So I’ve worked as an analyst or consultant for the past ~12 years, and made (and read) many such analyses. Most a lot less technical than the ones discussed here. They’re all well within “All models are wrong but some are useful” territory.
When I read ones that are very technical and use a lot of data, they’re a great source of what assumptions I should use in my own thinking, but they tend to overlook something critical that makes the final output much less useful than the inputs and intermediate results. Like assuming automotive OEMs will pay aerospace prices for a material instead of using a cheaper grade. Or using a linear approximation for the impact of vehicle weight on MPG that implied a weightless car would only save 1⁄3 of the fuel of a normal one. Or assuming some comparison metric won’t also be undergoing iterative change and improvement over time. And so on.
Heliostat costs have come down by >2x, but it has nothing to do with the “learning curves” finance types like to point at, it’s just a matter of how much time smart people spent thinking about them
That’s pretty much the standard explanation of what learning curves are, abstracted away from the specifics of a given process/product/industry.
But to your specific question here: I would definitely like to see more experimentation with solar thermal, especially for things like industrial process heat. Seems underexplored.
On electricity generation though, I think there are a few factors that make me think it’s unlikely to compete well with PV.
Electrical energy storage costs are falling. Realistically Li-ion will be <$100/kWh by ~2030, and they have much higher round trip efficiencies (>95%). I doubt vanadium flow or sodium ion or anything else will be at that kind of scale by then, but those could bring it even lower or limit how much costs spike with rising Li demand. We’ll also be building a lot of them no matter what, for EVs, home batteries, and the like, and many of those will interface with the grid fairly intelligently. We’re already starting to see some utilities install a few hours of battery backup, because they can already be more cost competitive than gas peaker plants.
As much as there’s ridiculous overhype about hydrogen, we’re likely to be making a lot of it in the future, because some applications will need hydrocarbons either for chemical feedstocks or for liquid fuels: eSAF, methanol for marine fuel, ammonia production. This means we’ll be way overbuilding renewable energy generation relative to immediate electricity demand on average, which will make it easier to deal with less-than-optimal production delays using demand response from electrolyzers. In principle these carriers can also be used to effectively ship solar power long distances using the same kinds of tanker and pipeline infrastructure we have today, but I doubt much of that capability will be used for electricity production except maybe in remote areas.
Fundamentally silicon has a near-optimal bandgap for PV and had a headstart because of its abundance and use in other electronics, and other technologies (III-V, CdTe, OPV, etc.) all have/had glaring weaknesses. I’m pretty hopeful about perovskites in a way I never was about those others. I think if you look back on this post in 2030 you might find we’re in a world of solar cells that are 1⁄4 the current price, have more stable output across different temperatures and light levels and under indirect light (and so produce more hours/day), weigh much less, and are multi-junction with overall efficiency >30% and plausibly >40%. The physics nerd in me hopes that someone will figure out cheap metamaterial waveguides that let us make thin film multijunction concentrated PV which would easily get us to much higher efficiencies, but I have no sense of a timeline for something like that.
As for your point about high electricity prices not holding things back, I think you might not be thinking the counterfactual through enough. In a world where electricity prices were 1/10th as high and mostly from renewables, what else would change over the following 10-30 years as people make choices based on this new information? A lot of things! All of a sudden:
Buying 40 kWh of electricity looks a lot better than burning a gallon of gasoline or the equivalent amount of natural gas in your car or factory.
It makes a lot of sense for houses even in more extreme climates to be built or renovated with air source heat pumps instead of furnaces.
Large-scale desalination and indoor agriculture start looking affordable enough to help a lot of people improve their quality of life, improve access to a varied diet of fresh foods, reduce the need to damage ecosystems to expand agricultural output, and improve our civilizational resilience to climate change.
Direct-air carbon capture stops looking absurdly expensive (or at least reduces to a capex problem of the kind that industry and engineering regularly overcome with normal kinds of efforts).
Extracting valuable minerals from waste (or seawater, etc.) becomes viable even if the process is energy intensive.
Data center operating costs fall by 50-60%.
Those are just top of mind. Markets assign prices, but costs come from atoms, joules, time spent, and ideas. Drop one of the input costs to almost zero, and that ripples through everything else by changing the tradeoffs.
First off, I’d like to say that we probably mostly agree. But...
That’s pretty much the standard explanation of what learning curves are, abstracted away from the specifics of a given process/product/industry.
There’s a big difference between curves based on “time smart people spent thinking” and curves based on “money spent”. That was my point.
Realistically Li-ion will be <$100/kWh by ~2030
No, I don’t think so. Maybe you were looking at BloombergNEF prices, but those were heavily weighted towards subsidized Chinese batteries for domestic vehicles. US battery pack prices were ~2x the BloombergNEF global average a couple years ago, and Tesla charged $265/kWh for their grid storage without supporting infrastructure.
Also, LiFePO4 lifetime is overstated because cycling and calendar life degradation interact because cycling cracks the anode SEI.
As much as there’s ridiculous overhype about hydrogen, we’re likely to be making a lot of it in the future
Water electrolysis requires slightly bigger subsidies than the IRA ones, which were just meant for bootstrapping. Natural gas will continue to be where hydrogen comes from.
Capital costs are bigger than the electricity costs. I wrote this.
less-than-optimal production delays using demand response from electrolyzers
Per above, that doesn’t make sense.
methanol for marine fuel
You mean, for dimethyl ether? For marine diesel engines? If anything, that makes more sense for diesel trucks because particulate pollution is a bigger problem with those.
other technologies (III-V, CdTe, OPV, etc.) all have/had glaring weaknesses. I’m pretty hopeful about perovskites in a way I never was about those others
Huh? CdTe works fine, it’s been used on a large scale, it’s just not quite as good as Si. And the good perovskites are unstable, and I don’t see that changing. If anything multilayer Si/CdTe seems more likely.
cheap metamaterial waveguides that let us make thin film multijunction concentrated PV which would easily get us to much higher efficiencies
Sure, split-spectrum concentrated solar seems appealing in some ways, but it’s just not happening.
Buying 40 kWh of electricity looks a lot better than burning a gallon of gasoline or the equivalent amount of natural gas in your car or factory. It makes a lot of sense for houses even in more extreme climates to be built or renovated with air source heat pumps instead of furnaces.
This decision is based on residential electricity prices. And again, California charges 10x the cost of production.
Large-scale desalination
Even there, capital costs > electricity costs, tho there is some tradeoff between them. And desalination is already feasible.
I am not at liberty to share some of the details, but I’ve seen 3rd party accelerated testing data showing perovskites from some companies stable with expected 20+ yr module lifetimes, and real world multi-year testing data with very little efficiency loss. In addition, while the nameplate efficiencies are definitely lower, they have much more stable output curves under a wider range of weather conditions, on a curve which in many climates would result in greater total kWh output per day/week/year, and spread more evenly throughout the day, compared to Si.
And yes, CdTe works fine at scale, but it’s not something we’re ever going to scale to TWp/year, there just aren’t enough cadmium and tellurium we can readily mine. And to get to an actually decarbonized world, we’re going to need to increase total electricity production several fold, and a few times more as more countries become more developed, so multiple TWp/yr is where we’re going to need to be. We’re already above 1 TWp/yr silicon pv production capacity, mostly in China.
As for batteries, it sounds like you’re talking about prices, but I’m talking about costs. I really don’t care what Tesla charges, I care what it’s going to cost them and the next dozen manufacturers to make batteries as they scale production. And yes, I’m aware of the NMC/LFP differences. I still think we’re going to see more and more shifting towards LFP, and those problems continuing to get less severe.
I should also add: I know residential scale solar tends to be extremely inefficient from a balance-of-plant and installation labor cost perspective, but I just don’t see a way to maintain a 10x difference between production cost and retail price in a place with net metering laws, a planned ban on ICE vehicles, lots of sunshine and stable weather through the year in many regions, and very high overall housing costs that make financing home improvements seem much less onerous proportionally. Not for the long term, anyway.
while the nameplate efficiencies are definitely lower, they have much more stable output curves under a wider range of weather conditions, on a curve which in many climates would result in greater total kWh output per day/week/year, and spread more evenly throughout the day, compared to Si
Hmm, I don’t see how that could be the case unless you’re talking about a greater total area, and as you probably know, support structures + land costs more than the actual solar panels these days, so lower efficiency for lower panel cost is a bad deal. (If it even actually would be lower per output, and I have some doubts.)
there just aren’t enough cadmium and tellurium we can readily mine
Oh, that’s what you meant? Yeah.
really don’t care what Tesla charges, I care what it’s going to cost them and the next dozen manufacturers to make batteries as they scale production
You can look at the economics of some Li-ion battery producers. The margins aren’t huge.
yes, I’m aware of the NMC/LFP differences. I still think we’re going to see more and more shifting towards LFP, and those problems continuing to get less severe
I said LiFePO4, which is LFP.
I know residential scale solar tends to be extremely inefficient from a balance-of-plant and installation labor cost perspective, but I just don’t see a way to maintain a 10x difference between production cost and retail price in a place with net metering laws, a planned ban on ICE vehicles, lots of sunshine and stable weather through the year in many regions, and very high overall housing costs that make financing home improvements seem much less onerous proportionally. Not for the long term, anyway.
Heh, I agree—which is why I don’t think the net metering will stay.
Hmm, I don’t see how that could be the case unless you’re talking about a greater total area
Whether or not it checks out in the real world, it’s possible because PV conversion efficiencies are not constant. They’re a function of things including temperature, light level, direct vs indirect light, and incident light angle (even with antireflective coatings).
The power output from Si PV falls off quite a bit at high temperature, partial shade, or less direct light. Some semiconductors have much lower efficiency penalties under these conditions. So your Si might be, say, 22% efficient on a clear but temperate summer day at noon, and get you 220 W/m2. But it’s less than 22% efficient outside of the ~5 peak hours of daylight, or when the temperature of the panels rises above ~25C, or in winter.
So, an idealized panel that had a constant 16% efficiency all day, in all weather and all seasons, could make up for producing less power at noon by producing more power at 7am-10am and 5pm-8pm, and when there are some clouds, and when it’s very hot out, and in winter.
(Every time I think about this it reminds me of how in the 90s we compared CPUs on their clock speeds, and then the metric stopped making sense as we got better and more varied architectures and multi-core systems and such. The headline efficiency number just isn’t the only relevant point on a very multidimensional graph).
I also think we probably mostly agree. But to be clear, as I understand it, experience curves for production aren’t based on money spent, they’re based on cumulative units of product ever made.
And you’re obviously right about capex today and in the near future. My central point is that capex for any application is something that we should expect to fall over time with iteration and scaling, because that’s what industries do. But we’ll never get started if opex is so high no one bothers.
And no, I meant methanol for marine fuel, because of the rising orders for dual-fueled ships, ports talking about becoming methanol hubs, and projects being announced to make methanol for this purpose.
experience curves for production aren’t based on money spent, they’re based on cumulative units of product ever made
Eh, I’ve seen both. It doesn’t really matter here, right?
And you’re obviously right about capex today and in the near future. My central point is that capex for any application is something that we should expect to fall over time with iteration and scaling, because that’s what industries do. But we’ll never get started if opex is so high no one bothers.
I don’t think that perspective makes sense if you consider the economy as a whole. Most opex is someone else’s capex, and capex depreciation is sort of opex. I don’t think raw material costs have become relatively more important over time vs processing costs, either.
And no, I meant methanol for marine fuel, because of the rising orders for dual-fueled ships, ports talking about becoming methanol hubs, and projects being announced to make methanol for this purpose.
Huh, that is a thing. But it’s a smaller thing than LNG fuel for ships, which makes sense, because LNG is economically better, with higher conversion cost but maybe half the fuel cost of methanol. I suspect methanol fuel is more of a cheap hedge against potential EU regulations. If it actually gets bought as fuel, it would probably be chinese methanol made from coal, and meanwhile they’d be proclaiming their readiness for e-fuels, lol.
Eh, I’ve seen both. It doesn’t really matter here, right?
True, but one is a much close proxy for time-spent-thinking-about-it and real-world-feedback-obtained than the other.
Most opex is someone else’s capex, and capex depreciation is sort of opex.
Also true, but if you’re starting from an assumption that something is infeasible because its total cost is high, and capex is the biggest but not overwhelmingly the biggest component of that, then dramatically reducing the price of the non-capex component reduces the problem from “This makes no economic sense whatsoever, and it isn’t something our industry can fix on its own anyway,” to “Anyone who manages to get capex way down can disrupt this.” It’s removing a systemic constraint on the value and usefulness of other innovations.
So I’ve worked as an analyst or consultant for the past ~12 years, and made (and read) many such analyses. Most a lot less technical than the ones discussed here. They’re all well within “All models are wrong but some are useful” territory.
When I read ones that are very technical and use a lot of data, they’re a great source of what assumptions I should use in my own thinking, but they tend to overlook something critical that makes the final output much less useful than the inputs and intermediate results. Like assuming automotive OEMs will pay aerospace prices for a material instead of using a cheaper grade. Or using a linear approximation for the impact of vehicle weight on MPG that implied a weightless car would only save 1⁄3 of the fuel of a normal one. Or assuming some comparison metric won’t also be undergoing iterative change and improvement over time. And so on.
That’s pretty much the standard explanation of what learning curves are, abstracted away from the specifics of a given process/product/industry.
But to your specific question here: I would definitely like to see more experimentation with solar thermal, especially for things like industrial process heat. Seems underexplored.
On electricity generation though, I think there are a few factors that make me think it’s unlikely to compete well with PV.
Electrical energy storage costs are falling. Realistically Li-ion will be <$100/kWh by ~2030, and they have much higher round trip efficiencies (>95%). I doubt vanadium flow or sodium ion or anything else will be at that kind of scale by then, but those could bring it even lower or limit how much costs spike with rising Li demand. We’ll also be building a lot of them no matter what, for EVs, home batteries, and the like, and many of those will interface with the grid fairly intelligently. We’re already starting to see some utilities install a few hours of battery backup, because they can already be more cost competitive than gas peaker plants.
As much as there’s ridiculous overhype about hydrogen, we’re likely to be making a lot of it in the future, because some applications will need hydrocarbons either for chemical feedstocks or for liquid fuels: eSAF, methanol for marine fuel, ammonia production. This means we’ll be way overbuilding renewable energy generation relative to immediate electricity demand on average, which will make it easier to deal with less-than-optimal production delays using demand response from electrolyzers. In principle these carriers can also be used to effectively ship solar power long distances using the same kinds of tanker and pipeline infrastructure we have today, but I doubt much of that capability will be used for electricity production except maybe in remote areas.
Fundamentally silicon has a near-optimal bandgap for PV and had a headstart because of its abundance and use in other electronics, and other technologies (III-V, CdTe, OPV, etc.) all have/had glaring weaknesses. I’m pretty hopeful about perovskites in a way I never was about those others. I think if you look back on this post in 2030 you might find we’re in a world of solar cells that are 1⁄4 the current price, have more stable output across different temperatures and light levels and under indirect light (and so produce more hours/day), weigh much less, and are multi-junction with overall efficiency >30% and plausibly >40%. The physics nerd in me hopes that someone will figure out cheap metamaterial waveguides that let us make thin film multijunction concentrated PV which would easily get us to much higher efficiencies, but I have no sense of a timeline for something like that.
As for your point about high electricity prices not holding things back, I think you might not be thinking the counterfactual through enough. In a world where electricity prices were 1/10th as high and mostly from renewables, what else would change over the following 10-30 years as people make choices based on this new information? A lot of things! All of a sudden:
Buying 40 kWh of electricity looks a lot better than burning a gallon of gasoline or the equivalent amount of natural gas in your car or factory.
It makes a lot of sense for houses even in more extreme climates to be built or renovated with air source heat pumps instead of furnaces.
Large-scale desalination and indoor agriculture start looking affordable enough to help a lot of people improve their quality of life, improve access to a varied diet of fresh foods, reduce the need to damage ecosystems to expand agricultural output, and improve our civilizational resilience to climate change.
Direct-air carbon capture stops looking absurdly expensive (or at least reduces to a capex problem of the kind that industry and engineering regularly overcome with normal kinds of efforts).
Extracting valuable minerals from waste (or seawater, etc.) becomes viable even if the process is energy intensive.
Data center operating costs fall by 50-60%.
Those are just top of mind. Markets assign prices, but costs come from atoms, joules, time spent, and ideas. Drop one of the input costs to almost zero, and that ripples through everything else by changing the tradeoffs.
First off, I’d like to say that we probably mostly agree. But...
There’s a big difference between curves based on “time smart people spent thinking” and curves based on “money spent”. That was my point.
No, I don’t think so. Maybe you were looking at BloombergNEF prices, but those were heavily weighted towards subsidized Chinese batteries for domestic vehicles. US battery pack prices were ~2x the BloombergNEF global average a couple years ago, and Tesla charged $265/kWh for their grid storage without supporting infrastructure.
Also, LiFePO4 lifetime is overstated because cycling and calendar life degradation interact because cycling cracks the anode SEI.
Water electrolysis requires slightly bigger subsidies than the IRA ones, which were just meant for bootstrapping. Natural gas will continue to be where hydrogen comes from.
Capital costs are bigger than the electricity costs. I wrote this.
Per above, that doesn’t make sense.
You mean, for dimethyl ether? For marine diesel engines? If anything, that makes more sense for diesel trucks because particulate pollution is a bigger problem with those.
Huh? CdTe works fine, it’s been used on a large scale, it’s just not quite as good as Si. And the good perovskites are unstable, and I don’t see that changing. If anything multilayer Si/CdTe seems more likely.
Sure, split-spectrum concentrated solar seems appealing in some ways, but it’s just not happening.
This decision is based on residential electricity prices. And again, California charges 10x the cost of production.
Even there, capital costs > electricity costs, tho there is some tradeoff between them. And desalination is already feasible.
No, they really don’t.
Also to add:
I am not at liberty to share some of the details, but I’ve seen 3rd party accelerated testing data showing perovskites from some companies stable with expected 20+ yr module lifetimes, and real world multi-year testing data with very little efficiency loss. In addition, while the nameplate efficiencies are definitely lower, they have much more stable output curves under a wider range of weather conditions, on a curve which in many climates would result in greater total kWh output per day/week/year, and spread more evenly throughout the day, compared to Si.
And yes, CdTe works fine at scale, but it’s not something we’re ever going to scale to TWp/year, there just aren’t enough cadmium and tellurium we can readily mine. And to get to an actually decarbonized world, we’re going to need to increase total electricity production several fold, and a few times more as more countries become more developed, so multiple TWp/yr is where we’re going to need to be. We’re already above 1 TWp/yr silicon pv production capacity, mostly in China.
As for batteries, it sounds like you’re talking about prices, but I’m talking about costs. I really don’t care what Tesla charges, I care what it’s going to cost them and the next dozen manufacturers to make batteries as they scale production. And yes, I’m aware of the NMC/LFP differences. I still think we’re going to see more and more shifting towards LFP, and those problems continuing to get less severe.
I should also add: I know residential scale solar tends to be extremely inefficient from a balance-of-plant and installation labor cost perspective, but I just don’t see a way to maintain a 10x difference between production cost and retail price in a place with net metering laws, a planned ban on ICE vehicles, lots of sunshine and stable weather through the year in many regions, and very high overall housing costs that make financing home improvements seem much less onerous proportionally. Not for the long term, anyway.
Hmm, I don’t see how that could be the case unless you’re talking about a greater total area, and as you probably know, support structures + land costs more than the actual solar panels these days, so lower efficiency for lower panel cost is a bad deal. (If it even actually would be lower per output, and I have some doubts.)
Oh, that’s what you meant? Yeah.
You can look at the economics of some Li-ion battery producers. The margins aren’t huge.
I said LiFePO4, which is LFP.
Heh, I agree—which is why I don’t think the net metering will stay.
Whether or not it checks out in the real world, it’s possible because PV conversion efficiencies are not constant. They’re a function of things including temperature, light level, direct vs indirect light, and incident light angle (even with antireflective coatings).
The power output from Si PV falls off quite a bit at high temperature, partial shade, or less direct light. Some semiconductors have much lower efficiency penalties under these conditions. So your Si might be, say, 22% efficient on a clear but temperate summer day at noon, and get you 220 W/m2. But it’s less than 22% efficient outside of the ~5 peak hours of daylight, or when the temperature of the panels rises above ~25C, or in winter.
So, an idealized panel that had a constant 16% efficiency all day, in all weather and all seasons, could make up for producing less power at noon by producing more power at 7am-10am and 5pm-8pm, and when there are some clouds, and when it’s very hot out, and in winter.
(Every time I think about this it reminds me of how in the 90s we compared CPUs on their clock speeds, and then the metric stopped making sense as we got better and more varied architectures and multi-core systems and such. The headline efficiency number just isn’t the only relevant point on a very multidimensional graph).
I also think we probably mostly agree. But to be clear, as I understand it, experience curves for production aren’t based on money spent, they’re based on cumulative units of product ever made.
And you’re obviously right about capex today and in the near future. My central point is that capex for any application is something that we should expect to fall over time with iteration and scaling, because that’s what industries do. But we’ll never get started if opex is so high no one bothers.
And no, I meant methanol for marine fuel, because of the rising orders for dual-fueled ships, ports talking about becoming methanol hubs, and projects being announced to make methanol for this purpose.
Eh, I’ve seen both. It doesn’t really matter here, right?
I don’t think that perspective makes sense if you consider the economy as a whole. Most opex is someone else’s capex, and capex depreciation is sort of opex. I don’t think raw material costs have become relatively more important over time vs processing costs, either.
Huh, that is a thing. But it’s a smaller thing than LNG fuel for ships, which makes sense, because LNG is economically better, with higher conversion cost but maybe half the fuel cost of methanol. I suspect methanol fuel is more of a cheap hedge against potential EU regulations. If it actually gets bought as fuel, it would probably be chinese methanol made from coal, and meanwhile they’d be proclaiming their readiness for e-fuels, lol.
True, but one is a much close proxy for time-spent-thinking-about-it and real-world-feedback-obtained than the other.
Also true, but if you’re starting from an assumption that something is infeasible because its total cost is high, and capex is the biggest but not overwhelmingly the biggest component of that, then dramatically reducing the price of the non-capex component reduces the problem from “This makes no economic sense whatsoever, and it isn’t something our industry can fix on its own anyway,” to “Anyone who manages to get capex way down can disrupt this.” It’s removing a systemic constraint on the value and usefulness of other innovations.