidt
At least some degree, exactly. And what's interesting is, of course, these are not completely independent, so it's unthinkable of having a complete ... let me quickly explain for the sake of being clear here what we mean by manufacturing complexity or process complexity. It's the number of steps — so, it's kind of related to the number of components and design complexities. Here we have the number of steps that you need to produce it. And again, how integrally linked are they, right? If I change one step in the manufacturing, do I have to change others and do I know how to change them or do I have to try out?
When you have this high complexity, then, learning by producing, you could say, is very valuable. Building this experience. And if you think about a very complex product like, let's say, a nuclear power plant or a combined-cycle gas turbine or so, it's not very likely that you will ever be able to produce it in an integral way, right, where you have mass production, where all these steps are in one line.
David Roberts
So there's just fewer points for improvement in the manufacturing process.
Tobias Schmidt
Exactly. But there's still a lot of learning by using. So learning during the use phase, that's when you have this design complexity. The problem though, is often that you don't build a lot, right. If you improve something, then you'll only be able to implement it in the next big, large facility, and then maybe it doesn't turn out the way you wanted it to turn out again. And then you have to wait until the next big, huge plant is built. Whereas, if you optimize a manufacturing line and you ramp up the production of the technology that the manufacturing line is producing, then you have a lot of new ways of trying out, right? You have trial and error and learning by doing what you are producing.
David Roberts
So, Abhishek, then let's talk about the second axis, which has to do with the degree of customization and the three levels you have here are: "standardized" to "mass-customized" to "customized." So, again, just give us a little sense of exactly what we mean by that and maybe give us a few examples so we can wrap our heads around it.
Abhishek Malhotra
Yeah, luckily this axis is a bit easier to understand, at least for me. So again, just to build on the previous example that we've already talked about, an example of a very standardized product is, again, a solar module, which looks the same everywhere. An example of something that's mass-customized, which maybe has a standardized platform that you can build on, but still there are tweaks that you need to make depending on the specific site. You can think about rooftop Solar PV systems, where the sighting, the installation, the mounting structures need to be somewhat adapted to where you're deploying the technology.
David Roberts
Right.
Abhishek Malhotra
And something that's really customized, something that's really site-specific is something like your building envelopes, right. And especially when you're retrofitting buildings, it really depends on what the building already looks like, it depends on the climatic conditions, it depends on personal preferences. So a lot of site-specific factors that come into play, and basically it's not a standardized thing that you can just mass-manufacture and apply everywhere.
David Roberts
Right. You mentioned inherent features of these technologies and I guess some are inherent, but I'm also, I guess, curious how fixed you think technologies are on this particular axis. I mean, presumably, it is possible to become more standardized over time, at least for some technologies, is it not?
Abhishek Malhotra
Definitely. And that's part of the learning curve, right? These technologies are dynamic in these axes. They do move around. So early on, when you have a lot of design uncertainty, you might have a lot more variation in design. This is something that we've seen happening in wind turbines, right, where early on, you had a lot of designs which were adapted to the local conditions, the local wind regime. But gradually, over time, you saw platforms emerging and you had platform-based generations of turbines where you could have a standardized core, but the coatings on the blades, for example, would be adapted to whether it's a cold climate or whether it's a desert condition.
To optimize those things, you have adaptations based on the local wind regime, but that builds on a standardized core and that's something that evolved over time. So, definitely, I think a big part of learning is how can you standardize whatever you can and use that as a base to build on and minimize the adaptations that need to be done for a specific site.
David Roberts
Right. Okay, so we have these two axes: "degree of design complexity" and "need for customization". And they each have three levels each. So I got this nine-square layout here, and this is ... I want to talk a little bit about how we should think about this in reference to policies, sort of how this should inform our policy approach to these things. And this is obviously not going to be as simple as yes or no, it's on a learning curve or not. Obviously, real life is more complicated than that, but so you have this gri, and at the bottom left, you have technologies that are both "simple" and "standardized." That's your Solar PV panel. On the top right you have technologies that are both "complicated" and "customized," like current nuclear power plants and biomass with CCS. And then, sort of, in the center, you have medium complexity, medium design intensity.
Let's talk a little bit about how to translate this into policy. So, I'm going to try to convey another visual image to listeners. I hope I'm not taxing people's visual imagination too much. So, you also have three types of technologies. If people can just picture the lower left square, right, simple and standardized, that's Type 1. The three squares around that, which are mass-customized and design-intensive, or both, are Type 2, and then around the edge, the top and the right, are Type 3 technologies.
The reason I try to explain this is because it's relevant to how you think policies should focus on this. So the first thing I want to ask is, sort of, like, the most familiar policies we have for encouraging clean energy are your basic deployment policies, subsidies for deployment. So let's talk about which kinds of technologies that works for and which kind doesn't. And more generally, just sort of like what kind of policies work for what types of technologies? Here, I'm sorry, it's a very broad question.
Tobias Schmidt
You can use this framework in probably three ways from a policy perspective. One is to say, "Okay, where are the key winners?", right? Rightfully or not, it's actually no good empirical evidence, but we always blame policymakers for being bad at picking winners. You remember Solyndra case and so on?
David Roberts
Oh yes.
Tobias Schmidt
Tesla was also paid, so. This could be a good starting point for identifying winners, right?
David Roberts
Well, what do you mean by winners? You mean winners that technologies that can be expected to...
Tobias Schmidt
To learn fast.
David Roberts
To learn fast.
Tobias Schmidt
We still have to ... nowadays we keep forgetting that often. But PV, until like ten years ago or even less, used to be the most expensive way of producing electricity.
David Roberts
Right.
Tobias Schmidt
Since then it's become the cheapest, if you just look at it pure LCE basis, without considering system integration. But that's amazing, right?
David Roberts
So does that show us that the bottom left box here, "simple" and "standardized", is just going to be fastest? All things being equal.
Tobias Schmidt
If — that's the important thing now for your policy question — if you find governments who then say, "Okay, this is going to learn fast, so we're going to spend quite a bit of money to bring it into the market, to allow for all of this experience being built up, and therefore to allow for all these cost reductions". And that's for instance what a couple of governments did, right? For instance, the Germans and the Spanish and so built massive feed and terror schemes, massive subsidies around PV, and that brought this industry out of the niche and thereby helped this industry really, really get to those cost reductions.
That's the first thing. You can identify those technologies that are very likely to become much cheaper. You can also identify potential losers, right? Technologies that are very unlikely to become cheap. Now, if you say, I still want some of these technologies because, I don't know, they are so relevant to the system or because of my industrial policy strategy or whatever, then you can use this framework in another way. And you can use it by saying: "How can we reduce the need to customize things?" For instance, by making sure we standardize, or we make sure that there is a standardization process tied to this deployment policy, or — that's even much harder, I think — you can try to even push industry, or provide incentives that industry reduces the complexity of technology. I guess the most prominent example would be here to move from current nuclear technologies to smaller modular reactors, which would, kind of, in an ideal world, from a nuclear perspective, would bring nuclear reactors from the top right, which is kind of worst spot to be in from the learning side, to — at least as close as possible — to this Type 2, to the center of this framework, right?
David Roberts
Right. Well, this gets to a set of questions I want to get to later, but let's go right there, which is: Very often these Type 3 technologies — which are sort of around the edge of the grid, which are very complex or very customized — are very big. And that makes it, for reasons you discuss in the paper, sort of difficult for them to learn one to the next because very often, a given country will just build one or two, and another country will build one or two, and those two countries are not necessarily sharing the learning.
Spillover is not happening between countries. It's, sort of, more difficult because of the large gap between generations, as you say, and then also the large geographical gap often between technologies. It's just difficult for big things to do this. Thusly, the policy that the paper tends to recommend — if you want to encourage these technologies that are Type 3 — heavily depends on international cooperation. Abhishek, maybe you want to jump in on that and just sort of talk a little bit more about why these big things can't really develop quickly if they are kept national.
Abhishek Malhotra
Yeah, and I think Toby already mentioned that, to some extent, that just because of the limited experience that you have with the technology, they don't have the opportunity to go down their learning curves as much as something that's standardized and mass-produced. But this becomes particularly important at the early stage of the technology, right. We're talking about huge projects which, at an early stage, the typical policy tool used to support them is to fund demonstration projects, right?
David Roberts
Right.
Abhishek Malhotra
An example of that is, again, CCS plants, which are huge, capital intensive. We have a bit of experience, globally, with deploying demonstration projects for CCS. But again, there are studies that have shown that there has not been a lot of interproject learning. And so there have been calls to have a global platform and globally coordinated deployment of CCS projects, where you can have kind of a shared learning, so that the learnings from one project can spill over to the next one and to the next generation. Again, in theory that seems like a great idea, but it's really in the practice where that often falls apart because as soon as you have technologies that are close to commercialization, people, private firms obviously do not want to share the knowledge that they have.
David Roberts
That's right. You're attacking against financial incentives there.
Abhishek Malhotra
Exactly.
David Roberts
Yeah. The reason I bring this up is because you frequently hear fans of nuclear power saying, "Well, nuclear might not be on a learning curve because we're doing it in this dumb way, but if we just built more, it could get on the learning curve." And so I just want to emphasize that that is true insofar as there's a lot of international sharing of learning and information.
Abhishek Malhotra
Exactly. And it is true because we've seen that happening already, right? But in specific national contexts, if you take the example of France or if you take the example of South Korea, where they had national nuclear programs which were very centralized and had continuity, you did see some progress. You did have some learning, not to the extent that we saw with Type 1 technologies like Solar PV or LEDs, but still they were cost reductions. How well that translates into global learning is something that remains to be seen because every country has its own safety regulations, design approval standards.
And so, facilitating global learning then becomes a challenge because of these often regulatory barriers that technology faces in promoting global spillovers of knowledge.
David Roberts
Are there other, just sort of, top of mind examples of how sort of what form that kind of international cooperation might take? Are there good examples of where there has been some good international sharing of learning on technology development?
Tobias Schmidt
I can only think of military-related technologies, to be frank. It has to be with two things. It's this standardization also across international or across armies in that case, and I think NATO is a great example, right, where this happened over the last decades, where there is now NATO standards, and all the NATO members, all the militaries and all their technologies and all their technology suppliers, to some degree at least, conform to those standards. And that makes ... maybe it's not so much about learning because costs in military, the costs don't matter that much, maybe, but it's effectiveness, right? It's much about effectiveness of a military.
And you can see that even in Ukraine, now. I mean, I'm not an expert on this, but how I observed ... this is a bit. Okay, it's quite interesting how those different NATO members supplying weapons to Ukraine, how often those systems can, relatively, easily be integrated because they have these NATO standards, especially the newer ones. But there's a huge interest, right? There's a huge common interest. There was a huge threat that kind of brought all of them together. And I would argue climate change is at least as big.
David Roberts
Right.
Tobias Schmidt
But the problem is a bit that it's a slow moving crisis. It's not as imminent threat as, "Okay, tomorrow we'll be hit by a nuke."
David Roberts
Yes. And international cooperation, I mean, even on more basic stuff, has been slow to unfold. I can think of two implications you could draw from this research. One is, we can sort of now chart which technologies are likely to be fruitful if we pursue them most quickly. And so we should just pursue those. And insofar as there are gaps, they're more likely to be filled by further investment in those technologies than by trying to push forward these Type 3 technologies, which are like wagons in mud. Of course, the other way to take it would be we have to have these Type 3 technologies to fill the gaps, so we have to figure out better ways to nudge them onto learning curves. Do either of you have either a preference of those two ways of looking at it?
Abhishek Malhotra
I think there's a case to be made for both. There is a case to be made for being reasonable about what to expect from Type 3 technologies, from very complex and very customized technologies. And if expectations do not pan out, then we shouldn't be surprised. So, if there are modeling studies that have a big role for certain Type 3 technologies, like CCS, for example, in meeting climate goals, we have to have the understanding that that's going to come at a cost.
David Roberts
Right, I'm curious if with this sort of image in mind, if ... they try to represent learning curves, I think, in these models somewhat. I'm curious if you think that any of the learning curves represented in sort of like the big IPCC models are unrealistic, given this understanding.
Tobias Schmidt
We're currently starting a project funded by the European Union, how to improve IAMs.
David Roberts
Just for listeners, those are integrated assessment models, the kind of models that are used by the big agencies.
Tobias Schmidt
Yeah. And they're very politically, they're very, very powerful, I would say. They have a lot of impact. Those are also the models that are being used often in the IPCC reports. And they gotten much better, but they still have a lot of ... some issues. One of them is, for instance, the representation of finance, which is typically not there. Everything is being financed just like that. Money is just there. And that's one issue, right, if you think about these big, big, big capital investments here. And the other one is that innovation is not necessarily that well-represented. For instance, all of them, I would say, underestimate the role of Solar PV, but also smaller energy system models that only model the energy sector often are not able to even keep up with the actual deployment of photovoltaic.
And that's because those models, they typically deal with technologies in a very similar way, right. They're not, what's being called technology-rich. They're not technology-rich enough, so they don't really see those differences. Another thing is also they typically underestimate the way how many variable renewables we can integrate into the grid and so on. We're working on that in this project now. And one of those questions is exactly, "How can we better understand learning rates of early technologies where we don't have data," right? So for PV, it's very easy to draw a learning curve, but what about direct air capture, right?
David Roberts
Right. My next question was, do you think this schema could be predictive? Because ultimately to be really useful, especially if it could have informed IAMs, do you think it could be sort of quantifiable enough to actually inform IAM parameters like direct air capture? Like, what can we say about direct air capture based on this understanding?
Tobias Schmidt
I would say to some degree yes. And it certainly helps improve the current guessing that's going on. So there's papers and not too few of them. We would have a PhD student here who works in direct air capture and learning in that industry. And there's not too few papers, I mean, there isn't that many papers who look at the future cost of direct capture, but almost all of them that are out there right now, they assume learning rates that are just pulled out of somewhere.
David Roberts
Right. Well, where would you get them? You have to make one up to put in your model right?
Tobias Schmidt
And you could, I mean, what you can do is you can use technologies which we have, which are pretty similar, right, like analogies. But what they do is most of them, unfortunately, take photovoltaics. Like, we've seen cost reduction of about 20-25% in photovoltaics. So we assume the same here. And I'm like, "What? Stop." A direct air capture lens will never be a very small-scale, mass produced thing like a Solar PV module. So I think, in that sense, it can give you, I probably won't be able to tell you is it like a 12% or 15% learning rate, but it can tell you is it rather a 10% or 12% versus a 20% or 25% learning rate. And that's a huge difference, right?
David Roberts
Right. I'm just wondering how sort of quantitative you can get to like — I was actually discussing with Abhishek before we started, maybe you can get into this Abhishek — if there's ways of quantifying these two axes such that you can get a richer and more precise grid.
Abhishek Malhotra
Right. And that's something that we're working on. So, as Toby said, the grid can already act as a heuristic. And if you understand how the technologies work qualitatively, you can get a sense of how fast technologies are likely to progress. But at the moment, we're also working on a project, as Toby mentioned, together with Lynn Kaack of the Hertie School, in which we're using patent data to try and tackle at least one of the axes. So we're using patents and the text in the patents to try and quantify how complex the technologies are. So how many different knowledge components do they combine, how easy or difficult it is to combine those knowledge components, and use that, using machine learning and text analysis methods, to quantify the complexity of the technologies.
And, hopefully, that will allow you to test this relationship between learning rates and complexity in a quantitative way. And I think that's something that we're very excited about and ready to test for more nascent technologies where we don't have a lot of data from deployment, where we can't reliably say, based on past experience, about how fast the technologies are progressing.
David Roberts
Right, so attempting to predict a little bit. Is there any thinking about how you might even begin to quantify the need for customization? I'm not sure what metric you could grab out of there.
Abhishek Malhotra
Right, that's the tricky one. We've been thinking about looking at products that are out there and what different designs exists. So there are some studies that have, for different purposes, tried to quantify the degree of variability in technology designs. So picking key parameters and looking at the population of technologies that are out there in the market and trying to quantify the entropy in that. So that's one possibility that we're thinking about. But again, this requires a lot more thinking and a lot more work before we can actually make that work.
David Roberts
I encourage listeners to go to Volts so you can see this grid that I've been ineffectually trying to describe throughout the thing. And I really do think it's helpful as a heuristic. And I do think it's helpful, as you say, Abhishek, if you're projecting an enormous amount of CCS in your model, and virtually all popular models do, and you have good heuristic reason to believe that it will be a shallow learning curve at best. It does seem like we should be adjusting the cost expectations for those things, adjusting those parameters based on this heuristic, eventually. I don't know if you guys have a specific expectation of that happening, but just this feeding back into IAMs and starting to really change the culture of IAMs a little bit. Is that something you could imagine happening, Toby?
Tobias Schmidt
I hope so. This project that I mentioned is mostly...most of the leading IAMs in Europe. In Europe, we have a lot of integrated assessment models other than in the US, unfortunately. And a lot of them are in this consortium, so I hope to get that message. The problem, a bit, is if you look at a lot of this negative emissions that we will need, according to all of those models to stay within the 1.5, but even 2 degree in many scenarios, temperature limit, then almost all of these technologies will be expensive, right? I don't see any, necessarily, any Type 2 technology...
The only exemption is maybe direct air capture, where I can see some designs moving in this platform direction. But a lot of them have huge shares of bioenergy with carbon capture and storage. And I'm like, oh, this, at least in our thinking about it, and all the research we did on this technology to place it in this heuristic matrix, it's top right, which is the worst box in this entire matrix.
David Roberts
It is sort of the iconic "complex and customized". It's difficult to imagine a simple or standardized biomass plant with a giant CCS facility attached to it...
Tobias Schmidt
That's mass produced, right.
David Roberts
Yeah.
Tobias Schmidt
Yeah, unfortunately.
David Roberts
And it does seem, as you say, there's so much negative emissions in these things, that these are not idle questions like the learning curve that you might or might not be able to get direct air capture on is an existential question. So it would be extremely helpful, it seems to me, to have a good heuristic going in at least.
Tobias Schmidt
Because coming back to policy, another way of providing incentives for industry and research to come up with new solutions is really R&D support, right? And then you can say, let's make sure we design our R&D programs, maybe also deployment policies, in a way that we encourage vendors, innovators to move out of this Type 3 area into the Type 2, maybe even Type 1 area. Think about how can we simplify or standardize technologies and how can we reduce complexity. Really make that part of your R&D policy and your deployment policy rationale.
David Roberts
Yeah, I was thinking about that in the context of EV batteries. So you can think of a lot of reasons based on this heuristic, why you'd want to standardize them somewhat or simplify them somewhat, but the trend seems to be the opposite, right? Like you have Tesla now integrating batteries literally into the body of their vehicle, which is about as bespoke as you can get. Or do you disagree?
Tobias Schmidt
I somewhat disagree because I think there's two things. Like for instance, if you look at Volkswagen, they have all these IDs now, but they're all on the same platform, even the Audis and the Porches, it's all one platform. So they're really in this platform area and the shapes look different, the sizes are different, but the core of the thing is the same. And what's also important is what really determines the cost of an EV is really the battery, still. And there that's very standardized. There's different chemistries within lithium-ion, but within that, that's really relatively standardized.
And again, it's mass produced, as again, complexity on the manufacturing side and that's how we actually see those very high learning rates. But again, you have this, as you said before, right? You have this incentives, and obviously I said that before as well, you have this industry incentive to not you have to differentiate between your competitor. So standardizing across the entire industry is always hard.
David Roberts
Right.
Tobias Schmidt
But you see a lot of the standardization within those companies.
Abhishek Malhotra
Just one word of caution that I want to introduce here, just so that we're on the same page. I do not want this framework to be seen as a rationale for standardizing everything, which is often not great for innovation, right? So you don't always want to produce more of the same standardized thing, cheap in scale, because that can also act as a roadblock for innovation in other ways.
David Roberts
Yes. Well, people talk about this in the context of PV, right? The idea that the sort of rapid, massive build out of old school Solar PV kind of had the effect of shutting down a lot of innovation. Do you think that's true?
Abhishek Malhotra
Yeah, and to some extent, that's natural, right? That's something that happens in any industry over time, as you have a dominant design, as it's called in the literature, which is clearly better than all the competition. It dominates, it gains in market share, other variations are unable to compete. And that's good because then you can achieve scale, you can leverage other processes to bring down costs because you can scale up production. You can produce the same thing over and over and learn from it cheaply, efficiently, and that's good. That happens in every single industry over time. But I just want to caution against prematurely imposing standards, and that is a debate that happens in a lot of industries, right?
David Roberts
Right.
Abhishek Malhotra
In India, I can give you the example of battery swapping.
David Roberts
Yes. Right.
Abhishek Malhotra
So, because batteries have a high upfront cost, often a way to promote the technology is at least for batteries or electric mobility in India, is to have battery swapping so that you don't pay for the entire battery upfront and you don't even own it.
David Roberts
I did a podcast with someone who's doing battery swapping in scooters in emerging economies.
Abhishek Malhotra
And there's a big debate about whether you need standardized interfaces to enable that on a large scale so that every manufacturer of batteries can interface with every scooter or three-wheeler. But I'm kind of wary of that debate because I feel like it's a bit premature. I think, we need to still wait for what the best design is before we can impose a standard, and we don't want to stand in the way of innovation.
David Roberts
Sure. But you could theoretically argue that by standardizing the battery, you are enabling innovation in scooters and scooter rental business models. And it's only by standardizing the battery that you enable innovation in the surrounding ecosystem.
Abhishek Malhotra
That is true. But, again, then one has to be very aware of the implications of that standardization, right? You're redirecting the course of innovation.
David Roberts
Right.
Abhishek Malhotra
Then, there needs to be agreement about whether that's the right way to go or not.
David Roberts
Right. Yeah. Well, one thing the schema makes clear is that industrial policy is not a straightforward ... it's a complicated and delicate business. It's a lot more complicated than you might think. It needs to be thought through and customized perhaps.
Tobias Schmidt
Yes.
David Roberts
But the policy implications are probably a whole ... you could probably write a whole another paper about the...
Tobias Schmidt
Industrial policy implications. Yeah, we have other papers on that, but it's very true. It's not the easy answer, that all you need is a carbon price of $26.50. But I think this ... I hope this opens the eye, that we should not throw the same industrial policies at very different technologies that learn very differently and then be surprised about the very different outcomes that we observe. Unfortunately, that's typically the case, at least in the West, but we should get smarter about this.
David Roberts
Yes. Alright, well, I'm going to wrap it up there. Thank you two so much for coming on. This is a great complement to our initial early learning curves that we did here on Volts a few months ago. This is a great capstone to that. So thanks for taking the time, guys.
Tobias Schmidt
Thanks a lot for having us.
Abhishek Malhotra
Thank you, David.
David Roberts
Thank you for listening to the Volts podcast. It is ad-free, powered entirely by listeners like you. If you value conversations like this, please consider becoming a paid volts subscriber at volts.wtf. Yes, that's volts.wtf, so that I can continue doing this work. Thank you so much, and I'll see you next time.
Which technologies get cheaper over time, and why?