So, Sam Altman wants as much money as he can get—the number he chose to give was $7 trillion—to build more chips for AI as fast as possible. Including in places like the UAE. Thus undermining his argument that OpenAI is helping with AI safety by preventing overhang and maintaining a lead for America, which was never a good argument anyway.
Well, I warned some people about him. To avoid getting fooled by bad actors, you have to avoid getting fooled by good actors. (In Altman’s case, I had the advantage of more information about him making conflicting statements to different people.)
Anyway, this got me thinking about how I’d spend $7 trillion dollars. Such large numbers get hard for people to understand; you can say it’s about $1000 per person alive, but giving it away equally isn’t in the spirit of the challenge here. No, the point here is imagining megaprojects and massive economies of scale, to grapple a little bit with the enormity of trillions of dollars.
So, the below is how I’d spend that kind of money in useful ways. Much of this is stuff that’ll happen anyway, and listing that is sort of cheating, so I guess I’ll aim for over $7 trillion, and when I know someone with a notable possible improvement to something, I’ll mark that with a *.
solar
Doing work requires energy, and the modern trend is towards electrification. The world is now generating an average of ~3.5 terawatts of electricity. Making 2 terawatts of average solar power generation seems pretty reasonable. With single-axis tracking you can get maybe 30% capacity factor, so that’d be 6.7 TW of solar panels. Anyway, at least $1 trillion of capital investment in solar panel production seems justified, and that’s a good start on spending all that money.
If you just want a big enough production complex to hit the limits of cost scaling, that’s a lot cheaper, probably a mere $100 billion. The “Inflation Reduction Act” probably would’ve made more sense if it funded a big solar panel complex instead of stuff like subsidies for water electrolysis (which is a dead end for at least a few decades) and rooftop solar (which is expensive and dumb).
PV panel production includes:
making silicon metal
making SiCl4 and distilling it for purification
reaction with hydrogen
slicing thin layers with diamond-studded wire saws
treatments including doping
application of ITO transparent conductor* and wires
The vast majority of solar panel production is done in China. It doesn’t involve a complex supply chain like electronics production in Shenzhen. It’s highly automated and shouldn’t depend on low labor costs. Why, then, is it cheaper to make solar panels in China than America? My understanding is, that’s because building the facilities is more expensive in America. The machines used are about the same prices, so the difference comes from things like land, regulations, building construction, welding pipes, and management costs.
Why is that? Why is making manufacturing facilities more expensive in America? Maybe because everything is more expensive in America! If you have a choice, if you aren’t bound by the locations of real estate or natural resources or funding sources or existing equipment, it seems like it’s only worth doing something in America if the cost of doing it doesn’t matter much.
In other words, I think the problem with making (exportable) stuff in the USA is largely currency values. The PPP/nominal GDP of Japan is 1.5x that of the US, Poland is 2x, and can you really justify building a factory in the US if you get 1⁄2 as much factory for your money? Similarly, it’s a lot cheaper to get surgery in Mexico instead of the USA, or go for a long vacation in Thailand.
grid storage
Sunlight is inconsistent. So, having generated electricity with solar panels, it may need to be stored until it’s needed.
Grid energy storage is more expensive than adjusting the timing of hydropower or burning natural gas. It’s “too expensive”—but that just means it’s more expensive than current generation, and people in California are paying literally 10x the cost of generation for electricity, 3x what people in other states pay. If electricity generation costs were really as important as some bloggers seem to think, California would be abandoned.
Reasonable people talk about reducing global warming and improving energy security in terms of minimizing mitigation costs and maximizing what can be done with how much people are willing to pay, but a lot of investors and governments seem to want to hear “this is cheaper AND it reduces CO2 AND it improves energy security” and will invest in delusional people who tell them that. It’s illogical: adding extra constraints has costs.
People drawing straight lines to project future Li-ion battery costs were dumb. The best ways to do grid energy storage are:
compressed air in underground caverns*
chelate flow batteries*
power-tower solar-thermal* with integrated heat storage and possibly underground compressed air storage
For compressed air energy storage, water-compensated systems seem worthwhile.
Flow batteries are currently too expensive; cheaper membranes* would be needed to make them competitive, but those are possible. Very-large-scale production of the organic ligands* would also be needed.
Solar-thermal would only be practical in particularly sunny areas with electricity demand. Better designs than existing plants are necessary.
I could see $4 trillion of such energy storage being justified worldwide, but that depends on things like CO2 valuations, natural gas prices, and politics.
HVDC
Having supplied electricity at the right time, it needs to go to the right places. The best way to move electricity long distances is with HVDC lines. That requires:
high-voltage transformers
AC-DC conversion*
land rights
aluminum wires
At least $200 billion of investment worldwide on HVDC conversion and lines is probably justified. But there are some points to note:
Getting land rights can be difficult, especially in the USA.
Fuel pipelines are a more-efficient and cheaper way to move energy than HVDC lines.
HVDC transmission alone isn’t a good solution to inconsistency of solar power.
Here’s an influential paper concluding that:
US power consumers could save an estimated US$47.2 billion annually with a national electrical power system versus a regionally divided one (~1.1¢/kWh). This amounts to almost three times the cost of the HVDC transmission per year.
Here’s further analysis, concluding that $400B of investment in HVDC transmission and $2 trillion in new generation is justified.
batteries and supercapacitors
Having supplied electricity at the right time and place, it may need to be supplied to something moving with batteries, or provided as pulses from supercapacitors.
The battery market is now dominated by Li-ion batteries*. There are other potential battery chemistries*, but most of the ones I see news articles about are obviously not competitive. (Like with stories about potential “cures for cancer”, most people eventually stopped believing them.) Anyway, at least $2 trillion of capital investment in battery production seems justified.
Flywheel systems used to be cheaper than supercapacitors*, but costs are now similar, and I suspect supercapacitors will end up outcompeting them. For mobile robots in factories & warehouses that can be recharged frequently, supercapacitors are competitive with Li-ion batteries; both options work fairly well.
motors
Having supplied electricity to the right place, it must be converted to something else, usually mechanical power. Economic growth and replacement of engines with electric motors means more motor power per person will be needed.
At least $1 trillion of investment in electric motor production is probably justified. I expect more usage of axial-flux permanent magnet motors* in high-performance applications and more switched reluctance motors* in cost-sensitive applications. Both would need grain-oriented electrical steel*.
power electronics
To drive electric motors, electricity must be converted to the right voltage and frequency. Some motors are directly connected to an AC grid, but most applications require variable speed which requires power electronics.
Motor driver circuits* typically use semiconductor switches on special circuit boards with thick copper. They used to use silicon MOSFETs and IGBTs, but electric car makers are moving to SiC switches. I think GaN* and/or SiC will mostly displace silicon for motor drivers. Research on production GaN crystals using physical vapor deposition, ammonothermal methods, and sodium flux methods is ongoing.
Like motor production, at least $1 trillion of investment in production of power electronics is probably justified.
copper
Transformers and motors need copper. The world is currently producing >$150 billion of copper a year, and continuous investment in new mining projects is needed.
actuators
Having produced mechanical power with electric motors, it needs to be converted to the desired forms. Most applications require higher forces at lower speeds than electric motors provide. The basic types of actuators are gears* and linear actuators*, and both are needed. I tend to further divide each into these subtypes:
high-force (for heavy equipment / presses / etc)
precision (for robotic arms / etc)
miniature
Notable current actuator types include:
planetary gears
cycloidal drives
strain wave gears
belt and chain drives
ball screws
roller screws
hydraulic pumps and cylinders
The relative cost of actuators and electric motors varies greatly, depending on speed & size & precision.
A world with more electromechanical actuators per capita is also a world that needs more roller bearings. Well, the CAGR of roller bearings is expected to be higher than GDP growth. Current bearings work well and I don’t expect much about them to change, but maybe we’ll see more ceramic bearings in the future...? I think spark sintering makes silicon nitride bearings a bit cheaper...or maybe Sialon-TiN microwave sintering will outcompete that? That’s not my area of expertise so I can’t say; what I do know is that not having to replace stuff because ball bearings wore out would be nice.
How big a market is this? Well, the global market for robotic arms is currently ~$30 billion/year, which seems small relative to their economic significance, but if having a robotic arm in homes becomes common, that’d probably be more than $1 trillion dollars of actuators. And if some superintelligent AI takes over and replaces humans with robots, that’d be an even bigger market!
Then there’s heavy equipment; I expect future excavators to use electric motor driven actuators, and you can look at the global sales of hydraulic equipment, which is currently larger than but comparable in scale to the robotic arm market. (By the way, both of those together are currently smaller than the video game market.)
biomass
Liquid fuels have higher specific energy than batteries, and that’s necessary for some applications. Also, refuelling is faster than recharging, and liquid fuels can provide long-term energy storage. Making liquid fuels from biomass is much cheaper than trying to make them from electricity. It’s potentially cheap enough for its CO2 mitigation cost to be among the lowest. Also, there isn’t enough arable land to produce all the energy currently used by civilization from biomass.
Converting biomass to higher-value chemicals is better than making fuels. The processing of biomass to chemicals* generally costs more than the biomass used, and only low-cost processing is practical. (Sometimes I see an article about people using plasma or something to process biomass or waste, and...sure, you can do that, but it’s far too expensive.) Heating up biomass in water can produce a mix of:
furfural (can be used for furfurylated wood & bamboo)
levulinic acid (usage* would be complicated)
hydrochar (similar to coal, can be burned or buried)
With a good process for such conversion, and development of good uses for furfural and levulinic acid, growing over a billion dry tons of biomass worldwide for that a year seems plausible. (In general, I like specially-bred Miscanthus sinensis as a choice for on-purpose biomass production for conversion to chemicals.) In that case, at least $200 billion of capital investment in such biomass conversion plants seems justified. That would involve many medium-size plants to reduce transportation costs, rather than a few large plants.
methanol
Currently, over 100 million tons of methanol is produced per year; prices vary but $300/ton is typical. The main uses are formaldehyde (for urea-formaldehyde resins) and gasoline additives.
Rather than those uses, what I imagine expanded methanol production being used for is: dimethyl ether as a fuel for diesel engines, and fermentation to chemical intermediates. Dimethyl ether works with existing diesel engines, but has lower energy density and requires pressurized tanks. Useful fermentation of methanol would require better engineering of microbes for synthetic methylotrophy*.
If conversion of diesel vehicles to use dimethyl ether (to reduce air pollution) is pursued, OR suitable engineered microbes are developed, at least $100 billion of capital investment in methanol production seems justified.
plastics
With all the backlash against plastics, global production of plastics has...continued increasing rapidly, of course. (Bans on plastic straws and plastic bags aren’t significant, and they’re completely unnecessary since biodegradable plastics work fine for those uses at negligible extra cost.)
I generally divide future commodity polymers into these categories:
strong thermoplastics*
biodegradable thermoplastics*
resins*
thermoplastic elastomers*
strong polymer fibers*
Overall, at least $500 billion of capital investment in plants for making polymers (and their monomers) seems justified.
An example of a future commodity plastic would be, hmm...(biphenyl-4,4′-dicarboxylic acid + ethylene glycol + butanediol) which can give bulk thermoplastics with >250 MPa tensile strength as well as strong melt-spun fibers. I’ve seen a few routes* to that, some cheap enough to justify enough production for that to justify $500 billion of capital investment by itself, but I’m not sure if companies will figure them out.
By the way, some uses of polymers I expect to grow relatively fast include:
poly(lactic acid) modified by chain extenders* for biodegradable plastics
EconCore type panels
specialized resins for 3d printing for lost-resin casting*
a new city
Supposing there’s more production of stuff like plastics and furfurylated wood, maybe some of it should be used for construction.
Some people think America should pack millions more people into New York City and San Francisco, but I’m not convinced that their high average wages mean that marginal wages of additional people would be high. Personally, I think it makes more sense to make a new city somewhere.
Ideally, you want some place with:
a decent climate
flat land for building
a possible seaport within a few hours of driving
some existing cities within a few hours of driving
Maybe some place south of Raleigh NC would be reasonable?
Supposing you want at least $100k of buildings per person and 1 million residents, that’s already $100 billion, and I’m sure it’s possible to spend more than that.
America already has cheap housing in dying cities; if building more in a new location, there has to be a reason beyond “more housing”, such as:
better climate
better access to a port or highways
better road layouts* achievable with greenfield projects
reaching some minimum scale
I’ve seen some people say the entire population of Taiwan should be relocated to a special region of the USA; that would certainly be a project that costs trillions of dollars. That doesn’t seem plausible to me, but I could certainly imagine a couple million Taiwanese wanting to emigrate.
investment decisions
Adding up the numbers above...yep, almost certainly over $7 trillion. Mission accomplished.
Of course, this isn’t how resources actually get distributed. China does have a more top-down approach that’s a bit closer to this than what happens in the USA, but that’s still not an integrated analysis. Rather, the Chinese leadership chooses a set of strategic goals, and they distribute those to people, who further divide and distribute goals. For example, the Chinese gov decided it wanted to: be able to take Taiwan → deter US intervention → have better nuclear threats → build more nukes and develop maneuvering hypersonic vehicles → build nuclear plants and a hypersonic wind tunnel like JF-22. It decided AI would be important, and then that turned into a bunch of project branches.
In the US, on the other hand, CEOs and executive directors make most of the investment decisions. They might read magazines saying that “cloud and blockchain and IoT” will be important, and talk to other executives at conferences who agree with that view, and then make a statement saying their company “will be a trend leader for emerging technologies including cloud computing and blockchain”. Then they delegate the technical details to a guy, who hires a consulting firm, who finds someone who social consensus says is probably knowedgeable. A Nobel-winning scientist like Gregg Semenza would obviously be the best possible expert, but that’s not necessary; it’s better to find a professor whose work seems relevant, and hire them and some guy from Guidepoint for a few hours of consultation that Harvard grad employees can compile into a technical report.
The US government is presumably less involved in the economy than China’s, but between the federal + state + local governments, spending is around 1⁄3 the economy. Much of that is payments to individuals and hospitals, but how does something like IRA subsidies for water electrolysis happen? From the point of view of the non-technical leadership, there’s an overall consensus that electricity to hydrogen is viable in the near future. That implies that it just needs tech development. Legislators make deals and work out how much they want to spend on different things, and then congressional staff are supposed to work out good ways to use some amount of money with some restrictions. (Those congressional staff studied political science and their job experience consists of being a congressional intern.) So they ask their economics experts about how to efficiently encourage the relevant tech development, and they decide to temporarily subsidize different routes to hydrogen at slightly below their current extra costs. (I understand, but still, for me it’s like watching video of a toddler trying to stick things in outlets.) And then, since there’s no immediately-visible solution to the high costs, companies try to find loopholes instead of spending on research.
We’d all like 7T for projects we like. The difference in Altman’s ask is that investors would likely make their money back and more. That might be the case with some of your projects, but that’s the key thing to focus on besides wishing for wealth.
...no?
I’d expect an Altman-led AI chip maker to be worse than other companies and get outcompeted.
If $7 trillion of investment isn’t enough for oversupply or at least commoditization of chips for AI, then that’s a world in which property rights of AI chips are no longer guaranteed.
In any case, this post isn’t even evaluating investments by profitability, it’s evaluating them by societal consumer surplus. Those are very different: 1 involves incremental value from current production, while the other involves marginal value from future production with the investment.
Right—but that’s totally unrelated to the logic of the $7T ask. Which makes this an odd post. What is the point of speculating about the good that could be done with $7T that magically appears rather than speculating about how money could actually be garnered for good causes?
As an aside, Altman didn’t ask for 7T for one business he would run—I think he said it would take that much to create an alternate chip supply chain in total. See this excellent breakdown of the chip supply chain. There is a lot of money to be made in chips. It might or might not make back investors money.
Interesting point about the property rights. Of course I agree that we’ll probably hit RSI AGI before that money is made back and that things will get weird or stop
Edit: I did some more reading, you’re right that $7T is way more than you could reasonably spend on chip manufacturer. So I don’t and I’m not sure anyone knows what the $7T refers to, but I think it’s probably all capitol investment intended to turn a profit, not money from God.
Funny you should say that . . .
https://retractionwatch.com/2023/10/02/nobel-prize-winner-gregg-semenza-tallies-tenth-retraction/
https://www.nature.com/articles/d41586-022-03032-9
While he may be a co-winner he’s not exactly a household name of a ‘Nobelist’, and certainly in our sort of online circles he’s increasingly famous for non-Nobel things; OP is being sarcastic:
That make sense. If I were going for sarcasm I would have said Kary Mullis.
I agree with pretty much all of this. If anything I think it somewhat understates the case. We’re going to need a lot more clean power than current electricity demand suggests if and when we make a real effort to reduce fossil fuel consumption in chemical production and transportation, and the latter will necessitate building a whole lot of some form of energy storage whether or not batteries get much cheaper.
I don’t think that’s a good model. Instead of asking economics experts, they ask lobbyists about what’s necessary for the technology to be developed.
The lobbyists from BP and Shell tell them: We are happy to switch from fossil fuels to green energy and to do that we need your help by implementic policies X, Y and Z.
Modelling the way these policies are made in Westerns states or even in China without lobbyists is going to lead to poor understanding of the mechanisms.
You criticize Altman for pushing ahead with dangerous AI tech, but then most of what you’d spend the money on is pushing ahead with tech that isn’t directly dangerous. Sure, that’s better. But it doesn’t solve the issue that we’re headed into an out-of-control future. Where’s the part where we use money to improve the degree to which thoughtful high-integrity people (or prosocial AI successor agents with those traits) are able to steer where this is all going?
(Not saying there are easy answers.)
“This is about how to spend money on AI safety” isn’t the point of the opening of the post. It’s more:
Here’s some stuff I’d talk about anyway (like thinking of the economy in terms of energy flows) and a convenient way to frame it that was in the news!
Wow, we could also spend money on not maximally accelerating AI!
I see a lot of people saying that AI is urgently needed to solve [problem] like global warming, but here is how you solve some problems by solving the problems.
AI is something I’ve thought about a lot, but I think I’ve already posted everything about that that I want to, and people didn’t seem to appreciate this that much.
Thanks for linking it! I think one reason I bounced off this article the first time was that I had pattern matched it from the title with the abundant posts on this platform that mostly distill existing arguments.
One billionth of that would quintuple my net worth. So I’d first spend that on securing an environment for myself in which I could actually think at full force. Then, I’d probably spend on a mix of superalignment research and lowering the human death rate towards zero—unless improved cognitive circumstances allowed me to think of something better.