I don't get the passenger rail scepticism. There are reasons it might not be effective in specific countries given geography, culture etc. (e.g. in the States) but it clearly works and is highly utilised the world over. It's most effective for travelling medium-sized distances between towns, where it's quicker and more efficient than either a plane (no security and lengthy boarding times), car or bus.
I'll need to look deeper into why Glaeser thinks trains are less cost effective. If trains turn out to be cheaper per passenger mile that would change my mind!
In terms of speed, I would think cars would be faster (since there's no waiting), then trains, then busses as the slowest?
It depends on the trains and the cities, and where you want to go on either side of the station. But in general, driving though traffic is often slower than trains which can go faster and more directly. Let's consider two typical journeys:
Reading to London in the UK is 23 minutes on the train, but nearly two hours to the same point in a car. London is slow and stressful to drive into, the trains are quick and regular. There's no flight.
Tokyo to Kyoto is over five hours in a car and only about two hours twenty minutes on a train. It's only an hour on a direct flight (2.5 with a change) but when you factor in security, baggage etc., there's not really a time saving.
(Obviously you're right that there will be plenty of journeys where trains are much slower than a car, when there's no direct line or your starting or end point isn't near a station.)
Sure, walking and cycling are always going to be cheapest form of transport per passenger mile— not necessarily the most convenient depending on distance!
The points about fusion are interesting. I think youre likely right, though I think the technology will still be important for settling the solar system (eventually) and I really hope people built it regardless. Solar power is not very efficient farther out. Though there may be other good options.
With home robots I disagree. The extravagant 20k-100k home robot that can cook a five star meal and walk your dog might be far off, but I think the math will work relatively soon for plenty of applications. A dishwasher runs once a day for an hour or two and costs a few hundred to a thousand dollars. Easy to imagine a similar price point working for “tidy up my room” or “fold my clothes." 8k for a Weave clothes-folding bot is a bit steep for many people but I think plenty others will buy it.
That's true about fusion, much more useful in the outer solar system and on long interstellar flights.
On home robots, fair enough! I'd love to be wrong here and overall I'm quite bullish on robots automating ~everything. Perhaps people with a higher willingness to pay will drive adoption and scaling. I do wonder if the price of robots stays high relative to wages if people will rent them.
Re: supersonics: what’s the actual fuel burn per passenger-mile supersonic vs. subsonic? The specific impulse graph implies it’s about 3x fuel burn for a given thrust, but that doesn’t really translate. On the one hand, if it was 3x burn for 3x thrust for 3x speed, that would be equal efficiency to current airliners. But Supersonic has MUCH higher drag, and carrying more fuel means less payload. I’d like to see more detail around that. Great post though!
Yeah I'm not familiar with the fuel burn numbers here. I know the drag spikes in the transonic regime, so you either want to fly subsonic or past Mach 1.5, not in between.
Ah, this got me to look into it more. Thrust specific fuel consumption (kg of fuel consumed per second per unit of thrust)is inversely proportional to specific impulse:
Looking at the table on that page looks like the SR-71 consumes about 4x more than modern engines. But given that the speed is ~4x higher the total fuel consumed for a given distance is about the same. But, lift-to-drag is worse for supersonics. This page says the Concorde had L/D of 7 whereas a 747 is 17.
Great article, so many great numbers. I have to comment that a hugely under appreciated aspect of space data centres is reduced latency. Large compute and storage to relay data between Earth, the Moon, and Mars solves blindspots in communication. Qualitative rather than quantitative risk.
Also, I'm a little worried, looking at your starship/methane costs that, at scale, launching objects into orbit will have cheaper energy costs than UK heating! Another form of shrinkflation?!
Good catch! Two reasons my methane price is optimistic. First, I'm using methane prices in the US. The US leads the world in fracking technology so has some of the cheapest natural gas on Earth. Two, infrastructure to transport natural gas to consumers adds a lot to the cost, that's why consumers pay over $10/MCF in the US:
It makes complete sense that, over time, cost of kg to orbit drops to $5.5 of fuel. The real light bulb for me was pricing joule-per-kg to orbit (I assume a constant?) and engaging in arbitrate against the dollar price of fuels through launch sites, regulatory environments and supply chains. That would be the market dynamic of "the launch business". (I might be late to the party on that realisation...).
But then there is an implication: if the cost of anything in orbit is cheaper than the cost of anything on earth by more than the launch cost, it should go into orbit. Huge demand.
I had never realised that cost could be as low as heating a house. So, now we have competition for energy between launch and heating. The cost of energy really is the base driver for everything
One quibble. In discussing cooling costs in space, doesn't it matter that the ambient temp in space is super low to begin with? That's a lot of diffusive force.
Low ambient temperature in space is super important! It's included in the radiator calculation by assuming the temperature is 2.7 kelvin (about the temperature of the cosmic microwave background, near absolute zero).
This is why reversible computing is more feasible in some sense in space. It is naturally cold there, so maintaining a cold computer is easier relative to Earth.
I spend a lot of time doing technology diligence professionally, and then on top of that trying to codify my intuition into formal models for what is easy to do and what is hard to do. My quick take: This is all correct. Also, way more entertaining than formal models. Nice work.
For materials, titanium is pretty abudent and if it was easier to refine and thus cheaper, it would have plenty of uses infrastructure and consumer products.
Nice analysis. This is nice work. I think for solar and compute you need to include marginal land costs and consider the alternative uses. That is the major missing cost that I think might be material to your work. For solar near good transmission connections, the land is finite and may have other good uses. Same for data centers. Many of the alternative sources have smaller land footprints so it may well be a point of competition. So tell me why you leave out marginal land footprint costs for solar? Using your cost theory, the current land for solar is the cheapest it will probably be for any future installs.
While it isn't explicitly included in the solar cost calculation, the idiot index is intended to account for all sorts of hard-to-predict costs. My understanding is that land costs in particular is a relatively small share of utility scale project costs.
Outside of cities there is a lot of land and it's surprisingly cheap!
Current price equilibrium is not likely a great proxy for future equilibriums. The current solar build is likely 0.05-0.10X US generation.
Your theory of exhaustible resources would not classify this an “unpredictable cost”. This is a predictable cost. Look at the footprint needed to 2X US generation, assuming grid ready, transmission adjacent land costs stay the same in the face of steady increased demand seems unlikely. Also, consider where solar ready land is - Canada, New England, Minnesota??? Appalachia, not very flat. Hmmm. Geographically concentrated demand for an exhaustible resource is not “unpredictable costs” and probably not a linear increase either.
This is without permitting and community frictions in the face of a large footprint. You should consider that grid ready land may become the dominant cost for solar.
Tyler’s blog is called “Marginal” Revolution for a reason.
Can solar scale to double or triple current US gen with land costs constant is a valid question? Solar can do a lot but it's not a category killer either.
The theory of exhaustible resources suggests that “the real price of an exhaustible resource should grow at a rate equal to the real interest rate.” The data are consistent with this hypothesis1.
Innovations in search and extraction can dramatically lower prices in the short term, but in the long term, we converge on a correct way to do things. Once extraction is solved, prices return to their slow upward climb.
Terrestrial land for solar is an exhaustible resource. Your installed solar cost projection is way off.
Superconductors are neat, particularly as a new electronic component for computers. That said, I'm skeptical of superconducting computers or other unconventional computing replacing our current computing roadmap:
The most exciting application for superconductors in my mind is reversible computers. And in that case, we already have superconducting materials with good properties! Though some tweaks will be needed. These computers need to be cold for accurate computation, so high-temperature superconductors aren't a pressing concern.
Other applications include power dense motors for flying cars/planes and perhaps better MRI machines. Broadly, I see superconductors as a nice boost for motors, plasma control, turbines, electronics but not an Earth-shattering unlock.
For what it's worth, solar module pricing is already far below $0.40/W in China, closer to $0.10/W is my understanding. It's 3x-4x as expensive in the US because of tariffs :(
Back when cells were over $2/watt, panels were about a third of the total cost. Finance cost, labour, civil works, structures, wiring, inverters, control systems, grid interconnect, land acquistion, design and permitting were most of the rest of the capital cost. Maintenance is not negligible too.
Sam is being far too generous with his theoretical minimum.
I guess that’s what the “idiot index” is supposed to cover. Agreed that labor/install/permitting/overhead brings actual install costs for utility scale solar ~$1.20/Wdc
I think the Austin Vernon posts I've linked are relevant here! If solar modules can fall in price with scale I don't see why other components like racking and wiring can't also fall in price. Erthos is deleting a lot of these components by putting solar on the ground and pre assembling wiring. Jurchen PEG claims one project reached all-in cost of $0.71/W including modules, that's not far off from the goal of $0.4/W!
Racking is pretty much at minimum cost of steel + galvanising + welding. Etc., mutatis mutandis. Ground works at low latitudes, not at high. Agrivoltaics with N-S vertical panels may be better there. The engineers working in the trade are not dummies.
The way to cut costs is panel efficiency. Halve the number of panels required, halve the land, labour, structures, wiring. Does not affect design, permitting, legal, inverters or interconnect though. Doesn't even affect civil all that much.
Oh, so those are addressed in the "Non-solar energy" section. I don't think they have a prospect of beating solar + batteries for electricity. Both require heating a working fluid and spinning a turbine with the attendant energy losses and equipment costs.
Instead, these are a better fit for providing industrial heat/steam.
Did not catch that remark, thanks for the pointer. Nonetheless, I still think you have way too much confidence in batteries, at least unless / until they get to 10x density.
Would love some comments about the various Doomberg articles detailing why solar / wind / batteries is nowhere near good enough currently.
While energy density is certainly a problem for using batteries in transport, for grid-storage, the space that the batteries take up isn't as much of a problem. Particularly on a site that already has acres of solar panels. In these applications, cost per kWh of storage is more important (at the pack level of course, so including wiring, cooling, controllers, etc.)
From what I remember of Doomberg's articles, they seemed like an okay if pessimistic assessment of costs today. I'll admit, solar modules and batteries are not the dominant cost any more. But I don't see why we can't push down these other cost factors.
Another point of confusion is looking at solar costs under the requirement of 99+% reliability. Solar and batteries can't do that yet. That's why I advocate for a 90% clean energy system using natural gas for the last 10%. It can be 10x greener while being slightly cheaper. Then we can wait for other solutions to push the renewables percent higher.
For off grid applications willing to adapt to intermittency, much lower prices are achievable. The prices in the non-solar section were for 97% uptime. For this use case, $20/MWh seems possible. Solar and battery prices are close to where they need to be for this goal, next step is to automate and cut red tape.
Your LEO launch cost analysis has a few flaws. First, the table used from Orca Sciences defines idiot index as "cost of ore plus energy of reduction." The energy of reduction is the thermodynamic minimum. They are comparing the real-world energy required relative to the theoretical minimum, with the point being that "in a world with cheaper clean electricity...we should expect more use of AI and Mg, and much less use of steel." That table doesn't illustrate the ratio of "other costs" to raw materials. Second, Musk has said he thinks fuel costs for Starship could be as low as $500k and all in operational costs per launch at $2m ($13/kg) - an idiot ratio of 4 as you are defining it. Also, air travel requires much more labor so one could argue the ratio should be lower. The $10m figure you cited is a near term goal, not an end state. Third, your fuel prices don't incorporate the sort of bulk deals they are doing with Linde nor the probable end state of vertically integrating, which can get methane down to ~$0.05/kg and LOX to $0.08/kg. And finally the comparison to Falcon Heavy might be a bit unfair as well, as the whole thesis on Starship is that rapid full reuse of a much bigger payload is what will drive costs/kg down. Maybe rapid full reuse won't prove to be possible, it is no doubt an ambitious goal, but if they were to reach that point then cost/kg would likely fall well below $100/kg.
> Musk has said he thinks fuel costs for Starship could be as low as $500k and all in operational costs per launch at $2m ($13/kg)
While I appreciate Musk's ambitious goals, I don't see a reason to treat his statements as good forecasts. He has a track record of failing to meet highly ambitious timelines for example. Part of point of this exercise is to establish a viewpoint independent of figures like Musk who have incentives to mis-report expected costs!
> Also, air travel requires much more labor so one could argue the ratio should be lower.
It's not clear to me why flying a plane would require much more labor than launching a rocket. The rocket/launchpad/payload requires more engineering staff to design, build, test, fuel, and repair. It also has a shorter lifetime than a plane, meaning more amortized labor hours per flight. Aviation is an established technology, rocket science is still being developed. We should expect rockets to have a higher index.
> our fuel prices don't incorporate the sort of bulk deals they are doing with Linde nor the probable end state of vertically integrating ...
Why doesn't SpaceX (or any other industry) already enjoy these super cheap prices? They’ve already expended hundreds of thousands of tons of fuel. Why haven't they pushed costs down further at such scale? Remember, my estimate for their fuel costs is already 2-4x lower than their current fuel costs. They likely aren’t leaving money on the table here, particularly with their upcoming IPO.
For that matter, if 3x lower natural gas prices are achievable, why aren’t fracking companies doing this today? It’s a competitive market and a company with much lower extraction costs would be able to enjoy large profits at today’s prices. Despite innovations in fracking and the shale revolution, US spot prices for natural gas have rarely dipped below $2/MCF in the last 30 years. $0.05/kg requires prices below $1/MCF and zero purification/liquefaction/transport costs. This seems unlikely, particularly given recent geopolitical developments.
Fair points but I think the primary way one gets to a better estimate of $/kg to LEO is to look at the actual swing factors rather than apply idiot index ratios. The launch operation center for Falcon 9 is less than 300 people, so the direct labor per launch is pretty minimal at 1000 launches/year (<$100k/launch). Whether the fuel is $500k/launch or $2m ($20/kg) isn't the major swing factor either. I would argue the major swing factor is 1) the number of fully reusable launches per Starship. At 20, the amortized cost of the build would be $5m/launch. At the goal of 100 it would be $1m/launch ($10/kg.) Falcon 9 are up around 25-40 now and given the more durable construction Starship has a credible claim of more reuse. If they were indeed able to fully reuse stage 1 and 2 then maintenance would probably be sub-$1m based on Falcon 9 reuse maintenance. Nitrogen is another $1m/launch. So the whole thing swings on full reuse. If Falcon 9 was full reuse they would already be at ~$200/kg, and the whole point of Starship is the benefits of scaling up by 8x while getting to full reuse. You could argue that one should fully load the $15b development costs of Starship into the launch cost/kg. But at 1000s of launches/year even that doesn't prove to be the major swing factor - and that is also a sunk cost. The labor involved in production is factored into the amortization, and the other costs like G&A aren't big enough to matter at 1000s of launches per year. They will continue to spend a ton on R&D, but one shouldn't burden the Starship launch costs with that expense because it is focused on future projects (or pushing down the cost of Starship). Bottom-line is that I think there is good reason to believe they get to full reuse so I am optimistic they reach sub $100/kg.
You said, "The theory of exhaustible resources suggests that “the real price of an exhaustible resource should grow at a rate equal to the real interest rate.” The data are consistent with this hypothesis."
The article you linked to to support that says, "Of these 93 mineral commodities, 60 of them got cheaper between the beginning and end of their time series." That is in terms of inflation-adjusted dollars, over the last 120 years or so. Even if the real interest rate averaged a modest 1.5% over that time frame, then the price should have gone up by a factor or ~6, not mostly gone down.
It seems like the entire premise of this piece is just wrong.
If you look at the final chart of that post, it shows the annual rate of real price change in last 20 years. Looking at the exhaustible resources (minerals, fuels) the modal rate is 1.5-2%. Compare with real interest rates ranging from 0.5-2% over the same period:
Remember, my argument depends on mining technology picking the low hanging fruit and seeing progress halt. Now that the petrochemical and shale revolutions are complete, we should expect a return to the regime of exhaustible resource economics. And the recent data seems consistent with that.
Surprisingly, even the long term data show a median growth rate around zero despite a century of technological change. For natural gas in particular (the most important resource for the purposes of this post) Potter concludes that "[n]atural gas and oil got cheaper until the 1950s and the 1970s, respectively, and since then have gotten more expensive."
Given this, I feel okay assuming the price of natural gas will not fall dramatically in the future.
I think your cherry-picking the data. The second-to-last chart, which covers a lot more 20 year periods, shows average rate around 0%. More relevantly, the link you gave to Henry Hub clearly shows the price of natural gas declining precipitously over the last 20 years. Would your model have predicted that in 2006? Why would you think technology would just all of a sudden stop progressing in this moment in time in particular?
I would bet against natural gas prices dropping dramatically in the future, but that is with very low confidence
The Henry Hub data looks flat to me with various spikes and a period of high prices 2003-2008. But if we're worried about cherry picking, I don't think looking at one resource over one specific 20 year period is a good idea!
Let's separate two claims:
1. Real natural gas prices are expected to be about the same in the future.
2. Exhaustible resources rise in price at approximately the real interest rate assuming no technological improvements in extraction.
I think we agree on 1, which is the basis for some of my arguments in this post. If you prefer longer term data, the longer term data on natural gas and mineral commodities bear this out. See for example the data on mineral commodities in Brian Potter's post. Many of them look flat with data going back over a century.
Given long term data among a variety of exhaustible resources showing flat prices in addition to flat prices for natural gas, I'm fine with #1 as a load-bearing assumption for parts of this post.
As for 2, I see it as a lot more uncertain. The theory and the data seem to agree in the modern era (which is when the technological improvements in extraction have slowed). But that could change. If prices turn out to be flat long term like we saw in the 1900's, we'd need a new theory.
I don't get the passenger rail scepticism. There are reasons it might not be effective in specific countries given geography, culture etc. (e.g. in the States) but it clearly works and is highly utilised the world over. It's most effective for travelling medium-sized distances between towns, where it's quicker and more efficient than either a plane (no security and lengthy boarding times), car or bus.
I'll need to look deeper into why Glaeser thinks trains are less cost effective. If trains turn out to be cheaper per passenger mile that would change my mind!
In terms of speed, I would think cars would be faster (since there's no waiting), then trains, then busses as the slowest?
It depends on the trains and the cities, and where you want to go on either side of the station. But in general, driving though traffic is often slower than trains which can go faster and more directly. Let's consider two typical journeys:
Reading to London in the UK is 23 minutes on the train, but nearly two hours to the same point in a car. London is slow and stressful to drive into, the trains are quick and regular. There's no flight.
Tokyo to Kyoto is over five hours in a car and only about two hours twenty minutes on a train. It's only an hour on a direct flight (2.5 with a change) but when you factor in security, baggage etc., there's not really a time saving.
(Obviously you're right that there will be plenty of journeys where trains are much slower than a car, when there's no direct line or your starting or end point isn't near a station.)
Cost is not the only metric to compete in. Consider a train ride in Tokyo versus Baltimore? Safety and flexibility are other important dimensions.
Sure, walking and cycling are always going to be cheapest form of transport per passenger mile— not necessarily the most convenient depending on distance!
Loved to read this.
The points about fusion are interesting. I think youre likely right, though I think the technology will still be important for settling the solar system (eventually) and I really hope people built it regardless. Solar power is not very efficient farther out. Though there may be other good options.
With home robots I disagree. The extravagant 20k-100k home robot that can cook a five star meal and walk your dog might be far off, but I think the math will work relatively soon for plenty of applications. A dishwasher runs once a day for an hour or two and costs a few hundred to a thousand dollars. Easy to imagine a similar price point working for “tidy up my room” or “fold my clothes." 8k for a Weave clothes-folding bot is a bit steep for many people but I think plenty others will buy it.
That's true about fusion, much more useful in the outer solar system and on long interstellar flights.
On home robots, fair enough! I'd love to be wrong here and overall I'm quite bullish on robots automating ~everything. Perhaps people with a higher willingness to pay will drive adoption and scaling. I do wonder if the price of robots stays high relative to wages if people will rent them.
Re: supersonics: what’s the actual fuel burn per passenger-mile supersonic vs. subsonic? The specific impulse graph implies it’s about 3x fuel burn for a given thrust, but that doesn’t really translate. On the one hand, if it was 3x burn for 3x thrust for 3x speed, that would be equal efficiency to current airliners. But Supersonic has MUCH higher drag, and carrying more fuel means less payload. I’d like to see more detail around that. Great post though!
Yeah I'm not familiar with the fuel burn numbers here. I know the drag spikes in the transonic regime, so you either want to fly subsonic or past Mach 1.5, not in between.
https://en.wikipedia.org/wiki/Drag_(physics)#Wave_drag_in_transonic_and_supersonic_flow
Supersonic will always burn more fuel of course, so main question is whether that 2x-3x time saving is worth the higher fuel and design costs.
Ah, this got me to look into it more. Thrust specific fuel consumption (kg of fuel consumed per second per unit of thrust)is inversely proportional to specific impulse:
https://en.wikipedia.org/wiki/Thrust-specific_fuel_consumption
Looking at the table on that page looks like the SR-71 consumes about 4x more than modern engines. But given that the speed is ~4x higher the total fuel consumed for a given distance is about the same. But, lift-to-drag is worse for supersonics. This page says the Concorde had L/D of 7 whereas a 747 is 17.
https://en.wikipedia.org/wiki/Lift-to-drag_ratio
So ~2.5x more fuel consumption overall if I'm understanding correctly
Great article, so many great numbers. I have to comment that a hugely under appreciated aspect of space data centres is reduced latency. Large compute and storage to relay data between Earth, the Moon, and Mars solves blindspots in communication. Qualitative rather than quantitative risk.
Also, I'm a little worried, looking at your starship/methane costs that, at scale, launching objects into orbit will have cheaper energy costs than UK heating! Another form of shrinkflation?!
Good catch! Two reasons my methane price is optimistic. First, I'm using methane prices in the US. The US leads the world in fracking technology so has some of the cheapest natural gas on Earth. Two, infrastructure to transport natural gas to consumers adds a lot to the cost, that's why consumers pay over $10/MCF in the US:
https://www.eia.gov/dnav/ng/ng_pri_sum_a_EPG0_PCS_DMcf_a.htm
It makes complete sense that, over time, cost of kg to orbit drops to $5.5 of fuel. The real light bulb for me was pricing joule-per-kg to orbit (I assume a constant?) and engaging in arbitrate against the dollar price of fuels through launch sites, regulatory environments and supply chains. That would be the market dynamic of "the launch business". (I might be late to the party on that realisation...).
But then there is an implication: if the cost of anything in orbit is cheaper than the cost of anything on earth by more than the launch cost, it should go into orbit. Huge demand.
I had never realised that cost could be as low as heating a house. So, now we have competition for energy between launch and heating. The cost of energy really is the base driver for everything
Fun read!
One quibble. In discussing cooling costs in space, doesn't it matter that the ambient temp in space is super low to begin with? That's a lot of diffusive force.
Low ambient temperature in space is super important! It's included in the radiator calculation by assuming the temperature is 2.7 kelvin (about the temperature of the cosmic microwave background, near absolute zero).
This is why reversible computing is more feasible in some sense in space. It is naturally cold there, so maintaining a cold computer is easier relative to Earth.
I spend a lot of time doing technology diligence professionally, and then on top of that trying to codify my intuition into formal models for what is easy to do and what is hard to do. My quick take: This is all correct. Also, way more entertaining than formal models. Nice work.
Thank you for the kind words! And that sounds like a really cool job!
It's not really a job, it's part of a consulting portfolio that is really lots of jobs.
Also, yeah, it is the coolest job in the world.
For materials, titanium is pretty abudent and if it was easier to refine and thus cheaper, it would have plenty of uses infrastructure and consumer products.
You may be interested in Orca's efforts to make titanium cheaper! It looks hard but I hope it works out:
https://www.orcasciences.com/articles/there-has-to-be-a-better-way-to-make-titanium
Nice analysis. This is nice work. I think for solar and compute you need to include marginal land costs and consider the alternative uses. That is the major missing cost that I think might be material to your work. For solar near good transmission connections, the land is finite and may have other good uses. Same for data centers. Many of the alternative sources have smaller land footprints so it may well be a point of competition. So tell me why you leave out marginal land footprint costs for solar? Using your cost theory, the current land for solar is the cheapest it will probably be for any future installs.
While it isn't explicitly included in the solar cost calculation, the idiot index is intended to account for all sorts of hard-to-predict costs. My understanding is that land costs in particular is a relatively small share of utility scale project costs.
Outside of cities there is a lot of land and it's surprisingly cheap!
Current price equilibrium is not likely a great proxy for future equilibriums. The current solar build is likely 0.05-0.10X US generation.
Your theory of exhaustible resources would not classify this an “unpredictable cost”. This is a predictable cost. Look at the footprint needed to 2X US generation, assuming grid ready, transmission adjacent land costs stay the same in the face of steady increased demand seems unlikely. Also, consider where solar ready land is - Canada, New England, Minnesota??? Appalachia, not very flat. Hmmm. Geographically concentrated demand for an exhaustible resource is not “unpredictable costs” and probably not a linear increase either.
This is without permitting and community frictions in the face of a large footprint. You should consider that grid ready land may become the dominant cost for solar.
Tyler’s blog is called “Marginal” Revolution for a reason.
Can solar scale to double or triple current US gen with land costs constant is a valid question? Solar can do a lot but it's not a category killer either.
Exhaustible resources more expensive long-term
The theory of exhaustible resources suggests that “the real price of an exhaustible resource should grow at a rate equal to the real interest rate.” The data are consistent with this hypothesis1.
Innovations in search and extraction can dramatically lower prices in the short term, but in the long term, we converge on a correct way to do things. Once extraction is solved, prices return to their slow upward climb.
Terrestrial land for solar is an exhaustible resource. Your installed solar cost projection is way off.
Congrats on the Marginal Revolution feature
I largely agree, but you glossed over the #1 question I have: new materials for superconductors? Please say more on this!
Superconductors are neat, particularly as a new electronic component for computers. That said, I'm skeptical of superconducting computers or other unconventional computing replacing our current computing roadmap:
https://splittinginfinity.substack.com/p/semiconductors-will-see-an-end-of
The most exciting application for superconductors in my mind is reversible computers. And in that case, we already have superconducting materials with good properties! Though some tweaks will be needed. These computers need to be cold for accurate computation, so high-temperature superconductors aren't a pressing concern.
Other applications include power dense motors for flying cars/planes and perhaps better MRI machines. Broadly, I see superconductors as a nice boost for motors, plasma control, turbines, electronics but not an Earth-shattering unlock.
Agreed on launch costs and space data centers. Some commentary on the first of the SDC ideas last year: https://angadh.com/space-data-centers-1
For what it's worth, solar module pricing is already far below $0.40/W in China, closer to $0.10/W is my understanding. It's 3x-4x as expensive in the US because of tariffs :(
Whoa, $0.1/W is kinda insane. But then again OWID has modules listed at $0.26/W in 2024
https://ourworldindata.org/grapher/solar-pv-prices
At this rate, the racks and labor are the real bottleneck.
Back when cells were over $2/watt, panels were about a third of the total cost. Finance cost, labour, civil works, structures, wiring, inverters, control systems, grid interconnect, land acquistion, design and permitting were most of the rest of the capital cost. Maintenance is not negligible too.
Sam is being far too generous with his theoretical minimum.
I guess that’s what the “idiot index” is supposed to cover. Agreed that labor/install/permitting/overhead brings actual install costs for utility scale solar ~$1.20/Wdc
I think the Austin Vernon posts I've linked are relevant here! If solar modules can fall in price with scale I don't see why other components like racking and wiring can't also fall in price. Erthos is deleting a lot of these components by putting solar on the ground and pre assembling wiring. Jurchen PEG claims one project reached all-in cost of $0.71/W including modules, that's not far off from the goal of $0.4/W!
Racking is pretty much at minimum cost of steel + galvanising + welding. Etc., mutatis mutandis. Ground works at low latitudes, not at high. Agrivoltaics with N-S vertical panels may be better there. The engineers working in the trade are not dummies.
The way to cut costs is panel efficiency. Halve the number of panels required, halve the land, labour, structures, wiring. Does not affect design, permitting, legal, inverters or interconnect though. Doesn't even affect civil all that much.
I wrote a response, but it's too long for a comment, so here's an article.
https://donald532703.substack.com/p/a-response-to-splitting-infinity
Thank you for this! I haven't given it a full read yet but I generally agree with what I've seen.
Strange you did not mention geo-thermal. That and nuclear is the future, intermittent, non-dispatchable tech like wind / solar are not.
Oh, so those are addressed in the "Non-solar energy" section. I don't think they have a prospect of beating solar + batteries for electricity. Both require heating a working fluid and spinning a turbine with the attendant energy losses and equipment costs.
Instead, these are a better fit for providing industrial heat/steam.
Did not catch that remark, thanks for the pointer. Nonetheless, I still think you have way too much confidence in batteries, at least unless / until they get to 10x density.
Would love some comments about the various Doomberg articles detailing why solar / wind / batteries is nowhere near good enough currently.
While energy density is certainly a problem for using batteries in transport, for grid-storage, the space that the batteries take up isn't as much of a problem. Particularly on a site that already has acres of solar panels. In these applications, cost per kWh of storage is more important (at the pack level of course, so including wiring, cooling, controllers, etc.)
From what I remember of Doomberg's articles, they seemed like an okay if pessimistic assessment of costs today. I'll admit, solar modules and batteries are not the dominant cost any more. But I don't see why we can't push down these other cost factors.
Another point of confusion is looking at solar costs under the requirement of 99+% reliability. Solar and batteries can't do that yet. That's why I advocate for a 90% clean energy system using natural gas for the last 10%. It can be 10x greener while being slightly cheaper. Then we can wait for other solutions to push the renewables percent higher.
https://splittinginfinity.substack.com/p/net-zero-part-1-energy
For off grid applications willing to adapt to intermittency, much lower prices are achievable. The prices in the non-solar section were for 97% uptime. For this use case, $20/MWh seems possible. Solar and battery prices are close to where they need to be for this goal, next step is to automate and cut red tape.
Your LEO launch cost analysis has a few flaws. First, the table used from Orca Sciences defines idiot index as "cost of ore plus energy of reduction." The energy of reduction is the thermodynamic minimum. They are comparing the real-world energy required relative to the theoretical minimum, with the point being that "in a world with cheaper clean electricity...we should expect more use of AI and Mg, and much less use of steel." That table doesn't illustrate the ratio of "other costs" to raw materials. Second, Musk has said he thinks fuel costs for Starship could be as low as $500k and all in operational costs per launch at $2m ($13/kg) - an idiot ratio of 4 as you are defining it. Also, air travel requires much more labor so one could argue the ratio should be lower. The $10m figure you cited is a near term goal, not an end state. Third, your fuel prices don't incorporate the sort of bulk deals they are doing with Linde nor the probable end state of vertically integrating, which can get methane down to ~$0.05/kg and LOX to $0.08/kg. And finally the comparison to Falcon Heavy might be a bit unfair as well, as the whole thesis on Starship is that rapid full reuse of a much bigger payload is what will drive costs/kg down. Maybe rapid full reuse won't prove to be possible, it is no doubt an ambitious goal, but if they were to reach that point then cost/kg would likely fall well below $100/kg.
Hi, thank you for the comments!
> Musk has said he thinks fuel costs for Starship could be as low as $500k and all in operational costs per launch at $2m ($13/kg)
While I appreciate Musk's ambitious goals, I don't see a reason to treat his statements as good forecasts. He has a track record of failing to meet highly ambitious timelines for example. Part of point of this exercise is to establish a viewpoint independent of figures like Musk who have incentives to mis-report expected costs!
> Also, air travel requires much more labor so one could argue the ratio should be lower.
It's not clear to me why flying a plane would require much more labor than launching a rocket. The rocket/launchpad/payload requires more engineering staff to design, build, test, fuel, and repair. It also has a shorter lifetime than a plane, meaning more amortized labor hours per flight. Aviation is an established technology, rocket science is still being developed. We should expect rockets to have a higher index.
> our fuel prices don't incorporate the sort of bulk deals they are doing with Linde nor the probable end state of vertically integrating ...
Why doesn't SpaceX (or any other industry) already enjoy these super cheap prices? They’ve already expended hundreds of thousands of tons of fuel. Why haven't they pushed costs down further at such scale? Remember, my estimate for their fuel costs is already 2-4x lower than their current fuel costs. They likely aren’t leaving money on the table here, particularly with their upcoming IPO.
For that matter, if 3x lower natural gas prices are achievable, why aren’t fracking companies doing this today? It’s a competitive market and a company with much lower extraction costs would be able to enjoy large profits at today’s prices. Despite innovations in fracking and the shale revolution, US spot prices for natural gas have rarely dipped below $2/MCF in the last 30 years. $0.05/kg requires prices below $1/MCF and zero purification/liquefaction/transport costs. This seems unlikely, particularly given recent geopolitical developments.
https://www.eia.gov/dnav/ng/hist/rngwhhdm.htm
Fair points but I think the primary way one gets to a better estimate of $/kg to LEO is to look at the actual swing factors rather than apply idiot index ratios. The launch operation center for Falcon 9 is less than 300 people, so the direct labor per launch is pretty minimal at 1000 launches/year (<$100k/launch). Whether the fuel is $500k/launch or $2m ($20/kg) isn't the major swing factor either. I would argue the major swing factor is 1) the number of fully reusable launches per Starship. At 20, the amortized cost of the build would be $5m/launch. At the goal of 100 it would be $1m/launch ($10/kg.) Falcon 9 are up around 25-40 now and given the more durable construction Starship has a credible claim of more reuse. If they were indeed able to fully reuse stage 1 and 2 then maintenance would probably be sub-$1m based on Falcon 9 reuse maintenance. Nitrogen is another $1m/launch. So the whole thing swings on full reuse. If Falcon 9 was full reuse they would already be at ~$200/kg, and the whole point of Starship is the benefits of scaling up by 8x while getting to full reuse. You could argue that one should fully load the $15b development costs of Starship into the launch cost/kg. But at 1000s of launches/year even that doesn't prove to be the major swing factor - and that is also a sunk cost. The labor involved in production is factored into the amortization, and the other costs like G&A aren't big enough to matter at 1000s of launches per year. They will continue to spend a ton on R&D, but one shouldn't burden the Starship launch costs with that expense because it is focused on future projects (or pushing down the cost of Starship). Bottom-line is that I think there is good reason to believe they get to full reuse so I am optimistic they reach sub $100/kg.
You said, "The theory of exhaustible resources suggests that “the real price of an exhaustible resource should grow at a rate equal to the real interest rate.” The data are consistent with this hypothesis."
The article you linked to to support that says, "Of these 93 mineral commodities, 60 of them got cheaper between the beginning and end of their time series." That is in terms of inflation-adjusted dollars, over the last 120 years or so. Even if the real interest rate averaged a modest 1.5% over that time frame, then the price should have gone up by a factor or ~6, not mostly gone down.
It seems like the entire premise of this piece is just wrong.
If you look at the final chart of that post, it shows the annual rate of real price change in last 20 years. Looking at the exhaustible resources (minerals, fuels) the modal rate is 1.5-2%. Compare with real interest rates ranging from 0.5-2% over the same period:
https://fred.stlouisfed.org/series/REAINTRATREARAT10Y
Remember, my argument depends on mining technology picking the low hanging fruit and seeing progress halt. Now that the petrochemical and shale revolutions are complete, we should expect a return to the regime of exhaustible resource economics. And the recent data seems consistent with that.
Surprisingly, even the long term data show a median growth rate around zero despite a century of technological change. For natural gas in particular (the most important resource for the purposes of this post) Potter concludes that "[n]atural gas and oil got cheaper until the 1950s and the 1970s, respectively, and since then have gotten more expensive."
Given this, I feel okay assuming the price of natural gas will not fall dramatically in the future.
I think your cherry-picking the data. The second-to-last chart, which covers a lot more 20 year periods, shows average rate around 0%. More relevantly, the link you gave to Henry Hub clearly shows the price of natural gas declining precipitously over the last 20 years. Would your model have predicted that in 2006? Why would you think technology would just all of a sudden stop progressing in this moment in time in particular?
I would bet against natural gas prices dropping dramatically in the future, but that is with very low confidence
The Henry Hub data looks flat to me with various spikes and a period of high prices 2003-2008. But if we're worried about cherry picking, I don't think looking at one resource over one specific 20 year period is a good idea!
Let's separate two claims:
1. Real natural gas prices are expected to be about the same in the future.
2. Exhaustible resources rise in price at approximately the real interest rate assuming no technological improvements in extraction.
I think we agree on 1, which is the basis for some of my arguments in this post. If you prefer longer term data, the longer term data on natural gas and mineral commodities bear this out. See for example the data on mineral commodities in Brian Potter's post. Many of them look flat with data going back over a century.
Given long term data among a variety of exhaustible resources showing flat prices in addition to flat prices for natural gas, I'm fine with #1 as a load-bearing assumption for parts of this post.
As for 2, I see it as a lot more uncertain. The theory and the data seem to agree in the modern era (which is when the technological improvements in extraction have slowed). But that could change. If prices turn out to be flat long term like we saw in the 1900's, we'd need a new theory.