看多 TSLA：合并应能提振其股价，隐含溢价至少 30%（约 500 至 550，若达到 500 则看 650），交易将在 18 个月内完成；Cybercab 的增长，以及将能源和汽车业务转化为算力吉瓦，可能使盈利/现金流增长十倍，而现货算力价格是该论点的风险。

Right before the Tesla earnings call, I don't know if it had anything to do with it. definitely didn't have anything to do with the audio, but he was threatened by two left-wing nut jobs.

Um do you think a lot there was a lot of criticism about the Tesla earnings call?

Tesla calls are always a little depressing. In the end, the stock will move mostly because of Cybercap. I think that's what will drive the stock, but there's one more thing now in the stock, not Optimus because it's the same thing. you have to wait until it happens.

Now, the new driver is the commentary on the merger which happened on the SpaceX call. I think that sent a signal into the market and I think that will carry Tesla stock upwards because it's also clear in my opinion the price will be at least 30% above what it is now.

If Tesla goes up a lot, it will maintain a 30% premium. Otherwise, you can't do this. Which means we talking 500 to 550 from here on. If the stock goes to 500, we talking about 650. that is required to to get a yes vote on Tesla and there can be no tricks.

This will be the most uh litigated merger in history. but that wasn't priced in before. So now basically from where we are at wherever we go you always get a 30% implicit merger premium depending where the market price is the timing of this thing but everyone knows it has to happen in the next 18 months guaranteed in my opinion

But the problem is if if the Tesla price is at 350 we want 40%. So the the Tesla investor has their anchor point of the minimum that they would accept as long as they get to the upside.

I think it'll close somewhere around three SpaceX shares per Tesla share. That assumes the price are in the region of where they are today. the longer they drag their heels and if SpaceX does pull off their 10 gawatt of extra compute next year and we'll get we'll get to comput bonkers that business is the more they screw Tesla shareholders.

And don't forget there's also something called Tesla. So what happens if we actually get to 50,000 cyber caps next year? Of course 10 gawatt are much much more money than 50,000 cyber caps, but the multiple on the cyber caps will be much much higher in my opinion.

immediately I think Tesla, despite the terrestrial comput, and we'll get back to that in a minute, I think the Tesla Cyber Cab uh business is extremely interesting. And now they're what they're doing, 80,000 paid miles, is it a day at about 250 a mile, which is way higher than I expected.

And they're growing by 10% a week, which means we could get to a million miles by the end of the year.

and the satellite will be 100% guaranteed ready cuz Nvidia and Tesla they're working uh Nvidia and SpaceX they're working on on the orbital compute rack you know the Vera Rubin 72 rack for orbital compute scheduled for Q4 2027 I think these teams deliver stuff in time Nvidia and SpaceX so they will get this thing ready

We know they're crushing it. They're going to be adding a mini hard what, every 30 days till the end of the year, which will take us to the exact number? 2.7 gawatt approximately by the end of the year.

And 10 gawatt next year. Now, if they're monetizing those at $50 billion a gigawatt, that's a lot of money. A lot of money. So, do they even need space?

Yes, they do. because my prediction is first of all I still think that's a little too bullish but I started to model it in now because Elon I mean he was very confident so I have 12.5 gawatt live end of next year in the model let's see I find it a little crazy but they seem to be extremely confident about that crazy and they're buying what what was also crazy too that we heard this week is Elon he sees the bottleneck he sees cuz he's in bed with Jensen at Nvidia he sees all the chips he wants.

He's got all the money he wants. He's got the land. He knows how to build. His bottleneck is cooling and was it cooling and wiring are the two components where he is actually buying and securing and owning 20 gaws worth of cooling for next year.

And that's a tell right there because if he's ramping so much into the cooling, it means that's his concern. He's got everything else licked.

Now he's making blades. Then he figured out he drilled down and said what is the problem with the turbine manufacturers? They said dudes we can't get the blades and no one can manufacture the blades. only three companies on earth and then Elon said fine that's not very hard now they are manufacturing the blades shipping it to the turbine people and then get the turbine so that was genius

and and and they're using their Giga cast technology and capability to manufacture these blades correct yes exactly they use the the rocket engine casting technology to actually cast the blades so that's genius

Chips absolutely no problem. Memory no problem. They're locked it in with Nvidia, with Micron. They know how to build these things better than anyone else. They have figured out the cooling and secured it.

They have figured out the turbines. They can do everything. And then it comes down to this last question. And they have some form of hack that I don't understand to be frank because this is like a non-trivial political land grabbing literally land grabbing hack. And I don't know what it is,

Last year he did some illegal turbines or something. That was his first hack. Now he knows, okay, I can't do that anymore. I need to get more serious. I need to get permits. He is hacking the permits.

But every time you hack something like that, the door closes.

versus gigawatt compute whether it's on earth or not is the big thing because I would argue when I see the ramp of cybercab I know people say it's nothing but what we did pick up from the call if you listen very carefully to the CFO and by the way people criticize that CFO for having a thick Indian accent remember he somehow saved 5 billion of capex so he might be a financial genius that he doesn't get credit for like

But people people slam him and maybe you know Elon doesn't suffer fools. This guy this guy could be the most incredible CFO you've ever seen but just doesn't sound like it. And I just want to just defend the poor man there on this one.

Let's talk about compute demand. And we'll get to this in a minute. I want to talk about gigawatt terrestrial compute versus cyber cabs. Now, it's kind of a longer story, but you said a gawatt of compute is worth about a million cyber cabs.

I don't agree with that. And this is kind of rough. you know what you need, how many GPUs you need, the cost of them, the rental price per hour, the revenue you get, etc. And what that translates to in terms of cyber cabs, you guys can pause it, copy this, and I can share the notes in it later as well.

But this is where it gets interesting. If you can, and it could be, by the way, the gross profit per year per cyber cab might be 20,000, might be 15, might be 30. But when I see the latest numbers of the miles each car is doing and they're charging 250 a mile,

Yeah. So it's the demand for these things is off the hook. Like he's created it's like um do you ever see a fairground with novelty rides? I remember uh when I was a kid, they had these things called called bumper cars, you know, with the electric thing on the roof and you ride around, you crash into each other.

These things have become almost like a fairground attraction. People are flying to Austin to take a cyber cab ride and then they're tick-tocking it and putting their

But I think so. James, I give it to you. If they get to 250 miles at $250, of course that's crazy. Then you're not at 30,000 by the way. You're at something significantly higher.

Oh yeah. Well, well that's just because of the novelty right now and the premium surge rates and that's going to wear off in time.

I'm pushing back on. So on that side I you're right. I'm pushing back on the gigawatt profitability. I think it's more profitable. You had 11 billion per gigawatt.

That's the big factor. I assume these things only last five years. Yeah. And the capex of these things is very expensive. So that's that's the piece that you don't So a cyber cab can do a million miles. They said that.

but you have you have roughly 50 billion also in capex. Yeah. So you have 10 billion capex depreciation but it's spit the problem is the cyber cap spits off so much money compared to the cost of manufacturer which terminal cost to manufacture is 15,000.

If this true I agree I agree on the cyber cap side. I think I'm roughly on the same numbers only. I would even have high if these numbers hold up which I would doubt but I definitely agree with 30 30 billion but

I think compute gets a similar amount that's that's the difference you because compute the operating expenses are not that high so you can basically spend 10 billion operating expenses and then 10 billion on capex depreciation then you're getting to 30 billion

well we need to we need to hash this up because per my calculations and the depreciation before depre appreciation by the way it gets even more extreme. I think it's more around 300 400,000 cyber cabs is equivalent to a gigawatt of compute.

But the other question we have is how much do we get per gigawatt of compute which I think after expenses easy 15 billion 11 billion etc. after depreciation depreciation is a big factor here as well.

Um either way they're both very lucrative and I would I would say it is I would say it's in the realm of the possible but I wouldn't go beyond that. I would say it's in realm of the possible that 500,000 cyber caps are 1 gawatt.

if they really get that, you know, because I had pretty aggressive pessimistic assumptions about what they get paid per mile and how many miles they can drive. So I think they can get higher. I think you're too low here on profit.

Is this on a giga or what? No. What is this revenue per? This is the this is the cost of a gigawatt of AI compute. The number of GPUs required assumes 85% utilization. Cash and power is 15% of the variable costs.

Revenue it pulls in 11 billion. Of course, it's a lot higher.

I think that's totally off. I think you're talking about 40 billion, not 11.

Because think about they had half a gigawatt and they're making 26 billion. So I think on the cyber cap, yes, I might be into conservative views is pretty accurate. But here I think that is the dramatic insight of Elon that he sees, are you kidding me?

I'm spending 50 billion in capex and I'm making this back within a year in revenue.

And that's like people doubt that this will continue, but he said I don't just think it will continue, it will go up. Exactly. That's that's what I've been looking at because everywhere I look at like you look at all the coreweave disclosures and all the neocoscalers hyperscalers etc. I mean the hyperscalers to start with those guys.

None of them have compute. Microsoft, Google, Amazon, absolute monsters. They are like when I saw Google buying compute from Elon, I was like, are you kidding me? Really?

I mean, they're toast. They they they need compute. There is no option.

Look at my favorite new company. OM has like the future trading. So they created these compute futures where you can buy futures on H100 per hour cost on B200, B300 on all the GPUs of Nvidia and you see the prices and H100 is now higher than six years ago.

Yep. Exactly. They're going up. There's massive constraints and then you look at the the Neocalers, they the NeoClouds, whatever they call them, they cannot ramp. So Coreweave, it used to be two and a half years to make a data center.

Now they're looking at four years and Elon's kicking them out in what 12 weeks.

Yeah, exactly. I mean, it's that is also why I'm a little bearish or cautious or not invested in the NEO clouds. I think people get it all wrong. There are two types of people.

The bullish people who say there's enough demand. The bearish people, they say it's not enough of demand. I'm the third type. I'm saying like, "Dude, are you kidding me? Of course there's more than enough demand. The problem is they can't build them."

And every year they're delaying this. They're paying5 billion dollars in interest. Exactly. And they're at 15% at 15. Exactly. And imagine if you look at the business, they cannot afford to lose 5 billion a year.

So that's the problem. It's an exec. It's a supply build out problem for them to build it out fast enough. Every year you are delayed. You It cost you five billion.

And this is where like I I'm looking at Elon's edge on the compute side and it is a horrifying story if you are not him and you're trying to build coherent compute and also I heard a rumor back in July from a very valid source that said he has locked up 40% of the Vera Rubin supply. 40% one guy like if you if you if you Joe wanted to buy a Vera Rubin chip 3700 you can't get it if the problem is not buying one.

Yeah, exactly. So that's like if I go to Yans and say I want to buy 600,000 maybe he gives me also priority. Okay. So he's got So let's let's let's talk about the compounding of the edge or the moat or the competitive advantage he has that I see he has power behind the meter building turbine blades.

He's got a team of rocket scientists and I remember him saying, "Oh, building rockets is hard. Building data centers is easy, but it's still hard." But in relative terms, he's got a crack team of SWAT team of rocket scientists putting these data centers together for that coherent compute.

He's got coherent compute at scale. Nobody's ever threaded together more than 30,000 GPUs. He's doing 300,000 at a time. Bonkers, bonkers math. And then you have the other factor of speed to market and the ROI payback in less than 12 months. How can anybody compete with this?

You put all those compounding factors together. This is not this is not anymore. We are living in interesting times because this is not about competing. It's not about demand. It's not any of the conventional things are out of the window.

We in a situation it's like COVID a little bit. I was in COVID I had a genomics company and then people came to us and said can you do testing there was no issue about competition.

There was no issue about de uh demand. There was no issue about there's only one question. Can you do this Can you I need this next week I pay whatever you want.

And that's exactly the situation with AI. No, there is no competition. Everyone is just can you do this? Can you stand up my compute? If yes, here's all the money you want. Unlimited money.

And for Yensen, he's on the other side of the equation says, "Dude, I I am perfectly run company. I can manufacture millions of uh Vera Rubins next year, but how can I be sure that you little guys are going to buy all my Vera Rubins?"

And then Neoclouds comes like, "Oh, I promise, but you know, here's a clause and here." And then he says, "I don't believe you." Then Elon comes along like you have seen what we have done last week.

Can I reserve a million V where Rubans? He says sure you get them first. You get everything you want because you're Elon.

So you know this is I think that's the situation. I think the market gets it wrong. They think in old paradigms. They think like oh but what if there's no demand? What if the financing falls?

We're like dude this is not the problem. The problem is can you do this stuff?

In fact there's another little point here. There is a new element. Uh this is the agentic explosion. It's estimated that there could be a 10x increase in the demand for CPUs, not GPUs, which is good for the AMDs of the world, etc. because of the agentic explosion.

Nobody has any idea. There will be billions of AI agents running. You know, you probably have 200 of them running just an individual and it's early days yet for you. So the CPU market is going to explode and the inference and I said years ago in a video I said the future of compute is going to be like 99% inference not training or learning inference inference inference and then we years ago we we spoke about maybe inference in the car inference in the Tesla inference in the cyber cab there is an interesting element of the cyber cap by the way and the amount of cooling that it has way more than that's required for air conditioning.

A lot of people have a theory that the heavy cooling is actually for the GPUs. If they are charging and running inference at the same time, then you got the mega blocks and all that stuff.

This is where the world gets really crazy and nobody is ready for that.

So, it goes back to compute. Do you have enough compute? Who is positioned to build the most compute? Nobody. And then you have the other angle of spot market. It's it's a variable market with infinite demand and no supply.

Also, by the way, the interesting thing is when you think about this whole thing, I did the math on it now, uh, when you build out what was it again. If you build out 10 gawatt, that's roughly 600,000 Vera Rubins, right?

600,000 Vera Rubin GPUs, which is enormous amount of compute. But think about that. How many cyber caps will you have? Millions.

Okay. How strong is AI5 or whatever is in there? You know, for inference, probably not that bad. It's probably not that much weaker than Ver Rubin's. I don't know the exact ratio.

Let's say it's 2 to1 where Ruben to that thing. So, you are actually building out insane amounts of compute. You know, now it's it's crystallizing more my mind what how this translates.

Like basically, it could be that if you have 1.2 2 million cyber caps. It's the equivalent of 10 gawatt of inference compute, which is pretty stunning if that's true because that's an amount of money for Tesla.

You can't even fathom. I This is to I'm totally winging this because I don't know the exact math that's so easy to just push into these cars. But I think it is kind of.

And I don't know the translation ratio how much this compares to Ver Rubin, but they also have new chips and in there. If you do the math on that and and think about oh what happens if these things actually also become compute nodes even if they make only like if you run the math on top of it a million cyber caps a gigawatt of compute.

Yeah, but what if a if a million cyber are also 10 gawatt of compute?

Then it becomes crazy. And then the other thing is when you when you look at the supercharger infrastructure around the world, there's like 75,000 of them right now. Uh if you have a Tesla and you rock up, you could go to a Tesla Supercharger at midnight to 5:00 a.m. There is nobody there.

Nobody. It's tons of capacity. Afternoons, maybe three out of 18 stalls are full. And they have this mega block concept. They got the cyber cabs going to be charging there probably at night.

And then you've got the connectivity with Starlink. The synergy is insane.

Let's talk about uh mega packs for a second because I wrote a paper three or four years ago saying the mega business alone is going to be worth $1.1 trillion market cap. We found out on the earnings call.

It's already there because they have 130 to 140 gawatt of capacity. They sell them for a revenue of about 33 billion conservatively at 33% margin. That's 11 billion in profit. You multiply that by 100 PE, that's 1.1 trillion market cap value of a business in a company that's trading at 1.5 trillion.

It's bonkers. The mega pack business is always ignored because battery storage isn't very exciting. But you connect the battery storage with some chips for inference compute wired up at Starlink.

That that that's that's you could scale that out to beyond what the neoscalers can do very quickly.

The question is how much free energy is in these batteries if they're deployed, but there is some percentage you can use of that. And um I think Elon hinted at this stuff because he was dropping something like distributed compute.

And I I don't know what that means. I don't know if they're seriously considering doing that already next year. Probably that could be his hack. Um but uh who knows? I don't know.

I don't know. But I think this distributed compute I I'm not an expert on that. Is it is like sufficient? you know, can you how big need to be the compute clusters for inference?

But the whole point with these satellites is we know 240 kilowatts do the trick, right? These these satellites will be small. They will be like a quarter megawatt. So you need 4,000 of them for a gigawatt and they they are going to be distributed.

They have laser links, but that is way too slow. That's not a cohesive cluster, but you don't need that for inference, which means you can do it with mega packs and cyber caps.

So I do think that is the one story that now that I understand more and more the economics the crazy economics of spot compute uh that no one I think understands about Tesla once you take Tesla and translate everything they do in energy and then cars into gigawatts and then take any share of that even if it's 10% and turn into compute gigawatts you can't even imagine what happens next. This thing is like

I mean you're basically you know whatever tenfolding the earnings and cash flow of Tesla. What's crazy is because we know GPUs are like uh GPUs for inference is like smashing an ant with an with a hammer.

It's overkill. CPUs do it very effectively. But when you look at the architecture of AI5, even AI4, it's baked for inference from the ground up. extremely powerful, extremely smart, very and we shouldn't forget that AI5 is ready probably when it's going to be delivered in April.

April Q2. So, it's going to be very interesting. What if they actually package that with, you know, mega pack and everything? Do we know? Can they swap out, imagine they build, I don't know, 200,000 cyber cabs. Can they swap out the AI4 for the AI5 easily?

I think they can because I know I I had a Tesla Model X that I bought in 2019. I still have it, but they upgraded the computer for me. So, so I could really leverage the latest FSD.

And I think the the bottom line is with this whole situation, I think if you believe that compute is going to be the hot stuff for the next 10 years, I mean, they just cornered everything in space on Tesla in in micro hard like all over the place.

Yeah, it's crazy. And that's why, you know, I think Elon, he sometimes lets things slip because he's very, if people think he is a dishonest person or a bad actor or a liar, but just look at what he's achieved.

But I do believe he believes in truth and I do believe he lets truth slip through his lips and he's very open. He's almost like what do you call those people who sometimes curse without being able to control their mouth.

He just he just spits out breadcrumbs sometimes without being able to control his mouth on Twitter.

You know, when people talk 10 years out, 5 years out, a year and a half out, you can't talk about SpaceX in isolation because Tesla and SpaceX will be together. I think we all agree on that.

But then Elon said there's a ma in unless there's a major world war, it's way more than 100 trillion in less than 10 years. And I thought that was a very profound little nugget of information.

That's before the SpaceX call where he was also extremely bombastic about compute projections. And again, maybe he's factoring in these compute projections in the event of an acquisition of Tesla and also Mega Blocks and also in CA in CyberCab inference as well all connected together. We don't know. We do not know,

but when I model it, like when I model it out, I don't see any problems getting to 450 billion next year if they get to 10 gigawatt buildout because it's not, you know, they're ramping it up over time.

That's why it's not like a trillion or something.

and he leaked that as well in the call. He said half a half a trillion to a trillion dollars in revenue and like the CFO is having so even if he's say 50% wrong, 80% wrong, these numbers are still huge. And that's what

I mean it's completely nuts for especially when you look at the largest companies on earth like Amazon, Walmart in terms of revenue they have bad revenue. This is super low margin revenue.

Here we are talking about 50% margins and something like that 60%. So it's insane.

So yeah I think everything in this whole thesis comes down we know he can build this stuff. He know we know there is no one like Elon in terms of buildout and hacking all these problems.

Everything just comes down in the thesis risk to spot compute prices. Is it a bubble? Is it not a bubble? Will these prices hold? Will they go down? Will they crash? Or will they go up?

Robboaxi and Cyber Cab are doing more paid miles now than Whimo in Austin. So that tipping point has already happened and it's still early. So that's impressive. I think they're just really focusing on safety.

Got now 3.3 million miles, zero accidents, which is massive already. So, the safety is kind of blicked, but it's in that area, Austin. Once they start rolling out other cities in Dallas and Houston, I think those miles are going to go very vertical.

And I think what's like what's really holding Tesla back uh with the cyber cap situation is people just have no good feeling about like or intuition about how fast this can go.

They say like, well, it's going to take forever. Everyone is in this like whatever. We get a 100 per month for in perpetuity. this is never going to do anything.

That's the big question. When is that point where they can just deploy thousands and thousands and at that point we go nuts the stock? And the question is, will that be end of next year or will it be in December?

Well, you and I are actually almost bang on line in terms of the tipping point for Wall Street to wake up to Cyber Cab and that is 2000. Yes, we could easily get there by year end.

We could, but we could also not. Like that's the problem. Like that's very hard to predict, but I believe we get close to that number.

Cipher sniper Sean D&D Tesla K8.

If you in these times, if you just say, let's make some money with Tesla, you need to also watch your surroundings, right? Because if you make 10x with Tesla and then you end up in, you know, democratic socialism of America, guess what happens to your money?