00:02:54 · Saurav Shroff
Datasets requires a fully reusable rocket. So, I mean, if you think about the launch market, zero competition, one player that can service the present slash old market, and that is that is that is steamrolling the world, and racing towards a rocket that can service the next generation of spacecraft. Meanwhile, you have like 10 ish competitors, quote unquote, that are trying to catch up to the yesterday rocket, and basically just one competitor that's also trying to to build at the same time a rocket that can achieve the economics and scale of the future... To service the future market. So, why is the market bad? Sounds decent. Well, actually nobody... There are there are zero vendors. So, you know, call it a market is actually a misnomer. Right now, if you try to buy launch from SpaceX, they won't sell it to you, basically, under any circumstances. They may sell you like sort of obscure products, like they'll sell you launch to the moon, or they'll sell you, you know, have a new product called Starfall, which are great new products, but they will not sell you traditional launch services that are useful for either of these two applications, which collectively one plus 99 account for a 100% of satellites that humans will launch. Meanwhile, the other companies are also not in the market. Some want to be, but have no rocket, and so are sort of a selling, rocket launch service that doesn't exist. Most are not even trying to match, the capabilities of Starship, they're trying to match the capabilities of Falcon. And so, it might actually be the worst market of any kind, because it looks like there used to be market, there used to be supply, it looked like the supply was increasing, but actually there is no supply, and no matter how much money you have, you can't buy supply. So, very bad.
00:04:52 · Ty Morse
[00:04:52] S0: So with this sort of thing, if SpaceX is gonna be mostly putting up AI data center sats in a few short years, I think Elon said that the cost of putting compute in space in like three years from the data of the Dwarkesh podcast, it would actually be cheaper than here on Earth. Sounds about right. And if that's his view and he believes that roughly there's like no, you know, demand limit on intelligence, why would he like ever sell launch to anyone else even if it scales, you know, scales launch like crazy with Starship?
00:05:23 · Saurav Shroff
[00:05:23] S1: Yeah. And that's the perfect question to ask. If I owned a monopoly on a service and that service unlocked the scaling of what I personally and maybe others believe to be the most useful and valuable technology ever made. And the supply and demand relationship of that technology is that there is drastically more demand than supply, which is to say the prices are extremely high, the margins are extremely high because there is more demand than supply. Any rational economist would say that it's dumb to sell launch services, the bottleneck, to anybody else. And so, actually, the future that I see is there's really two possible futures. Future one is that, we create competition, and I think this is the, probable future. We have some very good reasons to believe that that's the case. And then future two is we continue to fail to create competition, and the result is that all of super intelligence capacity is owned by one company that isn't motivated to reduce price, and isn't motivated to increase quality. And, you know, if we just ask ourselves, this is like not some kind of like revolutionary question, which future is better? Just about every single person in the room, except for the holder of monopoly, would say, the one with competition.
00:07:02 · Ty Morse
[00:07:02] S0: If you had to think through your own business, if your entire business relies on the ability to launch your product to Orbit, hopefully in a few years. Right. How do you think through, like, basically solving that problem or working with one of the other providers, so it kind of makes sense where you can scale at a reasonable rate?
00:07:21 · Saurav Shroff
[00:07:21] S1: I guess it's best to answer your question by breaking up the future into two relatively distinct time periods. We're gonna call the first time period, which is in the immediate future, the period where AI data sets are not the majority. And then we're gonna call... And we... I think that's, like, roughly now till, like, 2029 or something, or 2030. Then we're gonna call the period after that the period where AI sets are the majority. And when I say the majority, I mean the majority of satellites sent to space measured by their size or power. Okay. The aggregate size of power of all the satellites. During the first phase, our businesses really has no problems. As in, SpaceX, before exiting the launch market, which they recently did, sold lots of launch, using their current rocket Falcon nine to many companies, most of most of whom are our customers. And, there are other rocket launch companies who will be moderately successful in catching up to Falcon nine's capability, which is to say 10 to a 100... Well, I guess Falcon's higher than a 100, but 10 to a 100 flights a year, 10 tons per launch, and the economics of a rocket that is not fully reusable. And all of those things will be true over the next two to three years, and during that period of time, our products will continue to sell and continue to provide immense value to customers. So we're not worried about this phase. This phase one. Phase two is where there's some reason to be concerned. Let's look at what an AI dataset is, why it's better than a terrestrial data center, aka a data center on the ground, and then what's required to make it good on the rocket side. So AI data satellite, very simple. Take a chip, take a memory, basically take a compute module, not memory, take memory, sorry, take a chip in memory, and put it inside of a satellite that's quite a bare bones satellite that's extremely large and generates a huge amount of power, but is very simple in that it doesn't have, you know, great pointing accuracy, it doesn't have the ability... It doesn't it doesn't have very many unique capabilities. It's basically just a a powered shell. So just a box that has a huge solar panel, a tiny battery, some interconnection, you know, networking, devices, some attitude control devices, a small thruster, but basically a very bare bones satellite. That's what an AI data satellite is.
00:09:41 · Saurav Shroff
Roughly speaking, an AI data satellite needs to be about $5 per watt for it to be competitive with a data center on the ground, because the data center on the ground costs between 10 and $15 a watt, and lasts a little bit longer. Okay? So if you wanna make it better, has to be about $5 a watt, and that's all in. So that's building the satellite, that's every single component that's out the door and into space. So roughly speaking, to make a $5 a watt satellite, something like $1 per watt needs to be launched. Needs to be the launch cost. So if we break this down, and we look at the numbers, and look at, let's say, a 200 kilowatt class data satellite, this satellite should cost about a million dollars in total, of which something like 800,000 will be the cost to build the satellite, and 200,000 should be launch. If 200,000 is launch, and the satellite weighs something like 1,000 kilograms, which is quite optimistic, actually, for the weight of the satellites, actually very optimistic for the weight of the satellite, then launch needs to needs to cost $200 a kilogram. Right now, SpaceX, well, when they were in the market, would sell you launch on Falcon Nine for single digit thousands of dollars a kilogram, and it would cost them something like $1,000. So you can see how Falcon Nine is not a sufficient vehicle to make... It's five x off. Yeah. It's five x off. Exactly. Additionally, if you look at any of the companies that are currently deploying data centers at large, okay, look at OpenAI, Anthropic, any Neo Cloud, SpaceX themselves actually with XAI. They're deploying data centers on the order of sort of one to 10 gigawatts a year. So let's do some more math. If a 200 kilowatt satellite is one ton, and a rocket like Falcon Nine can carry about 10 tons to sun synchronous orbit, which is where AI data sets need to be in order to be useful, each Falcon Nine launch can carry two megawatts. What that means is that to get to two gigawatts, okay, which is a medium number in comparison to in comparison to one to 10 gigawatts is a thousand launches. So, Falcon nine is 10 x short on scale, and five X short on cost. And so what this means is that you need a roughly one order of magnitude improvement in both scale and cost, just to get to the baseline, where the AI data set is both better at cost wise, and even modest in scale. Right?