00:03:05 · Casey Handmer, Ty Morse
[00:03:05] S1: No idea. And it actually... To be really blunt, like, the the the reason you would do this is to provide some of the mass for Starmind for the for the orbital AI constellation. But it's it's hard to imagine, like, what set of circumstances would would make it, like, obviously economically preferable to do that as opposed to building more Starships, more launch pads, more factories on the Earth where everything's, like, thousands of times cheaper and just launching more stuff from the Earth. And I think the constraint might be something like if you're launching 10,000 Starships a day, people will say, hey. Come on. But, like, if they don't, then just keep launching from the Earth. Because, like, Starship will ultimately get launch costs into that constellation down to a few $100 a kilogram, which is not very much compared to the costs of a mass produced high performance GPU. So if you look at, like, the embodied cost of the of the item in orbit, most will be the part that does the thinking, the GPU, and then a bunch of other mass associated with radiators and solar arrays and communication. But even that is not all that much, and almost all those systems are spiritually two dimensional. So you can fold them up or make them really thin and not very heavy. So it'd be interesting to see. [00:04:26] S0: Elon on the Dwarkesh and John Collison podcast basically said that in three years from the date of that podcast, he believed it would be cheaper to launch, like, space compute Yeah. Versus building it on the ground.
00:04:26 · Ty Morse, Casey Handmer
[00:04:26] S0: Elon on the Dwarkesh and John Collison podcast basically said that in three years from the date of that podcast, he believed it would be cheaper to launch, like, space compute Yeah. Versus building it on the ground. [00:04:40] S1: Yeah. It's like one of those, like, monkey boy twitches thing. [00:04:42] S0: Well How does that equation work? [00:04:44] S1: Well, Starship is getting a little cheaper. Well, I mean, me and everyone else have made a little spreadsheet to try and figure this out. Starship will ultimate ultimate make launch cheaper, but also much more launch available. So if you're saying, well, we wanna do a gigawatt of deployment, that's just not really feasible even with Falcon. But with with Starship, that's possible. And then at the same time, I think Elon correctly foresaw that large scale deployments of data centers on the surface of the earth would run into headwinds that would increase their cost. This didn't didn't require, like, an exceptionally accurate crystal ball back in the day because there's just so many marginal constraints that sooner or later, one of them will come to bite. [00:05:22] S0: It kinda seems like the data center thing on Earth, it's like this unicorn where the companies that are building them will almost pay any amount to just have them built, and it seems like there's a whole bunch of other shit that's just getting, like, layered on on top of the data centers.
00:05:37 · Casey Handmer
[00:05:37] S1: Yeah. All the local jurisdictions are saying, well, we'd like our piece of the action. And Exactly. I I think they're in a... What's called a price discovery phase where they will, yeah, shortly find out what the market clearing rate is for the appropriate sized official bribe to make these things happen. I mean, we're actually working on that at at Terraform as well, so it's a neat thing. We we we currently have a 1.8 megawatt solar array that we built earlier this year, And we can... Turns out you can you can run your synthetic fuel plant on it during the day, and you can also hook up the batteries and and then run a continuous... Much smaller but continuous load like a... An AI computing rack or data center rack or something off that at the same time without these two users conflicting. So you basically, like, more than double the economic output of of your fixed solar infrastructure, which in terms of land use is is the dominant contributor. At the end of the day, almost all AI compute over the next, say, ten years will converge... Will kind of... Will will be done with this silliness with natural gas and and nuclear and so on, and it will converge on at massive scale, basically, like a a giant Starlink satellite in space, which needs, you know, a a decent radio antenna and a couple other bits and pieces to work properly in space and a GPU. Or basically the same thing on the ground where you... In addition to... You can basically delete the giant radio transmitter and replace a battery. They put a battery there. So the kind of the... The real question mark there about, like, asymptotic cost is whether or not batteries are cheaper than launch. Right? Because one of them requires a launch, the other one requires a battery to last... To to run at night.
00:07:17 · Ty Morse
[00:07:17] S0: Interesting. And so it's basically, like, batteries plus bureaucracy?
00:07:21 · Casey Handmer
[00:07:21] S1: Yeah. Plus plus land. Land is, like, ridiculously abundant, though. You know? Like, Nevada is, like, what, 65,000,000 acres or something. It's it's absurdly enormous amount of land, and and and a lot of that land is still federal because back in the day, they they couldn't even give it away. Right? It was so economically marginal. And now this, like, absurd technological deus ex Makena has kind of landed in all our laps, where it's a system that's essentially apparently unbounded economically... Economic productivity. So it's, like, way better than farming was in, like, 1860. And the only thing it needs to operate is is is power, which you can get from from the sun. That's it. Basically, like, The United States, like, income taxation and so on basically only occurred in the last century. Prior to that point, the United States government was was much, much smaller and was funded primarily by tariffs. And so we've had this, like, massive tax and spend thing that was largely put in place by Roosevelt in the nineteen thirties, and it's grown and grown and grown. And and you can see in other countries that, you know, they're starting to run out of places to find revenue to fund the government. Right? But if it... If if instead, like, The United States had managed to retain, like, its small government, like, fairly, like, light touch on the economy through to the present day, and they start saying, oh, we're the feds, and we need a source of revenue to run our government. The feds will be getting into building data centers on, like, federal land in Nevada and then leasing it out to the highest bidder because the revenue available is just absurd. So it's kind of a funny thing that we're in this situation now.
00:08:52 · Ty Morse, Casey Handmer
[00:08:52] S0: It kind of seems like over time, large scale projects are almost becoming sort of illegal. [00:08:58] S1: Yes. Very much so in the last fifty years. [00:09:00] S0: Do you wanna go into that?