00:19:09 · Casey Handmer, Ty Morse
You know, this is where Kelly Johnson built all those airplanes, and he was able to go out and get the landing gear made and and so on, like the specialists who do just those things that weren't within the plant itself. They gotta rebuild it, or we just won't have wealth. The Australian economy made this mistake. Right? The Australian economy lost its middle, and now it's just tertiary services, which is extremely low leverage because it's just one on one on one stuff. Right? And primary production, which is just digging up dirt and shipping it to Asia. And none of those things... You know, basically, like, you get your first 1% and your last 1% of value generation, and you miss the 98% in the middle. Australia's economy is flat. That's insane. Like, it's a peaceful, modern, you know, western liberal democratic system with well educated, relatively young, demographically vibrant workforce, good culture, no violence, no no threats, really. Like, New Zealand's not gonna invade, and unlimited natural resources, their economy is flat. Lotus numbers. [00:20:02] S0: Let's talk through the build out of Starmind, the economics behind it, and what the timeline looks like for that. [00:20:10] S1: It's interesting. So I'd say it like, I wish there was moon dirt here. Yeah. Yeah. It'd be so cool. Starmind. I don't like the name very much. You should talk with Bev. Is it his fault? [00:20:24] S0: Yes. [00:20:25] S1: Okay. Like, StarThink was right there. [00:20:30] S0: You like StarThink?
00:20:31 · Casey Handmer
[00:20:31] S1: Well, it runs Starmind. This... You got that joke potential. Right? Alright. It's an interesting one. So when I when I first, you know, heard of, like, oh, we're gonna do data centers in space, I was pretty skeptical. There's a couple of good reasons to be skeptical. The first is that, you know, the juice is never worth worth a squeeze when it comes to, like, oh, but there's infinite free power in space. It's not infinite, and it's not free. And, you know, if you're in in most orbits, you'll eventually be shadowed by by the earth, etcetera, etcetera. But there are a couple of ways to kinda get around that, and I think Starlink really shows the way. So people who've read some of my blogs will know that that, like, space based solar power is kind of dumb because you're taking a product where it's expensive and you're selling it somewhere where it's cheap, which is not what a business should do. So in space, power is really expensive, and then you're beaming that power as undifferentiated microwaves down to the earth where you're collecting them, and now you're having to compete with solar and nuclear and coal and gas and everything else for, like... I don't know. Like, what does what does your merchant power producer earn in California? 10¢ a kilowatt hour? Like, not very much. You pay more than that, but you're paying for distribution. However, Starlink works because instead of just beaming, like, a shitload of microwaves at the earth for power, they modulate those microwaves with, like, gram something or other, like, some quadrature amplitude modulation radio signal thing that allows them to transmit not power, but but information. Right? And so if you if you run the math on, like, what is the revenue per watt of transmitted microwave power for Starlink? It's a billion times higher than space based solar power. So while SpaceX is making money hand over fist with Starlink, and it's worth kind of putting a pin here in the fact that other people have tried and not succeeded, but space... SpaceX has, their margins might be, I don't know, a 100% or something optimistically. If you reduce their revenue by a factor of a billion, they are no longer in business. So, basically, you take that model and you say, well, we can make money being the middleman transmitting data from your content delivery service or an edge edge node or whatever.
00:22:41 · Casey Handmer
You know, Netflix is down the road here. Via our network, which initially was just this bent pipe system, and now they've got LaserLink set up so they can they can transmit a lot of, you know, a lot of data in in the shell, you know, the shell of Starlink satellites and then land it again at some other place, either at a consumer place or or, you know, like a consumer satellite dish or or a fiber hotel or, you know, a gateway, what they call a gateway, which is like a gigantic dish, which is, you know, where all the data goes in. They can make money doing that, which is very cool. Then you say, well, what if we, like, mess up our latency a little bit? And so we can put all these satellites in some synchronous orbit so they're always in the sun, except for during lunar eclipses. And then we don't need anywhere near as many batteries. And we use laser links to transmit data to and from those via the Starlink network. Although if you didn't have that, you could also transmit to the ground. And instead of charging people a $100 a month to relay, like, Netflix, where, like, the value of the marginal bit is not all that high. In fact, Netflix, like all these systems, uses error correction, so you can delete 10% of the bits, and it'll be fine. You won't notice. And instead of transmitting Netflix, you instead transmit tokens of some hyperintelligent model that you can update with software. So once you got the hardware flying, update software. And then you say, well, what is the comparative value between your your average Starlink bit versus your average Starmind bit of data? And it... It's, like, probably at least a thousand times more valuable, like a token. Like, what do you pay for tokens versus, like, your subscription to Netflix or something? Right? It's it's not even in the same category. I was doing a project on on Saturday where I rediscovered that, like, nested loops consume resources when I basically had an AI go out and scrape scrape the entire Internet for all sources of data on on 200 separate topics and then categorize that data across 200 separate principal components, which then resulted in, like, 40,000 times. Yeah. It's like $260 of tokens. Like, unbelievable. Like, this is... It gave me better results than a team of expert professional analysts would do in a... With a year's work.
00:24:48 · Casey Handmer
And this is just, like, just as an idea, just like me mumbling into my phone and then forgetting about it. Yeah. So that's that's real money. That's real money right there. It'll be interesting to see if economic demand for intelligence in the form of, you know, AI tokens can saturate. I suspect not. Yeah. So that's the idea. Giant giant solar array GPUs. No. And and, like, data center in space, you have this, like, massive thermal problem, but you don't necessarily have that if you do it right. Like, instead of building, like, a giant box with full of racks of hardware and cooling systems, you can actually just, like, sprinkle the GPUs out along your solar array. Right? Because, like, each solar module generates a kilowatt and each GPU consumes a kilowatt and then just have, like, local Wi Fi or or or optical fiber link or something to transmit data up and down the the solar array, you know, train or or or chain or whatever, and then you've got, you know, a star tracker and a couple of radio antennas and stuff in the middle. Then if your GPU dies, who cares? And it's like, it's bolt on to a solar array, and the GPU is the expensive part. Does it make sense? So it's like, well, how do you replace a GPU when it dies? In a in a data center, you got all these fixed costs. And if your GPU dies, you gotta go find it and rip it out and put a new one in. In space, you're like, no. No. Like, I've got a 100 GPUs on this particular satellite, and one of them just died. And then 99 of them are still going fine. And, eventually, you know, maybe you're down to one or two or they're very old and you just deorbit, burn up, crash in Alaska.
00:26:19 · Ty Morse, Casey Handmer
[00:26:19] S0: If you believe that Elon has the same view where token demand is effectively infinite, and if you just make more tokens or, you know, put more GPUs up there, someone will buy them, and either it'll be him or someone else, then how do you think he's thinking through the problem of just maximum scale as fast as possible? [00:26:39] S1: Well, it's not exactly a secret. And there's one other aspect of this, which is to say, Starlink as a business model is possible if you have, like, a legal monopoly lock on, you know, a plurality of the world's most talented engineers and a highly mature launch system that is extremely affordable. And as we've seen with Chinese, Russian, European, and other American attempts to compete in this space. It is extremely difficult to do if your satellites are not best in class, like, very good value, and then you don't have access to, you know, preferred launch, basically. So how do you do this as quickly as possible? There's probably some optimization to be done in terms of, like, maximizing tokens per kilogram, but, ultimately, it's about maturing Starship design and then scaling up production. So Starship Starship, Louisiana. Mhmm. Right? You got one one launch pad in Texas, two actually now, a couple going up in The Cape, and now they're saying 10 or 12 in Louisiana, and and who knows where else? [00:27:42] S0: This is another thing. He's building this massive launch facility in Louisiana.
00:27:42 · Ty Morse, Casey Handmer
[00:27:42] S0: This is another thing. He's building this massive launch facility in Louisiana. [00:27:46] S1: And Well, I don't think any I don't think any, like, DOTA's been turned over yet. Mhmm. Again, like, you have to go through the federal agencies because it touches the coast, and it touches the Intracoastal Waterway, and it touches the Mississippi system. So the Army Corps of Engineers is gonna get a say, and it's adjacent to wildlife refuges. So, like, national parks is gonna get a say, then and the state's gonna get a say. And, like, by the time... Even even at Elon's speed, it may take a long time to get the thumbs up to really start moving on that. [00:28:13] S0: You think so? I kind of feel like he... You know, if if I think about the infrastructure projects that Elon has said that he's going to start, I can't think of a single one that didn't... Maybe Hyperloop might be one where he kinda said, like [00:28:27] S1: He didn't do that personally. Yeah. If you think about, like [00:28:30] S0: But with these [00:28:31] S1: Well, they started work on Starbase in, what, 2012 or something, and, like, not much really took place there until 2018. And then once they started working in in earnest, they were held up for months at a time. [00:28:40] S0: Do you not think that he's... He has, like, built teams that are much more effective in, like, ways that are much more effective at getting these large scale projects done? [00:28:49] S1: I mean, yeah. Definitely, like, the reps, you don't get worse at it for doing more of it. [00:28:52] S0: Yeah.