Most investors who want to get exposure

in AI think it's all about the models.

Open AAI, Anthropic, Google each have

their own models. But what if the real

winner isn't the company with the best

model? What if it's the company that

owns the entire stack? That's the

argument laid out in this incredible

SpaceX analysis by Bran Wang. Think

about it. Tesla provides the energy,

batteries, transformers, mega packs,

solar chips, vehicles, robots, and real

world data. SpaceX AI provides the

models, reasoning, digital workers,

payments, and software layer. And of

course, let's not forget the launch

system, Starling connectivity, AI

infrastructure, orbital compute,

satellite communications, and eventually

AI data centers in space. But here's the

thing. Brian believes that AI could

become a majority of SpaceX revenue as

early as 2028 with AI infrastructure,

Starlink mobile, digital optimist,

enterprise software, and orbital compute

all feeding the same ecosystem. In other

words, Elon is building a vertically

integrated AI civilization stack. And

wherever the network effects and pricing

power ultimately accumulate, Elon may

already own the entire value chain.

Brian is a futurist thought leader and a

popular science blogger with 1 million

readers per month. Check out his YouTube

channel at Next Big Future and his blog

next big future.com is ranked number one

science news blog. Welcome Brian.

>> Great to be here. Exciting exciting

times. A lot happening.

It is it here we are. A lot of the shows

we've been doing was preparing for the

SpaceX IPO and this is a great analysis

you put together. You called it the

SpaceX IPO AI in space. We we obviously

Elon was being interviewed several times

this week and uh he shared a lot of new

information and generally it is faster,

bigger, more efficient than what

everyone else was uh forecasting. And so

now with those new numbers, I believe

you created a deck here to tell us what

it all means. Right.

>> Right. So in the picture we see there we

have details of the um AI1 uh

satellites. So they will make more

versions of this. They will improve it

beyond this first version. This first

version they will try to get launched

next year. So in the interviews Elon

indicated a goal was to get one gigawatt

of these satellites up in 2027. So that

is a huge statement because I can you

can take there's a number in the bottom

left 70 kilowatts per ton, right? So 70

kilowatts per ton if you want to get one

gigawatt

that means you have to launch like um I

think 13,000 tons or something like

that. So basically you need to have 130

launches of 100 ton payloads in order to

get to one gigawatt, right? So that's

the ballpark of what they'll do next

year. And then each year thereafter,

each 18 months thereafter, they'll be

trying to 10x that.

>> That's the next year.

>> So they go from like 130, right? 30 a

year after that or 18 months after that

1300 the year after that 13,000. So that

that's the the pace which they want to

go and if things get slower that is the

direction that they are going in.

>> That's crazy.

>> Um

>> let's go through the revenue. You got

some more big important slides here.

>> Yeah. Yeah. So 2.5 more revenue. So 2.5

more revenue um than 2026 already. AI

majority only rented 40% of Colossus 2

at the 26 billion dollars. So they still

have another percentage. So Elon said,

you know, he didn't like um um Aaron's

numbers. So he said

too conservative. You're discounting out

to 240, nowhere near our goals. So here

is the numbers where roughly we hit the

goals of 1 gigawatt in 2027, 10 gawatt

in 2028, 100 gawatt, sorry, in 2029. Um,

but I only have the rental line for

space AI for 40% because I'm assuming

that I'm using more of the data center

for the cursor gro and the digital

optimus,

>> right? So, but then we hit to basically

two trillion, 1 trillion, 3 trillion in

2029 and 1 trillion.

>> Where is that? Where's the 1 trillion?

Right. Uh what what do you mean by one

trillion? Market cap or or revenue?

>> What?

>> Revenue?

>> So, you're thinking 1 trillion of

revenue by 2028

>> and then 7 trillion by 2029,

>> right? So this is like if Elon hits the

goals

>> 1 10 100 right because Elon was

complaining he didn't he you know no one

was give me the goals so I said okay

>> Elon you say your goals are this if you

hit these goals and if cursor Grock is

growing every four months right then we

hit these numbers which is

>> more revenue more profit than the top 20

companies today

>> combined the top 20 companies today is

5.

>> You're talking more than Nvidia and

Google and Amazon added up together

>> in four years. There's a potential if he

hits his goal.

>> This is too much. You're telling me that

he's going to be making 7 trillion per

year in revenue

by 2029.

>> That's if he hits his goals,

>> right?

if his goals versus the companies today

and the the other companies today I

think will drag up and and will have

more but you know if he hits his goals

you know versus all the other companies

20 top 20 you know your mag 10 and 20 or

whatever then you hit that number and

the profitability I'm assuming 70%

profitability because it' be so

profitable in this stuff because the AI

is profitable space is profitable then

he has $4 trillion of of ibida which um

the ibida of all these companies now is

about like 1.2 trillion.

>> I mean,

>> so he'd be three times.

>> So you're saying 3,000 three three

trillion in rental space-based AI.

That's crazy number. But then the three

trillion in digital Optimus and Grock,

which is basically a dig it's AI SpaceX

AI part of it. And that part I I it's

hard for me to agree on that part.

>> Yeah. Well, think is I I one thing to

get you should agree. I'm this is me

answering Elon's complaint that no one

really my next slide

>> my next slide I downgrade it by you know

you know I downgrade it by 80%. you know

to only one trillion in 2029 but you

know if it's 1 trillion 2030 whatever

right I downgrade it where it's like he

only gets you know like you know five

gigawatts or four gigawatts up in 2029

right so things take longer the 5

gigawatts takes you know like you know

3,000 launches right over you know a few

years so I downgrade that and then I

downgrade the cursor revenue and then I

the AI rental revenue

is physical stuff on here. So, I don't

upgrade that much. Starlink is basically

him getting, you know, the the 200

million dishes out there, right? And the

DTC direct to cell phone is him getting

the B3 satellites up there. So, this is

my more reasoned thing, but because

Elon's complaining that, hey, no one's

run the numbers, I ran the numbers. So,

my previous this is my I think they

should be able

>> This is crazy numbers. So, first of all,

I just did a quick look. Apple revenue

is 451 billion in annual revenue per

year. Nvidia is a 200 million um 200

billion. So, so you're talking,

you know, Apple size revenue by 2028 and

then like blowing everybody away by 202

>> and like like four or five revenue

streams that are invidious scale today.

>> Wow.

>> Right.

>> I guess so. I don't know, man.

Well, the thing is even if I zero if you

zero out space AI, zero out digital

Optimus, those things aren't happening.

It zeros them out, right?

>> Then just on Starlink and DTC and AI

rental,

>> I'm at 500 billion. I'm as bigger than

Apple from that stuff. And then the

launch is guaranteed. The launch thing

is they're the US government is going to

do uh Golden Dome. They already have $6

billion worth of contracts for Golden

Dome, right? Golden Dome being gonna not

have all these um you know stop missile

stop drones and stop stuff, right? They

plan to spend $185 billion over the next

three years on Golden Dome.

>> Who do they give the money to on that?

There's no one else can give it to on

launch. No one else can give it to on on

salary production, right? So at least

oneird if not one half of 185 billion.

So that would be line number one, space

enabled solutions, right? So they're

already at $6 billion this year. How do

they not go to 20 $30 billion per year

if all the Golden Dome contracts come to

them? And then there's NASA contracts.

>> Well, it's not just the launch because

Golden Dome will want Starshield, which

is a version of Starlink.

>> They'll want Starlink Mobile, the Star

Shield Mobile, a version of that. And

they want AI data centers because you

don't want the US government's, you

know, AI infrastructure in in a data

center on Earth that they can easily be

blown up by launch a rocket. And you

need,

>> right,

>> you need to also Yeah. just just

and and stuff like um they already have

like 89% of all satellites, all ground

stations, all whatever. So if I go up

you know 10 100 times they have 99% of

all ground stations 99% of all other

space all ground infrastructure right so

again there's the monopoly is complete

there's no one else to go to to get this

stuff if you try and give it to someone

else no one at scale right so that money

comes to SpaceX and I you know again I

didn't even push that hard I think the

space enabled solutions should go to

like 30 50 And thing is the

Congressional Budget Office says that'll

be $1 trillion or more over 20 years to

do the Golden Dome, right? But the thing

is whatever

um US government project landed on

budget and they're going to spend way

more than $1 trillion on on the Golden

Dome stuff and then every other country

will need a golden dome, right? So

anyway, so this is again somewhat

conserved. So then based on the total

addressable markets, they're getting to

significant fractions of this. If the

the Elon plan happens and they get to

like 7 trillion, they're at 25% of these

tamps. So the reason that they're

rolling out these TAMs is they're

actually going for this. They're going

for the these TAMs, right? So even in my

more conservative case, you're at, you

know, 5%, you know, 8% of these TAMs,

right? So the thing is they want these

TAMs in the next 10 years,

right? So So again, people who think,

well, I just dismissed these TAMs. Elon

just saying, I'm deadly serious.

>> Yeah. I mean, I don't know. It's cra I

mean, like, you know, if you just if you

didn't have this trillion dollar thing

here, people, you wouldn't be able to

say SpaceX is worth two trillion today.

So that's why, you know, they threw some

massive number. But you're saying, hey,

there's some real reality here. This is

not some madeup number. This is

>> right. this is road map going towards.

>> It's like, you know, he didn't just make

it up because he wants to,

>> you know, pump up the IPO. He This is

like actual bottom up

>> enterprise AI could be worth this much.

>> It's great.

>> Right. Right. And you know, Starlink

broadband saying, you know, again, 200

billion, you know, Starlink mobile 200

billion and then they expand and grow

that they can get to those numbers. So,

and they totally own that a

infrastructure.

>> They're building it and they have a big

chunk of it. U consumer subscriptions

that's X and XAI the XA XAI part you

know which we just dis dismissed if I

have three billion five billion

smartphones with my communications

they will be using first thing they see

will be X and X payments right so

they'll have three million they'll be in

the Google meta range

>> and and and given what we our discussion

on the on the Starlink you know,

satellites and all that. I have no doubt

Tesla is going to be making many

different form factors of bots, not just

humanoid. And there's going and and

SpaceX is likely going to make many

devices. Not a phone, it's an AI device.

So he goes, I'm not making a phone, but

he's making AI device, which is a slab,

a necklace, a wrist thing, glasses,

everything will be connected. We'll have

instant AI access. Trust me, believe

this. They are doing this. So, anybody

who says, "Yes, he's not going to create

a phone, but it's like it's going to be,

you know, the Elon the Steve Jobs thing.

It's a phone. It's an internet access."

>> Yeah.

>> Yeah.

>> It's a music player. No, guys. This is

an AI device.

>> Every form factor.

>> Okay. So, so then other things about

detail. So, this goes to the master

orbit target, right? So again, the

master orbit target in 2028,

it's looking like about 200,000 tons to

orbit, right? So that would need 1,200

ton launches or 2,100 ton launches

there. And that's like the year after

next year.

>> So how how many flights does that mean?

So how many flights start

>> that? That's I think they convert to I

think they'll be converting to the 200

ton payload version four rockets in

2027. So that means 200,000 would need

uh 1,000 launches.

>> 1,000 Starship launches of version four.

If they haven't switched to version

four, you need 2,000 launches of the

Starship. They're going to try and

launch um they try and launch like a

couple weeks ago and it will try and

launch again. Seems a little aggressive

to me because if you're talking uh

>> how many how many did you say a thousand

flights?

>> A thousand is three a day.

>> That seems crazy for 2028.

>> So they have four launch towers that

they're working on, right? They have

three that will basically be working

next year and the fourth one sometime in

next year or earlier. Um, so the money

raised, this goes to they need more

launch towers. Like in if you think

about the launches, you could get to,

you know, a launch per day. People see

that they're already at like a launch

every two days right now at 170. So

Falcon 9, they're launch every two days,

right? And they have like two main

launch facilities. They have um

Cape Canaveral and they have Vandenberg,

right? So um so if when you think about

this is how many launch towers they have

and the the more they want if I wanted

to do two launches per day then I need

to have the window for the uh shutdown

period where I'm messing up air traffic.

I need to have that shortened up to 30

minutes or less which is what they've

achieved with Falcon 9. they have a

short launch window, but basically they

say, "Okay, everyone stop flying for the

next 10 minutes and then 20 minutes the

rocket's gone and then they open the

airspace up again, right? So there

there's a bunch of things around that.

The other thing would be that so

basically to achieve this I think they

would need um about four new launch

towers uh four new launch sites and and

launch towers and ideally ones away from

air traffic because um you know if I

have less air traffic then I can launch

like four eight times per day less issue

because I don't have to stop air

traffic. It's just like can I get them

going? But then I probably would need to

have four to six launch towers at that

launch site because I want to stack I'm

stacking 5,000 tons worth of ship with

fuel on each one and I'm filling it with

200 tons worth of of satellite. So I

probably have a line of them ready to

go. Like this one's ready to go. Next

one ready to go. Next one ready to go.

Next to go launch, four hours later

launch. Launch board later. So the

logistics of doing this involve it's

like you're creating the um the

shipyards where you have big cranes

lifting up um you know 40 ton 80 ton um

rail cars you know onto ships where you

got like the mega ship with 18,000

things. Those are the kind of operations

we're looking at right like shipyard

containers of crap moving in.

>> It's not as hard as people think.

Famous words of course not. It is very

hard right? Uh but it the hard part is

the reusable rocket and they've they're

on their way and then from now on is

scale. Right. So Okay.

>> Right.

>> Mhm.

>> So when you payload up like all those

numbers convert basically divide them by

200. That's the number of launches in

each year. And so that means out in 2029

they're looking 10,000 12,000 13,000. So

they're going beyond that. That's their

their plan. And then megawws per

Starship. and think about how many

megawatts, how many kilowatts per ton

and then figure out I have 100 tons or

200 tons and then that means how many

megawws in order to get my gigawatt 130

launches 13,000 launches 10,000 launches

for u you know 100 gigawatts um and then

yeah AI half by the the the back half of

2027 and more than half of sorry I has a

spelling mistakes there um so AI is

actually already about half of their

revenue because people keep ignoring the

anthropic Google deal which is $26

billion per year. So them making two and

a2.2 billion per month starting from

October like $1.2 billion now and then

going to 2.2 2 billion means that they

will have you know 12 15 billion dollars

in the back half and that does not

include cursor and the the the data

they're doing with cursor is going to be

bought for 2 to 3% of the shares of

SpaceX

>> so that will happen they will own that

>> cursor has doubled revenue

>> in the last four months I saw

>> two billion in just the last month

>> if you keep d

>> yeah if you keep doubling every four

months that means in In 12 months you go

up eight times.

>> So then they'll be at 36 32 billion per

year by this time.

>> Yeah. Could be. It's a fast follower

toropic. Yeah. Yeah.

>> Brennanthropic. That's right. So we can

go to the next slide. So then they show

details about like what the the details

of the AI salad will do. So you have in

the left these three racks of of things

and they may have different number of

racks. They're just saying they're gonna

put multiple racks into each module that

will attach to the body of it. So they

have cooling and separation of these

things. So then this the larger view is

the one on the right where you didn't

zoom in. The the line of things in the

center body between the solar panels is

where all these racks of of AI are. So

you just have these little blades kind

of popping in modules of it with cooling

and with all the AI just kind of sitting

there just powered up on it.

>> And the thing is they can show details

about this because no one else can do it

at scale.

>> No one can do it. So it literally is an

AI uh server instead of an inside of a

ware factory.

These are just racks. They're literally

just racks. It's so simple. Look how

simple that is.

>> Shaming the

>> Yeah.

>> Right.

>> And the thing is they can show all the

detail like like they won't show the

detail of the Tesla bot because figure

might or or you know unitry might make

their their things but they can show

this cuz no one else can do it. They can

say I'm going to show you everything.

I'm going to open the curtain behind the

Wizard of Oz because there's nothing you

can do. I

>> I disagree. I think the Chinese

obviously are going to copy this

instantly and they're going to be up in

space and they'll be competing with AI

data centers in space. There's no doubt

about that.

>> So, but they can't do it at the scale

like the the 100,000 tons because

because

>> right last year 4,000 tons was launched

to space. 3,000 tons was was SpaceX. So,

the rest is China at 500 tons. But

unless China had the fully reusable

rocket to go from 4,000 tons to 500,000

tons in three years,

>> they're not going to get it done in

three years. Yeah, they'll be a a slow

follower, but they'll be the first one

there.

>> Okay.

>> Right. Right. They'll be the first one

there. Yeah.

>> Okay. This is a comparison done by um

one of the other people on X for the

version 1.5, version two, version three,

and now A1. The other thing is that

>> the A1 won't be the last one. like in

order for them to hit these numbers,

they'll need something twice as big,

four times as big. Um like 2028 and

>> twice as big,

>> right? So they they'll be

>> twice as big. Yeah. In terms of the

solar panels, other stuff.

>> I showed a photo where this is the size

of a Boeing 7878.

>> These are the wingspans. Yeah.

>> The same size. 70 m total.

>> It's massive.

>> Yep.

>> But in space it's tiny.

>> Okay. That's right. In space was tiny.

Yeah. So the reason this is huge demand

is this is the terrestrial buildout of

data centers in the United States. Okay.

So the black means under construction

and the orange is the announced but not

under construction. Right? So they

wanted 12 gawatts in 2026

>> and they wanted 10 gawatts in 2025.

>> Each year they're only completing maybe

5 gawatts. Right? So, and the thing is

some of the black is under construction

and not completed because it takes most

other places four years instead of one

year for XAI, right? So, the other part

of like why is Anthropic and Google

paying $26 billion to XAI is because

they have 20% of all AI data center

that's actually working, right? So

that's why they're ahead of with the $26

billion, they're ahead of Amazon on AI

cloud. Amazon makes $150 billion on all

cloud, but they only make $15 billion

per year annualized on AI cloud. So

they're ahead of them. They're ahead of

Google's entire cloud business. Google's

entire cloud business is $20 billion.

They're $26 billion. So it's not like

future all that stuff has to happen even

on terrestrial data center they're

kicking butt if they said okay I give up

on cursor and grock what they have I

rent out Colossus 1 and Colossus 2 that

is 64 billion per year of revenue of the

stuff that they have built and they're

adding they're talking about adding

another 220,000 chips 400 megawatts

probably in the next 6 months that

replaces is all of the stuff they rented

out to uh Anthropic and Google. They can

repeat that because they built another

every six months they're adding that

stuff up.

>> So, this is the insane economics they

have and they already have the deal for

like 2 gigawatts of of natural gas. They

have the third location so they can

expand this stuff up. So, the fact that

you know that they're leaders in

terrestrial AI right now. Can you repeat

what you just said because that was

mind-blowing to me. You're saying that

SpaceX AI terrestrial colossus data

centers that they're building in 6

months they can have $60 billion

per year instead of $24 billion per year

run rate. They could rent out more to

equal $60 billion a year by the end of

six months. Like they have that much

space if they decide to rent it. If they

decide to, you know, whatever.

>> Yeah.

>> Yeah. your slide here with the US the

terrestrial US data center buildout

shows that every company who wants to

have a data center are struggling to

even get it under construction. It'll

take two to three years and so there is

a tremendous demand. I think Microsoft

probably is going to be the next company

to announce that they're a partnership

and they're going to buy billions of

dollars worth,

>> right? Who like the the the demand went

vertical for uh Anthropic um you know

where they went from like um 9 billion

to 47 billion. So 5x in five months 5x

in five months

>> and they if they have more data center

they could go to 100 billion to $200

billion.

>> Where did you get that number from? So

that's

>> if if enthropic

>> I got that from Gavin Baker.

>> If Enthropic had more data centers for

inference specifically, not training

anymore. We're talking inference just

because when I want to ask a question,

it needs to have inference chips to give

me the answer back. They keep throttling

users because the demand is so high.

They don't have enough data center. So

they they throttle it. They make the

they make the models dumber. They

purposely have to make it dumber and

then they give you

>> but if they had they could they could

buy more from SpaceX AI they could I

think who said that I think Jensen Hong

the more chips the more money you make

it simply says the more data centers you

build the more money you're going to

make simple as that and so

>> right

>> of course Anthropic will go ahead and

pay another two billion a month

>> to rent from data centers from XAI

because they're going to make

>> 8 billion a month whatever five five

billion a Right. Dylan Patel says the

demand is five times supply. So semi

analysis using $7 million of analysis of

all tokens says if we could 5x our

supply the demand would meet it. So we

can go to like you know 300 billion$500

billion like next year or even beyond.

So so the demand is just unlimited which

is why he said every data center will

rent out. If you just want it, you just

say your price, go rent it out. First

tier, second tier, third tier data

center.

>> Yeah. Okay. Can you clarify again that

you is it true that after Elon clarified

what he says is going to happen that

2027 that you're actually going to have

1 gawatt of data center already in space

in 2027. That's next year. Is that true?

Because that's been pulled forward from

2028,

>> right? That's his goal to do it. So it

depends upon um version three starship

working. So by working I mean um

deploying um deploying satellites which

means restart the engine. So that the

last time they went up it returned they

return starship they didn't return the

booster. So this next goound they

relight the engine. So the relight the

engine means that they can if it goes to

orbit I can restart it. I can take

>> control descent. Yeah. coldest because

if I can't take it out of orbit, I have

a 200 300 ton hunk of space junk I can't

do anything about which risks their

10,000 satellites. They can't have that

lying around potentially damaging their

satellites, right? They will take it

out. So once they relight the engine,

then they can put up and they've done it

three or four times. They can put up

real satellites now. So then even if

they don't bring the booster or the

starship back once I just relight the

engine say um I I prove the relight next

time one after that I relight and take

and go to orbit I'm deploying real

satellites in the second mission after

this current one right one more and then

the next one after that so let's say

they go in June in July they start

launching real satellites version three

satellites right and they'll launch a

full Starlink though. Yeah.

>> Right.

>> Still

>> that's Starlink version three sally.

Right. But then and then that's money to

them and then recovering the booster.

They have done that three times. All

these people say nothing's ever proven.

They've already recovered three boosters

and they've relaunched two of them in

the 12 tests they've done. They already

did that. So they probably can do that

again. So one once they do any

deployment of anything from Starship

they immediately get cheaper than Falcon

9

>> and Falcon 9 is five 10 times cheaper

than anything else. So they immediately

get cheaper than anything else even if

they blow up the the booster and the

Starship.

>> Y

>> I don't recover anything. I just launch

it and then that's $500 per kilogram. I

recover the booster three, four, five

times. I'm at $100 per kilogram. Don't

recover the Starship, right? I'm only up

to the point where we're at now at now

where I recover the booster. So then I'm

expending six engines per launch. I

don't cover Starship yet. I'm only

recovering the 33 engines of the booster

or 36 engines and I have the the the six

engines. So then I'm launching at $100

per kilogram. I'm replacing six engines

every launch. I can launch a 100 times,

lose 600 engines, and I'm still fine. I

can launch. I can make 1,200. They can

make 5,000 engines a year.

>> How do you know that? That they can make

five Oh, 5,000 engines. How many How

many rockets?

>> Raptor fours, Raptor 3es, Raptor 3es.

The the the engines on each one of the

rockets.

>> It's not the rate limiting step.

>> Yeah.

>> It's not the rate limiting step, right?

So then I have to make as enough of the

of the upper stage in order to do it.

But obviously I don't want to do that

because I can go even faster if I can

replace them both. If I replace them

both, then I'm down to like 50 bucks, 20

bucks for a kilogram. Okay.

>> Right. And I'm flying.

>> I'm just trying to get to the dates. So,

you're you do believe that there's a

real chance that they can get to one

gigawatt into space AI data center

viable by next year. And what is 1

gawatt equivalent to? I know that he

said it's two times electricity of of US

uh you know, consumption right now, but

what is it equal to?

>> No, no, one gawatt is one nuclear

reactor. One gigawatt is one nuclear

reactor. One gigawatt is Colossus. Um,

>> okay, there you go. That's what I'm

saying. So, it's equivalent to Colossus

2. So, which could deliver what? How

many billions per year? Per month

>> if you put Rubmanships in them. I

believe Rubman ships will rent for at

least double what a B200 is. So then

basically I can double what I'm charging

to um

>> anthropic

>> um

>> Google Anthropic, right? So I would say

it' be 500,000 chips. And so you have

550,000 chips of whatever. If you rent

it out at the current B200 price, that's

$50 billion per year. The price I'm

paying charging for Google is $50

billion. So if I double that, that's

hundred billion because I'm doing Reuben

ships.

>> This is crazy. Wait, this is crazy

numbers. Are you serious about this? So

if he puts up one gigawatt in 2027,

he could get just that one gigawatt

could be rented. If it's Reuben chips,

he can get a hundred billion dollars per

year renting that space out because I

I'm freaking out as far because 100

billion is what how many cars Tesla

sells.

all Tesla revenue

revenue right now that took how many

years to get where we're at is 100

billion a year which is yeah I I think

uh I think some of the big boys you know

they're making 100 billion every quarter

but still this just instantly makes

Starlink and this just one gigawatt and

they could keep like boom boom boom

repeating it

>> right right

>> crazy and they'll keep you doing the

newest chips so it'll be Reuben Fineman

each time getting more and more value

Right. Oh my god.

>> And then they can still make another

gigawatt and 2 gawatts on Earth cuz

Colossus 3 macro harder. They have the

land. They've been preparing it. They

just need to throw up the shell and and

put, you know, do what they did for

Colossus 2 right there.

>> And then of course they can use it for

themselves. That's the whole thing too.

And that's why that is being written off

by everyone else that Grock is a loser,

not there. But hey, if you give Grock

all that power, now this is for

inference, isn't it? It's not for

training,

>> right? Not for training.

>> So, how do you get Grock to be smarter?

I thought the whole point was the bigger

the data center, the bigger the

training, the bigger I mean, I remember

CEO Dario, he said, you he has a tough

decision all the time. He has to decide

how much of the chips I have, do I do

for training so I can smarter

intelligence or do I do it for

inference? And I think he says 50/50

kind of thing. it's, you know, he that's

he doesn't know how to do it. Right.

>> Right. So that's why the the ground um

earth-based data centers like Colossus 2

are still needed because they'll still

do training in it, but it's not going to

be 50/50 anymore. It'll probably be like

2080 20% training, 80% inference because

as you recall, um Elon and and SpaceX

rewrote the the training stack into C.

So this it's uh 90% more efficient. So

now I can do twice as much training

instead of doing 50% I'm down to

>> they're just innovating left, right, and

center everywhere, aren't they?

>> Their ability to build fast but also uh

they they rewrote the ability to yeah to

training faster in just smaller number

of chips required right

>> Jesus it's crazy.

>> Yeah. So this is again one of the

modules three tree three and I think I

don't think necessarily you can should

count how many just representing how

many um racks.

>> So this is the rack right here or what

is does this expand through this?

>> They they're covering it they're

covering it with a protective shell and

probably also has something to do with

heat dissipation or other crap. So

they're covering their their racks with

a bit of protection so that you know a

little bit of dust doesn't break it up.

And then the Giga Satellite factory. So

this is um you know how they're going to

have 1,000 acres nearly two square

miles, 11 million uh buildings square

feet. So as as big as um Giga Austin

will be all these things making solar

cells, wafers um terminals um PCB

terminals which are the big money

makers.

>> This part right here is the existing

one, right? Just this is the one that

>> right the next slide shows

>> this is what's existing now as in it's

already making the

the the

>> dishes and satellites for

>> uh space starink for starlink and then

they plan to build this for the AI

satellites solar cells sol

>> and AI satellite production

>> right

>> when how long will it take to build all

this out

>> um I'm sure they get the buildings up in

in a year. Um, filling in the equipment

might take another year or two. Like

roughly, I would think, well, if they

want to get this thing going, then they

have to be working by 2028, right? That

was the goal. 2027, you probably need

certain things, at least part of it

working at the line level. So, they need

to put it up in in less than a year and

then get fully

>> AI satellite, AI AI satellite one needs

to be made by 2027 to get that one

gigawatt up.

>> Um, you need um I think you need about

13,000 That's a lot.

>> Let me just double check that right now.

>> That's a lot.

>> Yeah. 10,000. Yeah.

>> So, I don't know if they're going to get

that done by next year.

>> That's a lot.

>> That's the goal. That's a goal. It's a

lot. Yeah. It's a lot. But but they they

they're making version three salaries,

which are very similar,

>> right? And they they're very scaling

those version 3es to about 5 10,000 per

year.

>> So, they put the communication guys up

first. Yes.

>> Right. And then they'll get these things

going as well, which they say are easy.

>> True. So they'll be diverting they're

diverting B3 production into AI

satellites

>> and the and the the the goals they have

indicate half like before I thought okay

I'll put up all 5,000 all 30,000

communication satellites up first right

>> there Elon basically saying I'm

prioritizing AI even maybe over that one

because if I'm I have enough capacity up

there to have double the customers right

so If I if I only need to add 2,000 more

V3s in order to match I want to go to 50

million customers by the end of next

year, right? I can only I only have to

launch like, you know, 40 launches or or

30 launches for communications and then

I can have the rest for this other new

thing which needs more stuff. So this

could become 70 80% of the launches next

year because they can say, "Okay, I I'll

get less on the communication side, but

still enough to give my my customers

what they need."

>> Okay, great great points here because I

interviewed Aaron Bernett of Mach 33 and

he has the most comprehensive SpaceX

business model out there. Elon loves it.

Arc Invest and he collaborated and what

he said was that you know the hardest

part for his decision-m is to try to

read Elon's mind and on and where you

would allocate the funding where would

you allocate the funding would you and

he spent he prioritized Starlink V3 and

had that going out to until 2028 where

you got the AI satellites Elon comes out

does an interview and says what you just

said no no no we're going to get one

gigawatt of AI satellite out next year

and then you said which is he said this

too the starting V3 they're already

doing it now there was a clip that I

played where he and the engineer said

that actually the same technologies to

build the V3 is this it's similar to the

A1 in fact they actually said it's

easier to create these A1 satellites

than it is these V3s because they're

bigger and they're they're simpler

technology so they feel very confident

that like you just said they're already

making you said 5,000 of these

Now,

>> so to be able to then make, you know,

these it's not it's not crazy numbers as

I was alluding to earlier,

>> right? And I I'm thinking that they're

cannibalizing

the intended V3 production.

>> I was going to make 10,000. Now it's

like screw that. I make 2,000 of the

V3s, 8,000 of the AIS,

>> right? Um the other thing is I've I've

done work with Aaron and with Vlad um

his his main guy on some like aer some

articles where we did research on on the

SpaceX side. Another disclosure is I own

you know SpaceX and not I own other

stuff around that. So just

>> you own the SpaceX stock already. So

just your and other things around it. So

just FYI. Yeah.

>> Okay.

>> Yeah. Just FYI. Yeah. I did you know

SPVS and stuff like uh you know six

years ago. Yeah. So, so in terms of my

analysis, it is um um Mach 33 Aaron

Bernett level because I actually have

worked with them.

>> Okay. Well, I can tell. Yeah, you're

you're one of the best out there. Thank

you for sharing that info with our

audience. This is terrifying. Yeah, it's

massive.

>> Right. Right. So, 10 times the um the

size of the Austin Gigafactory, you

know, which is many times the size of

the Pentagon. Well, I mean the Giga

Texas factory was supposed to be the

largest factory in on the US except for

the Boeing factory. I think Boeing is

the only one that's bigger.

>> And now they're creating tariff which is

10 times bigger than that there. Yeah,

this is unbelievable,

>> right? And they'll have, you know,

double the output of the US um energy

consumption. So the 500 gawatts United

States, one terowatt for this thing. And

again, just um initial plans, you know,

they just want to keep going. I know of

course he has bigger ideas or bigger

missions.

>> So something else that appeared in the

in the video interview is he gave the

TAM a total addressable market um in

rough terms for highspeed internet. He

said a few hundred million of these

terminals which are the dishes to

receive highspeed internet. These are

the new version five highspeed internet

dishes. A regular one and a mobile one.

That's what you're watching right now.

And they are thinner than the current

dishes. So they're getting rid of, you

know, their inventory of old dishes and

they're converting to the V5. Um, it

also will be able to handle military and

government workloads.

>> So this these were taken apart at the

firmware level by a Ukrainian guy

because you can't get support in

Ukraine. So they have to take it apart.

So then they at the firmware level

they're analyzing this thing to do it.

There is no GPS in these new version

fives. So that means Starlink location

is better than regular GPS. That's what

another thing that means. These things

are the construction of these at boss

electronics.

Think of them as a flat screen monitor.

That's what that's what those those

things are in terms of the economics and

in terms of all that kind of stuff. So

how much does it cost for a small

screen? You know, if it's just a small

screen, it's less than 100 bucks.

>> 300 bucks, 500 bucks for these uh,

>> you know, 21 by7, you know, monitors

that I have.

>> Yeah. Right. But then the smaller

monitor is even cheaper. You can get it

down to like 30 bucks or 50 bucks. So

those are economics. And then production

getting to hundreds of millions, tens of

millions, 100 million, the TV, the

monitors built every year. So they can

get to that level. So right now they're

at about a million per month. They're

scaling up to 1.5 million per month.

They will double again to three million

and double again to 6 million per month.

At 6 million per month, that's 72

million per year. Saying a few hundred

million terminals is where they're going

towards.

>> So these things do more less power. One

to two hour battery.

>> So and also battery included. So

normally for a dish you need to plug it

into the wall in order to run it.

There's a lithium battery in this thing

to give it one to two hours of power. So

I can like have it not plugged into

anything

>> and you're already

>> and I can run it run it for one to two

hours,

>> right? And I have a USBC adapter. So if

I had a a car or anything that the ash

uh the cigarette lighter,

>> I can power it with a USBC.

>> So all the connections are standard now.

>> Um yeah, so just it's a consumer

product,

>> right? And able to handle military grade

workloads as well. uh finance, banking,

government, these things can do it all.

And the thing is they'll won't stop at

these these two products. They can make

a super mini mini that might be like,

you know, the size of that other thing,

you know, like sm of a phone and then

they'll pump that out. So there'll be

variation on the products that they will

have here. So anyway, so 100 million few

hundred million terminals, let's say 400

million terminals, right? That's the

kind of what a few hundred million might

mean. That would be about $200 billion

per year. At $50 per month or $40 per

month, they can get to $200 billion of

just high-speed internet revenue. And

they can scale up again the dishes out

in two or three years.

>> This is crazy. That's unbelievable. So,

this is basically all you need to do in

order to get high-speed internet is buy

one of these things and if they can make

it as cheap as a flat screen monitor, uh

they might even just give it to you for

free. I I've seen some promotions now

and then you they charge you some much

much cheaper price than right now it's

about equivalent to you know Comcast or

Xfinity and these other internet

providers. Uh but now this will make it

easier and cheaper and then you you just

you just point this up to this guy and

uh you got internet access at high speed

>> and you're good to go.

>> Yep. Sticking to every car everywhere.

>> That's what this is for, isn't it? This

mini is going to be in the cyber cabs.

We already saw every cyber many cyber

cabs we're seeing them bolting in one of

these things

>> and they're pretty big but if it's this

small it could just be part of the roof

even know it's there

>> pretty simple

>> every every optimus as well. Yeah, this

could be. What do you think? Well, no,

but it's got to be facing up. It can't

be on the head, right? So, what on the

back?

>> It can be on the head, but it can be on

the back. And then if it needs to angle

up, it can bend over a bit. You know, it

it can do whatever it wants.

>> Maybe

>> whatever orientation you want.

>> Maybe. I don't know.

>> Okay.

>> And yeah, so

>> I guess it's like a backpack, right?

Like if you have military use or you got

these optimists that are d, you know,

going up the mountains or out working on

the farms. Yeah. You don't want to be

plugging them into an electricity

>> and so they they've got to like have a

backpack in the back which is this thing

and it's getting internet access.

>> Yeah. And just some little lever thing

or arm and just angles it up a bit.

>> Yeah. So the other thing is the Echoar

Spectrum that they got

>> is used in the new dishes.

>> This means more bandwidth. So not only

would the XR spectrum used for the

direct to cell phone which you can use

it for it also enhances highspeed

internet. So if you and that means they

can go symmetrical. Right now the upload

download speed is I upload at 10 to 20

megabits per second. I download at 200

megabits per second 400 megabit.

>> That's what I get here. Yeah.

>> Right. Right. But now with the extra

spectrum of the new version five uh dish

>> I can have

>> Oh my god. This is crazy. 100 up, 100

down, 300 up, 300 down, 500 up, 500

down, gigabit up and down.

>> They can go to gigabit up and down. Uh,

with that, so it's faster and

symmetrical. So that means any kind of

like, you know, gamer application, other

stuff like that. I can do it 20

millisecond latency. So be fully

competitive with fiber.

>> So I can get the gigabit fiber, latency

of fiber, up and down of fiber, fully

competency.

I mean, I can't even get that with my

wired freaking

I'm so upset with me. Look at you. Your

your internet's terrible, dude.

>> I know. I know. I'm pushing over to Sky.

>> Yeah. My problem with me is that they

will charge me I think at one point

$1,000

uh extra fee just because because of the

demand is so high in the city here.

>> Then I think they lowered to 500 bucks.

So, I might just pull the trigger. But,

>> you know, it's like they need they can't

be doing that. they got they can't be

charging me extra just because and so I

think with the V3s going up that would

eliminate this congestion that they have

of limited bandwidth limited offering

and then maybe the bigger cities can you

tell me or that because you know how

they've always positioned it that this

is really for rural it's not for big

cities where there's too many people.

>> Yeah. So they can by having 10 times 20

times more capacity that means they can

go from a little bit in urban to a lot

more in urban

>> and if I say okay well I want to limit

it to 100 megabits per second I can

still get all the customers so you can

get to you know a nice V4 V5 but thing

is B3 and 15,000 V3s 20 times better

than V2 is not where it stops. He

already said V4 and B5.

>> And he said 100,000. So now if I have

100,000, I have 10 times more dishes and

I've gone to 100 times more speed. So a

thousand times more speed. So now I

think they can do urban. I think they

can then do urban with that because

they'll have a thousand times more more

capacity, right? So then they'll be

going straight into urban areas

competing on everything.

>> I mean they're just going to win it

straight out. They're going to win it

straight out. This is crazy. lowest

price, highest, right?

>> You know, enough bandwidth that people

are happy. The key thing was the

latency. Yeah. And uh they're there. The

new technology is, as you just said,

right there.

>> So, direct communication everywhere,

direct AI everywhere. This is the

perfect timing, right? Right. When

everybody's increasing their token

usage, asking questions, getting answers

back from the AI, you need that

everywhere and anywhere you go. And now

you you can do this

>> high bandwidth,

>> right? Right. Yeah. Yeah. So, you're

getting every AI access you need, every

communication you need, every GPS, it's

everything that they're delivering

everywhere. Uh, so no restriction. And

then some people say, well, they have

less spectrum of this or that. They'll

be making crap tons of money. They're

going to buy more spectrum, right? So,

so then they'll be able to compete on

everything. I think they'll be

converting to the 200 ton payload

version 4 rockets in 2027. So, that

means 200,000 would need Okay. Whoa, you

got more.

>> Yeah. Yeah. Um, this is just showing the

speed which they built the classic one

class

>> compared to other guys is big. Yeah.

More more power. Yeah. Next. Um, yeah.

So, number two in AI rent only open

AI/Microsoft has more the Golden Dome

part of it skipped over. We discussed

it.

>> Uh, next one. And then you know how much

ARR the companies have bottom line of

the first column. Amazon $15 billion run

rate for AI revenue on the bottom line.

Yep, there it is. AI revenue. $15

billion run rate. So, they're already

almost double that at $26 billion run

rate. In October, they'll have that. And

then um Microsoft has less overall

cloud, but they have more AI AR because

of uh Open AI at 37 billion. And then

all cloud revenue for Google is 20

billion. So, we're already beyond them

in in with AI the better cloud revenue.

>> Yeah. Just instantly all of a sudden

leaprog these guys.

>> Wow. That's right. Okay.

>> Yeah. And then the next one. Yeah. So

then this is just um adapting into Tesla

which we can discuss later next time is

that Tesla can rent Cortex. Tesla can

rent AI they put at superchargers and at

power walls and mega chargers. So

there's all these chips that Tesla can

also rent out. There's a bunch of free

money that uh Tesla will be taking that

will be a substantial boost. So, next

time we can discuss how Tesla can do the

same thing and how that folds into what

SpaceX is doing because they're, you

know, Siamese twins.

>> Tesla can run, too.

>> You know, uh,

>> yeah,

>> Tesla and SpaceX are gonna probably

merge. So,

>> SpaceX is going to run, too.

>> That's right.

>> Okay. Right. Wow. This is powerful and

uh you did a lot of work as always and

we appreciate it. This is uh as the

elite level analysis uh you you dive

deeper than everyone else as you you

contribute to Mach 33 uh and others and

so you you're right up there. So, wow.

This was there was several areas today

that was just mind-blowing and uh I

can't I can't wrap my head around it and

I don't want to cuz I still I think it's

too crazy.

>> But yeah, that's that's Elon, right? And

>> that's right.

>> Thank you for sharing. Yeah,

>> you're welcome.

>> Uh yeah, check out Brian at his uh blog

at nextbigfuture.com.

Thanks everybody. I've created a website

that is the most comprehensive resource

for the Tesla investor. Please check it

out. Simply go to my website at

herbertalm.com.