[00:01] There is something really strange going on with SpaceX and we got to talk about it because yesterday SpaceX reported earnings. Today the stock tanked and tomorrow there going to be some really massive stock lockups that open up which
[00:20] means potentially more people will sell. But the real strange thing I want to start with uh actually looks like a little chart and it's not a stock price chart. Instead, it has to do with um selling compute, which seems to be most
[00:37] selling compute, which seems to be most of what SpaceX really wants to spend their money on right now. In fact, about 85% of all of the money that they collect, they're not burning it on rockets or V3
[00:52] Starlink satellites to get that 10x faster. faster. Well, Starlink, no, it's going into data Well, Starlink, no, it's going into data centers. Yes, Micro Hard and Mini Hard.
[01:05] But there's a little weird thing going on and I feel justified in asking some questions. Start with the chart. Let's say on the
[01:17] left side of the chart you have annual revenue, okay, from your data centers. This will be measured in megawws per uh uh you know month, let's say. So
[01:29] we'll call it MW per month. Okay? And this will be in millions of dollars. So, I'll put a little M there. Let's say you make $4 million per megawatt that you're
[01:41] able to sell of compute and the scale is 0 to 30 and let's say you make four. Well, that would put you somewhere over here at the dotted line, right? Pretty
[01:55] low, pretty close to the bottom. That's understandable. Right now on the x axis we're going to have the cost per megawatt to build these facilities. And
[02:08] dollars and it's also going to go from zero to 30. And in this case let's say the average cost for like an you know a the average cost for like an you know a nebus or coreweave is something like $15
[02:22] million per megawatt. Well then you'd be somewhere in the middle. So we'd have this intersection point somewhere over here. Right? So the average data center should be somewhere low and in the middle of this chart now let's say
[02:38] you're a total freak of nature. Let's think about this for a moment. Total freak. And you blow all your money on the best chips in the world. and you literally pay the highest premiums possible
[02:53] and you make average revenues. Well, in that case, average revenues would be down here in the chart and you'd be paying, you know, a premium you'd be over here to the right. Right? So, this is average.
[03:08] This right here would be paying a lot and earning very little. Now, let's say you were a different kind of freak. Let's say somehow you were able to
[03:20] convince the world that you could actually build data centers for close to this side of the line, cheaper than anybody else. Yet somehow you were able
[03:32] anybody else. Yet somehow you were able to make six times as much money as all of the other people selling the same stuff. I mean, that'd be crazy, right? In that case though, you would be somewhere over here on the chart, right?
[03:49] That'd be crazy. So, if you overpay and under earn, you're here. If you're under earn, you're here. If you're average, you're here. And if you're well, crazy, I guess, or or magnificently
[04:04] crazy, I guess, or or magnificently crazy, then you're up here in the chart. Okay. So, why did we just go through that exercise of where you might end up sitting on the chart? Well, what if I told you there was something going on
[04:18] told you there was something going on with SpaceX that puts it in a very unique position on this chart? Uh, what we would call a statistical outlier.
[04:30] we would call a statistical outlier. Let's pull up a real chart. Holy smokes. Let's pull up a real chart. Holy smokes. So we've got Coreweave down here at a normal, you know, cost to build per megawatt. Nebus over here at a normal
[04:43] kind of cost to build per megawatt in this middle range right here. And then earnings are somewhere between, you know, $2 million to $5 million per megawatt uh per month, right? Well, actually in this case, it's annual
[04:56] whatever per month per annual, you get the idea. Okay, the idea. Okay, SpaceX actually shows up right here at SpaceX actually shows up right here at the top earning six times its peers and
[05:11] the top earning six times its peers and somehow building for 15th of its peers. This is a statistical anomaly. Let's look at the actual data here. SpaceX is telling everyone that they can build for $2.7 million per megawatt. Uh, okay. and
[05:29] that their annual revenue is somewhere around $30 million. Annual, I did say per month on my little sample whiteboard. Meant to say annual. Um, Nebius is able to collect revenue of about $4.5 million. Coreweave is able to
[05:44] about $4.5 million. Coreweave is able to do about 1.65. Core Scientific can do 1.75. The average hyperscaler collects 2.3. So somehow SpaceX is out earning all of them and they're able to buy it all for
[06:00] substantially less than everybody else. So that got me to be a little bit So that got me to be a little bit curious. What's going on here? How could this be possible? Is this an Elon premium? No. Is this being an Elon
[06:15] hater? No. Let's actually analyze what's going on. See, in the earnings call, Elon Musk suggested that they are able to install compute with less than a one-year cost
[06:28] incredible claim, so I wanted to stress test it. And then I found out where this $2.7 million per megawatt actually came from. It's not from an auditor. It's not from an SEC filing. It's not in the S1. It's nowhere on a formal document. It's
[06:46] simply what SpaceX has suggested they were able to build the Colossus 2 data were able to build the Colossus 2 data center at not including any GPU hardware. It's literally just the shell. No
[07:01] audits, just a management opinion of what it costs for them to put the shell together and the power infrastructure. So, is it a lie? No. But it's missing a whole lot of context. See, Deutsche Bank argues to now fill this up with GPUs,
[07:17] you would have to say that the cost per megawatt's going to be somewhere between megawatt's going to be somewhere between 25 to $30 million, which is essentially in line with the industry averages, huh, interesting. So, in that case, what is
[07:32] Colossus 2? Well, Colossus 2 and Colossus one for the matter are actually considered brownfield data centers. A brownfield data center is different from a green field data center and it has nothing to do with green energy.
[07:45] Brownfield data centers are basically data centers that are built inside of an existing structure like an old manufacturing facility that used to be used for something else that can now be a data center. This is what SpaceX did
[07:57] back when it was XAI uh for Colossus 1 and 2. They took over existing factories with power shells in place, faster to permit, even though Elon still ended up using unpermitted gas turbines, and SpaceX has had to promise to remove
[08:11] those illegal gas turbines by July of 2027. So, in about a year from now, so the clock is ticking on that. That's a very different speedwise and costwise than taking a blank piece of land going through permitting for 18 months to 24
[08:27] through permitting for 18 months to 24 months compared to the 91 days for slapping together Colossus 1 and two which individually the the faster one Colossus 2 was done in 91 days which you have to admit is pretty freaking
[08:39] have to admit is pretty freaking impressive but when you hear that $2.7 million per megawatt you have to know that is a brown field deployment inside of an existing facility with a lot of power
[08:52] infrastructure and utilities already in place. It is not building up from the place. It is not building up from the dirt and it does not incorporate GPUs or the servers or the switching equipment or the other things that really make a
[09:07] data center tick. Now, that's really important because the marketing SpaceX investors, a lot of them are using is this $2.7 million per megawatt figure. and it's going to screw up a whole lot of estimates for people. So that is
[09:22] important. Now in fairness, that's just part one because even if it costs you say Deutsche Bank's estimate, take a midpoint of like $27.5 million per
[09:34] megawatt to deploy let's say mini hard or macro hard. Okay, let's just say you could pull that off. If Anthropic is willing to pay $30 million per megawatt,
[09:48] you know, per year, then Anthropic is basically paying SpaceX back the full basically paying SpaceX back the full investment in less than a year. That could then still be defensible, right? So, is that good? Does that mean
[10:03] we're ready to rock? The chart didn't mean anything. Not quite. See, the problem with this chart is that it has two concerning chart is that it has two concerning issues for SpaceX. Number one is what we
[10:18] started with, which is the marketing around cost to build. How are we on the far left side? Oh, got it. We're actually not on the far left side. If we want to use advanced hardware and actually consider the GPUs, we're
[10:32] actually over here on the far right side, but the marketing implies that they're over here on the far left side. All right, misleading, but hey, a little bit more research and we can get that. But there's a second problem. The second
[10:45] problem is how is SpaceX possibly sitting at this much revenue when everybody else is sitting so low on revenue. Well, to understand that, we have to understand the incentives.
[10:59] First, Colossus number one was rented to Anthropic. The entire facility was Colossus number two was partly rented to Anthropic and that's partly because on training and more on inference and then they rent it some to Anthropic at a
[11:15] about in just a moment. There were some rumors that Grock was having latency issues and SpaceX wasn't actually able to properly utilize these facilities for training uh uh Grock, but those are just rumors. So, we'll let that be. that was
[11:29] reported by the information which is often considered credible but that rumor never ended up getting verified or denied. It just sort of died. So the revenue rates for the anthropic contract are $30 million per megawatt per year.
[11:44] It's about a 10-month break even payback, right? Once you consider all of the uh computing infrastructure. Google then signed a contract with Google then signed a contract with SpaceX for some of their extra compute
[11:58] SpaceX for some of their extra compute at $50 million per megawatt per year, which is really interesting because 50 divided by 30 is a 67% premium to Anthropic. Google signed a 67% premium over Anthropic when
[12:11] 67% premium over Anthropic when Anthropic is already paying a 6x premium Anthropic is already paying a 6x premium uh over what these market uh uh rates are from Hyperscalers, Core Scientific, Coreweave or Nebius. Now, why is that?
[12:27] Is it just because they want it and and it's available at SpaceX? Come and get it, but you're going to have to pay a premium for it. and and you know Elon's just really good at negotiating maybe. Or
[12:41] maybe. Or could it be this? Google holds $94.1 billion in SpaceX shares. That was as of June 30th. So the valuation has certainly changed a lot. But in Google's last earnings report,
[12:55] the vast majority of their earnings came from appreciation in SpaceX shares. Of that $94.1 billion, $80 billion of their money is locked up on the short-term lockup schedule, $14.1 billion locked up on the long-term schedule. So, a lot of
[13:11] on the long-term schedule. So, a lot of lockups. The bulk of Google's net income in that quarter of their $14 billion, SpaceX appreciation. In fact, this was negative and people are like, "Okay, this is an anomaly is how it was
[13:26] described." But regardless of Google's earnings, this video isn't about benefits massively if SpaceX stock booms. Well, Google also has a buddy renting from SpaceX. You'll see why they're a
[13:40] buddy in just a moment. It's Anthropic. Colossus 1 is rented from Anthropic. Colossus 1 is rented from Anthropic. Google happens to hold a $124 billion stake in Anthropic, which is actually almost 30% bigger than the stake they
[13:56] hold in SpaceX itself. So, let's put this together for a So, let's put this together for a moment. Google makes money if SpaceX stock goes up. Google overpaying for SpaceX stock on a
[14:11] Google overpaying for SpaceX stock on a cancellable three-year contract could make SpaceX stock go up. Google has a $15.3 billion backs stop Google has a $15.3 billion backs stop tied to a data center for we don't know
[14:26] tied to a data center for we don't know who in Hubard, Texas. Well, who's [snorts] developing a data center in Hubard, Texas? Anthropic. Anthropic
[14:39] Anthropic might need some back stops. Might also need some money. Anthropic making money means they too can overpay SpaceX for SpaceX's Colossus,
[14:55] which then boosts SpaceX's stock theoretically theoretically and helps Google. [laughter] And if Google gets helped and Anthropic does well, then Google wins again
[15:11] because Google owns a ton of Anthropic. So, you got yourself a massive cluster So, you got yourself a massive cluster circle to explain this anomaly.
[15:23] So, the marketing of $2.7 billion per gigawatt excludes all the GPUs. The reality is they're not building these facilities cheaper. And even if they could get the shell like the 10% of the cost cheaper because they're using a
[15:37] cost cheaper because they're using a brownfield strategy to get speed going uh when they actually try to scale this and start going to green field which is turning into facility it's going to take a whole lot longer.
[15:51] That's problematic on the cost side. So the cost side is an illusion. You're really over here at the hyperscaler average. You are not unique on these costs. Jensen does not like you enough to basically commit a massive proportion
[16:05] to basically commit a massive proportion of his GB300 chips [laughter] to you at a discount. It ain't happening. happening. Uh and then these premium rates
[16:17] actually we don't know if this is true but appear to be from people who would substantially benefit from SpaceX stock doing really well. The problem with these agreements at these elevated terms is renewal or worse
[16:34] these elevated terms is renewal or worse cancellation. See, Anthropic rented 100% cancellation. See, Anthropic rented 100% of Colossus 1 with a 90day rolling cancellation rate, which means at any point they can give their 90-day notice
[16:48] point they can give their 90-day notice and essentially be done. There is a uh a sort of an initial waiting period. All of those initial waiting periods end by the end of the year, although there's one that could end even sooner. I'll
[17:00] talk about that in just a moment. Google uh well, and then Anthropic is also renting, in addition to renting 100% of Colossus 1 with 90-day rolling part of Colossus 2 with 90-day cancellation rights. Google is also
[17:13] renting from So, those were were anthropic. Google is also renting from SpaceX, and they also have the 90-day cancellation rights. So basically the cancellation rights. So basically the 90day cancellation rights are everywhere
[17:25] after the introductory periods 90-day lease cancellation rate. The problem is Google can actually cancel early if SpaceX can't deliver functional access
[17:37] SpaceX can't deliver functional access for Google's deal by September 30th, which that's in 56 days. Now a problem for this is memory. See,
[17:49] Elon doesn't want to lose that contract because it's going to hurt their stock, which hurts Google. So, Google might not cancel anyway. But Elon's already kind of building the force majour narrative of well, we we
[18:04] couldn't deliver because it's not our fault. How's he doing that? Well, right here. Then I look at the rate at which logic and memory is being produced. And one was always considered a limiting factor here. The limiting factor
[18:17] currently is memory. the memory output is increasing by about 20% per year. Normally, that would be fantastic. But if demand is increasing 20% uh but demand is not increasing 20% per year, it's increasing 200% per year. So, in
[18:33] for RAM. We already know that they're up, you know, 4x year-over-year. of a seed here that he might not be able to fulfill that Google contract. Maybe maybe I'm reading into that too much, right? we can't get the memory. But then
[18:49] it's also worth thinking about this. A lot of the capex that SpaceX is going to spend is going to go into at least in micro and mini hard. Elon just talked and in the earnings call, we're going to put 220,000 GB300 Nvidia chips and uh
[19:06] 800G NIC's. Okay, these are your Infiniban connectors into these data Infiniban connectors into these data centers. Each rack holds 72 GB200s. So that means we need 3,56 racks per facility. Each of these racks runs about
[19:22] facility. Each of these racks runs about $10 million. A $10 million rack. Damn, that's a hot rack. Literally, they got to cool the crap out of it. [laughter] But anyway, that's $30 billion for
[19:35] But anyway, that's $30 billion for racks. Damn, that's big. Uh, and then of Infiniban. that's going to run you a billion and a half. You got to, you know, get the facility up. That's going to be 4 to 8 billion bucks. You got to
[19:47] get storage. That's going to be, and I'm not talking about like, you know, for toilet paper, like actual um, you know, electronic storage. So, that's going to be another 1.5 billion bucks. So, your ballpark per facility here is going to
[20:00] be about $40 billion. You're going to spend about $80 billion from micro and uh, macro or mini and macro hard, which are really just puns on Microsoft. The problem with this is the only way
[20:14] those are going to prove to be a one-year payback period is if Elon can get these wild rates on compute which nobody else
[20:27] is being charged and nobody else is charging. while it's possible they could pull it off based on this initial information
[20:40] that I'm seeing here, I don't even like looking at the Wall Street speculation because a lot of the suits, the Goldman Sachs, the Morgan Stanley's, a lot of these folks have to play nice with Elon,
[20:54] invited to the earnings call anymore. They're not going to get invited to the events anymore. It's how it is. It's how they play the game. And Goldman has to be at the Elon events. JP Morgan has to be at the Elon
[21:08] events because they got a lot of customers at the banks that like Elon and want Elon information. They want to feel like their bankers are there for them. It's endless marketing. Every single time they pick up that that call,
[21:21] oh, here's your analyst with look. You know, our next question comes from Edison U with Deutsche Bank. Your line is open. marketing every freaking time. Marketing, that's all it is. So, I'm a little sus about what's going
[21:37] So, I'm a little sus about what's going on here with uh SpaceX's AI rev. And that's where we can actually get into some of the earnings because you could some of the earnings because you could see here earnings did blow up in AI.
[21:50] They blew up from 737 million to two uh 2.5 billion. A lot of that was due to a 2.5 billion. A lot of that was due to a $1.6 6 billion boost which is a very strange flat number as we began to offer cloud services to customers. Okay, so
[22:06] obviously it's a flat number like this because it's probably you know two one or two customers paying a flat deposit. It's probably anthropic and Google
[22:18] know with certainty because they don't give us any more clarity here. Okay, here. We're obviously still losing money over here from AI. We're, you know, I don't think anybody's expecting SpaceX to be cash flow uh or or net income
[22:33] are not. Uh, you know, over the last 6 months, they've burned almost $5 billion here. So, the net loss is is present. Their negative cash flow here is about Their negative cash flow here is about $25 billion in the last 25 months. And
[22:48] if we look at their balance sheet, I will give them credit. They've done a great job at the suckening in during the suckening where they raised basically $85 billion plus then they came in and raised uh a refinance
[23:02] facility afterwards or or you know a loan facility loan ladder I guess I should say afterwards they've raised a lot of money so here we can see $36 billion in debt $13 billion in bills it's approximately 50ish right and then
[23:16] they've got on top like once I subtract that out of these marketable securities ities and the cash they've got here. They, in fairness, have $50 billion of free cash. That's probably good enough for the next
[23:31] 6 months of capex. So, I'm going to assume uh that uh will be spent in the next 6 to9 months. So, in the next 6 to9 months, they will have burned all of
[23:43] their free cash. And that free cash is going to build facilities that Elon is telling us have fewer than a one-year break even economic set if you include
[23:56] cursor revenues. But when we actually analyze those contracts, we're really suspicious that those contracts are actually being made because they really want this Colossus compute or they being made because
[24:10] they're trying to pump the SpaceX stock, which would make it a very dangerous proposition to overbuild, assuming you're going to be able to rent out that you're going to be able to rent out that hardware, those GB300s at those prices.
[24:23] So, call me skeptical, but I think it's reasonable to be skeptical and and a little bit nervous about this. So, I don't know what to make of these projections of we're going to get to
[24:37] revenue. Eventually, we're going to get to a trillion dollars of recurring revenue. To me, it's hard to justify that uh when you actually dive into some of the initial notes that I made, which we've already gone through. contra the
[24:52] contract termination rights the incentives for way overpaying compared Corore and keep in mind the business that they're getting into this is a an extremely capexheavy business that they're getting into and
[25:06] the vast majority of it is AI infrastructure see look at that second infrastructure see look at that second quarter we spent $18.4 billion of which 85.8% 8% is AI infrastructure, the vast majority of it. Which means 14.2% of the
[25:21] money that they are lighting on fire is actually going to light money on fire via rockets. And I want them to light the money on fire with rockets. [laughter] Instead, they're throwing it into AI infrastructure.
[25:35] So, I'm a little bit disappointed by that because I think they're throwing a lot of good money after something that is potentially a misleading good. Now, why would I call it a misleading good? Well, and this is not here to bag on uh
[25:52] Well, and this is not here to bag on uh Nebius or Coreweave, but let's just be real about the actual numbers of those individual companies. And we can see, you know, one is doing a little better than the other, but both
[26:07] little better than the other, but both of them have some issues. So, let's go of them have some issues. So, let's go take a look at this first core. Okay, so Coreweave has a capital expenditures of about $7.7 billion in just three months.
[26:20] Free cash flow of negative $4.7 billion. We could actually see that uh right here, right? So there's our capex of 7.7. We subtract this, we're negative uh $4.7 billion of free cash flow. Net loss over here, $740 million. So, you know,
[26:37] not great. They did increase their receivables a lot, though. The current receivables are just from three customers. Potentially as much as 89% of Core Wee's receivables might be Microsoft, OpenAI, and Google.
[26:52] And it's not uncommon for those companies to pay 15 to 25% upfront. And then you actually get a smaller set of receivable income over time. The problem cash flow statement. We know they're burning the money on reinvesting. It's
[27:08] actually their balance sheet. Look at the balance sheet. Uh, I've got about $4.3 billion as of March 31st of capital billion as of March 31st of capital here. And I have bills to pay. I have
[27:23] long-term debt of 17.3 billion. Okay, that that already doesn't look good that that already doesn't look good because, you know, I've got again $4.3 because, you know, I've got again $4.3 billion and 17.3 of long-term debt. I
[27:35] just pretend they had no long-term debt for a moment. How much short-term debt payables, acred liabilities, their current debt, and other current liabilities. If I just add those categories up, we are sitting at bills
[27:49] categories up, we are sitting at bills to pay of 15.1 billion, which is 2 and a half times as much cash that they actually have. Now, in fairness, I'm pretty sure they've been raising money since this last filing, but that's the
[28:02] only way they could survive is through issuance. They have to issue debt and stock. That's the only way they could raise or or or survive rather. Like this is like this is not they're not in my opinion it's not like they're like hey
[28:17] going to go buy houses at a 20% discount. They're going we need to raise money to survive. [laughter] Uh it's a little different. In addition
[28:29] to that, they are now having to raise at over a 10% yield on their debt. The Wall Street Journal was reporting that they were expecting to raise at 1.25% lower than that. They actually had to raise it by 1.25% just to attract enough
[28:44] capital so they could end up getting uh what they needed. what they needed. Okay, not great. Then we can get into the Nebius numbers. So the core weave numbers eh it leaves something to be
[28:58] numbers eh it leaves something to be desired. Nebius is not as bad. They're desired. Nebius is not as bad. They're in a much I would call it they're in a of their balance sheet. They've got
[29:10] about $9.3 billion of cash and I've got current liabilities here of about 1.3 and that includes like $700 million of deferreds. They do have 8.4 of long-term debt. So, you know, there are some bills to pay in that, but they've got enough
[29:26] to pay in that, but they've got enough money, but they are also spending a lot. They're spending pretty much all of their cash flow, uh, spending $2.5 billion, and of course, they're also issuing notes, warrants, whatever they
[29:40] issuing notes, warrants, whatever they can do to keep raising or taking investments from Nvidia, which they did do that as well. This is a very capital intense business. The issue that I have is why are
[29:53] companies like Cororee and Nebus having to basically give away their compute relative to those contracts that Google and Anthropic have with SpaceX. And then and Anthropic have with SpaceX. And then it all seems to point right back to
[30:07] Google's ownership in Anthropic and SpaceX. It's all a big circle jerk. Not good. NOT GOOD AT ALL. UH SOMETHING'S PRETTY weird about that. Now, if I go to
[30:19] the other portions of the earnings call, just to see if there's anything else that's interesting. Yes, I do think it is exciting to hear that maybe we're going to have a launch cadence for rockets that will get to at least one
[30:31] flight per day. It's worth noting that their cadence in 2025 was about one flight every other day, but you know, I I generally am going to frames. So, if he says we're going to get to one launch every day, that might
[30:48] be years from now, uh, in fairness, by the 2030s, there are estimates that per day, and you'll end up having thousands of launches per year. The issue here is you have two types of rockets. You have the Starship, which is
[31:01] on test flight 14 will be the next one after they, you know, crane the Starship 13 out of the ocean. Um, you know, that's that's a test vehicle. Elon says optimistic about the heat shielding technology they have, which is great.
[31:17] But Falcon is no longer taking orders. The Falcon rocket will be flown through business because they only want to do Starship, which will take larger payloads. And Falcon's only going to do 70 flights at an estimate of 2028. Well,
[31:33] ready by then in certainly not in in mass flight production. This is why need four launch pads built for this. Um, and you know, these these plans are all in works, but uh in the works, but uh usually these things take a whole lot
[31:49] longer. And uh I think a lot unfortunately of their capex isn't actually going to the Starship or the launch side of the business. As we saw, it's going to AI and I'm skeptical that that is a
[32:03] and I think there are reasons for that. It did not sink, by the way. It is being towed right now. So, let's delete that. The uh Starship 13. I do think it's cool that they're going to release V3 satellites for Starlink, which is
[32:17] supposed to make the service 10x faster, which are is already really fast. So, that's pretty impressive. So, I'm not sure exactly by what levels they're counting that, but um this is this is sort of my overview of how I feel about
[32:31] sort of my overview of how I feel about SpaceX. Now, when SpaceX first IPOed, before it IPOed, we called for maybe a 30% runup and then a bleed out. And so far, that's exactly what we've had. This tripling down on AI Compute makes
[32:47] the stock even more risky for me given that I don't have good visibility into that I don't have good visibility into what realistic contract rates are. And I honestly think we're being misled. Thanks so much for watching, folks. So,
[33:00] at meetke.com. Pay once, you get lifetime access to the trade alerts, buying, selling, fundamental analysis, you name it. The courses on building it off on your taxes and we'll see you in the next one. Goodbye and good luck.
[33:14] told us here? I feel like nobody else knows about this. see how it goes. >> Congratulations, man. You have done so you. >> Kevin Pra there, financial analyst and
[33:26] YouTuber. Meet Kevin. Always great to get your take.