저자: Kevin · 외국인
요약 보기
구글-스페이스X 로켓발사 계약 협상은 SpaceX IPO 기대와 우주 데이터센터 구상에 유리하다는 주장. 구글은 SpaceX 지분 6.1%를 보유하고 TPU를 우주에서 검증 중이며, KPEG 1.86 기준 목표가 약 574달러(현재가 400달러 대비 43% 상승 여지)를 제시. 다만 발사비용이 kg당 200달러 아래로 내려가야 우주 데이터센터가 성립하며, SpaceX는 현재 250~600달러 수준이라 아직 멀다고 봄. 구글 가격결정력 둔화 우려(Klarna/Affirm 도입, 요금제 변경)도 언급.
다듬은 글
A holy molly, Google is in talks with SpaceX uh for a rocket launch deal. This is absolutely brilliant because it adds to the IPO pump for SpaceX, which we already think is going to be the most insane pump ever. So much so that it's going to make SpaceX so freaking rich that they're just going to end up putting a put option on well, frankly, Tesla because I frankly think SpaceX is going to end up buying Tesla because Elon controls most of SpaceX and that'll be the perfect way for him to scoop it up with a really hot floating SpaceX stock with probably less than 10% of its shares outstanding on the market.
But, we already know that. This video isn't about catching you up on that. You can watch my last SpaceX and Tesla video regarding my Tesla thesis. Just type into YouTube, Meet Kevin Tesla SpaceX and you'll find it. What's more important here is what this means for Google and SpaceX themselves. First of all, this is absolutely brilliant for Google.
See, Google has 6.1% ownership of SpaceX. They're pretty early investor. I'm grateful that my venture capital company is also an early investor in SpaceX, so forgive me I'm a little hyped about it all, but I also like being realistic about it all and that's what we're going to do is we're going to be realistic about this data centers in space idea, which is basically what Google's trying to go for here by securing a hey remember when we helped you out back in the early days?
Uh you mind getting our space data center chips to space first? We believed in you first. Now you owe it to us. It's It's kind of brilliant because at the same time SpaceX has filed for an FAA license to launch up to 1 million satellites for orbitable orbital Got to say that right. Data centers. No, space data centers.
And Google's been working on this for a while, so this isn't new. Google's also talking to other launch companies, so we're not sure if they would just use SpaceX or if they would use basically anybody who has available capacity. Maybe a company like Rocket Lab, which has also been going to the moon next to a company that we've really been investing in the course member live streams and course member group like Marvell Technology.
Marvell Technology totally topic for a different video on fiber optics and ultimately nodes node expansion to on device AI and 5G expansion. Sorry, I'm like a squirrel around here. We just got a lot to talk about. Let's stay focused. Rocket Lab's been going to the moon probably Blue Origin, which is obviously Jeff Bezos's private, so you can't invest in publicly.
But, Google's been working on this for a while. In fact, they have a 2027 prototype project called Project Suncatcher. Yeah, no, not like a sunflower or a solar panel. It's actually really just a project to put Google TPUs in space. And see, the whole reason you want to put data centers in space and we've talked about this before, so I'll keep some of this short, is that you don't need batteries or electricity.
Well, you still use electricity, you just get it from solar panels that can essentially capture sunlight 24/7. The reason for that is you can essentially ride the dawn and dusk cycle of what we see on Earth, but you can ride this zone called the terminator line and it basically means you get 24/7 sunlight.
It you never really get sunset. Which is kind of cool. And so, Google has partnered with this company called Planet. Technically, Planet Labs is their company and they have a newsletter called Planet Pulse, PP. I had to give them credit for the PP shout out. But anyway, this is a company that's providing two prototype satellites for Google and the whole purpose is to evaluate, hey, can these Google TPUs actually operate in space?
And a lot of people are really excited about this because people see Google as not just, you know, a profitable SaaS and advertising business where you're providing Google Workspace, right? The enterprise Gmail and Google Drive storage, you know, you sort of your traditional server storage, but you're also providing artificial intelligence enterprise access through monthly fees, seat-based, right?
Which is your classic software as a service business. But, you're also building and well, essentially designing your own TPUs. But, you're also investing in SpaceX. And so, you've got these incredible verticals on top of what's truly and originally an advertising business. So, Google is like I want to say the jack of all trades, but they're really like the master of all trades.
They've just blown through all of our extension lines over here. We're riding on 374 right here, waiting this 374 line, waiting to break out of 400. You'd actually see that resistance here at 400. I suspect once we break 400, Google's really going to skyrocket past that. Uh it pretty pretty rapidly.
Mostly because if you think about Google's valuation, these people ask me about, "Hey, okay, Kevin, you're so bullish on these, but what's the valuation then?" Okay, fair, fair. Uh let's pull Google's valuation. So, in the um course member stock tab, the last information that we have, uh our last analysis on Google was on April 30th, and we analyzed that Google currently has a KPEG ratio, so uh Kevin price earnings growth ratio using the forward 4 years of growth of 1.86.
1.86 gives the stock currently a 43% upside. At 400 bucks, this brings it to about $574 as a price target, fair price for Google. So, Google's crazily still selling for a discount. The balance sheet's okay, you know, momentum's not that desirable, but its pricing power is still growing, which is phenomenal because there's so much competition for artificial intelligence.
I mean, I I have some questions about how strong this pricing power is going to actually last for Google. In fact, for Gemini, they're changing some of their pricing strategies. First of all, they're offering Affirm and Klarna, which is starting to make me think, "Really? You have to go to buy now, pay later to get customers?" Okay, a little weird.
Probably because there's a lot of competition with Claude, right? But on top of that, they're removing the AI Ultra from the enterprise workspace side, which is actually what we use over at House Hack, and they're transitioning to this thing called AI expanded access, and it basically drops your price from like $100 to $200 per seat to like $24 a seat.
But I think what they're trying to do is they're trying to say, "Hey, so few people are signing up for the $100 or $200 plan. Let's just try to get a lot of people, like maybe 70% of people on the $24 plan, and then that'll net us way more than having like 5% of people." You know, I'm making up those percentages on on those other sides.
But it does show you that some of the pricing power is still cooling, but this company is really just this massive call option on on frankly the future. So, I really think it's going to blast through 400, and that's a call that I've made in the course member live streams as well. Remember, you can always join those at mekevin.com, but focus on this.
Here, just looking at their sheets for a moment. Both Meta and Google have similar free cash flow, and both of them have stopped doing buybacks, and their debt positions are similar. Meta is actually growing a little bit faster on top-line revenue, but Google has the Gemini and space data center push, and the SpaceX exposure that Meta doesn't have.
So, you know, Meta is sort of no SpaceX exposure. You've got still that high capex, no buybacks, but again, no SpaceX data center exposure, or or space data center plans really. We haven't seen those. And big, no TPU plan. And so, that's why I want to get back to sort of idea about designing tensor processing units.
One of the reasons Google has been winning in TPUs is because they've been building these for over a decade. And so they're actually way ahead of other labs trying to generate their own chips. In a separate video, we're going to talk about Cerebrus. Uh I actually brought, you know, a Cerebrus chip. Actually, this is actually just sort of like a waffle and Cheez-Its.
But we're going to use that as an example to explain what's going on over there with that IPO, which you know, that that's that's coming. It's going to be fun. But um what's fascinating about TPUs is they've already proven to work in space. And the problem with space is you have a lot of radiation in space.
And so chip equipment needs to be able to survive this radiation, otherwise you get this problem called bit flip. It's basically cosmic radiation. And what's actually most sensitive to bit flip is high-bandwidth memory, which is like your Samsung, SK Hynix, SanDisk, right? Like all of these might actually end up having a lot of challenges in space.
So they're not your best probably hardcore uh space data center plays. They're like your on-land data center plays. You know, we Blackwell chips use a lot of even external high-bandwidth memory, so really important. But you really need to harden chips for radiation. And Google has already proven that their TPUs can pull it off without bit flipping, which is where you get, you know, high-energy cosmic radiation, sun flares, or whatever flipping like a one to a zero and or causing some kind of electrical abnormalities where, you know, you get little surges, and then they fry the chip.
You ever wonder why like stuff in your house just dies? It seems like the more smart features you have on like your washing machine or your dryer or dishwasher, the less long it lasts. Like the 40-year-old crap that has none of these little microchips just last way longer. A lot of that honestly is because of little household surges.
Every time you kick on the air conditioner, you got another little surge and it's putting pressure on these tiny little microchips. And wow, what a surprise, the stuff doesn't last as long, which is, you know, great for the manufacturers cuz most people don't want to repair their appliances because they realize in a lot of cases it's almost just cheaper to replace it than figure out what freaking little, you know, 10 cent Chinese microchip actually broke.
Cuz you know, to get a service tech out it's like 200 bucks, it's crazy, right? Just to get them out. And then they got to order parts and they got to come back, it's another $200 and you got to pay for the part, it's crazy if they even diagnose it right the first time. So, you have the same problem out in space, except instead of little AC power surges you have solar flares.
Well, Google's already tested one of their TPUs in space. It's already survived what they call 15 kilorads. Basically like kilo radiation units, standard units. I don't know that much about radiation, but essentially it's a lot. And it's three times what they would expect over a five-year period of of a normal exposure amount that these sort of chips would get.
And they haven't had a hard failure on these. Now, that sort of opens up the second big problem for Google and really anyone trying to place space-based data centers is cooling. Now, this is probably the most misunderstood portion of space data centers where people are always like, "Hey man, Kevin, space is really cold." Yes, space has a lot of theoretical cooling advantages.
The ambient temperature could be as low as like -269°. But there's no air flow cuz space is a vacuum. And so, there's less of an issue with what the temperature is in space, even though it could widely fluctuate, especially if you get solar flares from like really cold, like 4 Kelvin, to really hot, you know, 200° hot.
So, you you need to be able to withstand those sort of extremes. But the most important part is actually radiating heat away from your chips. Because if you're running small little test chips, that's cool. But if you're running a data center, you're going to generate massive heat and you need to radiate that heat out.
Some say for one for every one megawatt of data center compute, remember you would have 1,000 of what I'm about to say for one gigawatt of data centers, you would need about for every one megawatt, you would need about four tennis courts simply to radiate out that heat. So for one gigawatt of compute, you need about 4,000 tennis courts of imagine just sort of like giant tentacle arms that radiate out uh that heat that you don't want burning up your chips.
So you could do this with chemicals, but there are serious mechanical points of failure here, very toxic ammonia-based chemicals. It's really hard. So a lot of this like space stuff, I like to call a call option. You know, that company I talked about Planet that's already partnering with Google, they're famous for their Dove satellites.
They're like these little shoe-box sized nano-satellites in a box. Uh in fact, NASA has a CubeSat video I'll go ahead and pull up. It's kind of cool. Uh but the the real question is are these little low-cost satellites going to be the way we want to get satellites into space uh for data centers or do we want something bigger and more robust?
And again, biggest issue is how do we cool it? Here's what the CubeSats look like. This is the company that's been partnering with uh Google. But basically, they just sort of draw it out here as you could have these little modular cubes. And each cube can actually fit in your hand. They'd be about the size of 1 kilo uh kilogram, low-cost, and there are about 2,300 of them in space right now.
Call it a CubeSat. I don't know. More stuff you can look up if you care. But uh right now that company Planet photographs the surface of the Earth basically daily and then they license that content to governments or really anybody who wants it. So, kind of fascinating. But, all of this brings us back to SpaceX.
See, the question of what's going on between SpaceX and Google is how low or how inexpensive can we actually get the cost to launch? That's the goal. And so, the viability of getting data centers to space all comes down to getting the cost to launch down under $200 per kilo. Once you cross that threshold, that's when space data centers can work.
Now, SpaceX says that they're targeting to get about 1 million tons of payload to space on an annual basis. If they charged $50 per kilo for that, that could potentially work out to 1 trillion dollars of revenue and get 100 gigawatts of compute per year into space. Now, that's crazy. I don't actually think we need 100 gigawatts of space.
I think the claims are a little sort of like Elon Musk is going to colonize Mars. And I don't know, maybe maybe it's just like call him crazy and watch him do it, but these are some really crazy numbers, especially since right now we're still at a cost of between 250 to $600 per kilogram for a launch.
So, we're not yet at the point where we can expect data centers or close. We need a lot more development at SpaceX and a lot more money at SpaceX. We really want to see the cost get down to like $15, maybe $30 per kilo at SpaceX because then SpaceX could charge 50 or 60 per kilo and have a massive fat profit margin.
Of course, there's going to be competition from the Chinese and from Blue Origin or Rocket Lab or whatever. That's fine. I think SpaceX will be a leader here, but it's really important to know that right now all of this is like a big whole play because we don't actually know that data centers can work in space with our technology now because we don't have the radiation capacities that to radiate heat when I say radiation.
There's also cosmic radiation, so the chips have to survive that on large scale. We're not sure about that yet. So there are a lot of ifs, ands, or buts here. And we don't have the Starship actually functioning on a regular reusable basis just yet at a cost that is below $200 per kilo, which is not there yet.
This is why SpaceX needs to subsidize what they're doing in space with deals. And one way they're doing that is by partnering with, which I think is brilliant, Anthropic. Now, forgive me for sounding really bullish here, but this was really smart. Because see, SpaceX's deal with Anthropic to supply 300 megawatts of new computing capacity over 220,000 Nvidia GPUs by the end of May is so smart because honestly I think xAI was losing money hand over fist.
And getting a deal with Anthropic shows that they're willing to sort of forego Grok. Who cares? And just sell to the one that's absolutely killing it right now. And let's just be real right now, that is Anthropic. Anthropic is totally dominating right now. Uh Anthropic usage at at some companies has four x'd.
Which was one of the reasons Microsoft wanted to escape ChatGPT uh in their exclusive partnership so they could start offering uh Anthropic models along with Amazon to people who want it. Actually I have this chart over here, which is an interview of uh people who read the information. Do you use the following AI agent or coding assistant services?
And we've got the skyrocketing of cloud uh code, and then, you know, more recently cloud coworker over uh over here, cloud coworker. Whereas uh these others have all been coming down cursor. Look at this. Look at this plummeting cursor. Opening eye, flat line. You know, Grok isn't even on here. He don't even register on here.
So, it's really smart for them to say, "Fine, let's go where the money is because we need to subsidize space because AI usage on the ground in land sort of on terrestrial areas is going to keep accounting for a lot more of energy sector consumption. Right now, it represents about 8% of commercial consumption as of 2024.
I know that's like a year and a half old. That's growing. That's expected to go from 8% to about 20% by 2050. So, you know, one in five kilowatts are going to data centers essentially over the next 24 years. Pretty remarkable because that's, you know, one in five of our kilowatts just going to data centers.
That's crazy. You know, compared to everything. Military, health care, finance, restaurants, malls, movies. Whatever. Maybe movies isn't the best example. Kind of your dying industry, right? But, you get my point. It's a big number. So, once we could figure out these problems and we could get the cost per kilo down, we're looking Gucci at SpaceX and Google's at call option play.
Google's got the upside on valuation. SpaceX is going to, in my opinion, squeeze on low float but not on fundamentals. It'll squeeze on momentum and hopium, not on fundamentals, which is a bit of a risk factor. Whereas, Google, in my opinion, probably has the biggest upside because even if it doesn't work out and we don't get these space-based data centers, Google's still cranking on Gemini and their cloud services and advertising platforms.
The only red flag is why are we now introducing Klarna and Affirm? Are we starting to lose the pricing power war to an fropping? Something to think about. Anyway, if you like this segment, make sure to subscribe to the video ]
Comments 0