Thread #108045553 | Image & Video Expansion | Click to Play
File: kdjshfjf.jpg (410 KB)
410 KB JPG
Elon just posted that the future of space is AI data centers in space? Is this reasonable and profitable
230 RepliesView Thread
>>
>>108045553
You do your space walk and then a special menu pops up on your video watch.
>do as you are told
>you will have 5 minutes to complete the tasks
>do not tell anyone about this
>...[ok]
>small portal opens
>you can see the gears that run the sky ceiling inside
>an alien hands you a plug
>you attach it to the datacenter
>as you were, astronaut
>you aren't even supposed to tell your Earth wife
>she wouldn't believe you anyway, SpaceX operatives spread rumors of homicidal space craziness to keep people from talking
>>
>>108045553
>Sending whole data centers up to space using chemical rockets where no maintenance can occur for months
>will have to de-orbit after 5 years
>latency will be hell
how does this guy get away with saying this bullshit time after time and yet all his stocks are sky high? Are masses truly this dumb to keep investing in a con man that lied about being best player in diablo plus path of exile and quake. This guy lies like he breathes.
>>
>>
>>
File: 1700282749855454.jpg (28.2 KB)
28.2 KB JPG
>>108045553
>Elon just posted
>>
>>
>>
>>
>>108045553
it is both reasonable and profitable
>>108045596
they are going to send the whole fucking wafer up there, probably working with Cerebras to make a rad-hard version of the WSE-3 and a compatible grok model quantized in BF16 or with a custom ARM chip to pull the sparse off the wafer and send it back to earth
sats will only demise if they are extremely damaged, the ones that get "old" and still work will have thrusters that send them further out into orbit, increasingly latency but maintaining connection- then a newer satellite with more compute will take its place
space cooling is easy
they are going to use Frore Airjets. The vacuum of space makes it easy to keep the wafer cool because there are no moving parts and radiating the heat is easy once you get it off the surface of the wafer
the irony of this design is that solar wind becomes the biggest problem and the sun has been active as fuck lately, so the likelihood of surface static building up is very high. no idea how they solve that
>>
>>108046037
>probably working with Cerebras to make a rad-hard version of the WSE-3 and a compatible grok model quantized in BF16
in end gorund based data centers will be able to serve larger pool of people at a mass scale? how many satellites would you need to serve latest grok model in space to millions of consumers? While your competitor is using all extra revenue to research more, build more youll have to waste money on going to space.
>>
>>108046037
>space cooling is easy
how do you propose doing this?
you know that the iss has only about 10 crew onboard at any time (~1000 watt of heating) and about 200 watts of computer and still needs about an acre of ir cooling panels on it right? imagine how hard it is trying to cool a megawatt datacenter
>>
>>
>>108046238
you have to understand their scaling model
>they don't need permission from a government besides launch orders from Space Force/NOAA to put compute in orbit
>they can push the old satellites out further and use them as networking nodes
>energy is basically free
>cooling is basically free
>training becomes cheaper than your competitors
>inference is only limited by network connection (that is, not at all for anyone with a starlink dish)
limiting factors were physical space, water, cooling infrastructure, power and local regulation. this solves all of those except for the surface static buildup issue.
and the ROI for each sat goes down for each die shrink. so that's why Elon wants to build the TeraFab. He can build chips for Optimus, Starlink sats, Tesla vehicles and the xAI datacenters.
shit, they could just send up a bunch of "old" chips that have already been used if they engineer a rad-hard enclosure. probably not the best use of spacelift money but for just testing they have plenty of chips already in datacenters they could sacrifice that are already depreciated in the asset schedule.
Currently OpenAI, et al. are limited by network location. xAI running in space means they are the only ones that can push models to other places but the Earth...that's why he mentioned the moon, Von Neumann probes. Elon wants to have the fastest AI model available across the world via satlink and the only model available in space.
>>
>>108046293
>>108046322
frore airjets and microgap cooling with chassis conduction.
hypothetically even if the airjets fail the chips could just downshift on clock assuming pumps still work. if both fail yeah you'd have a problem
I assume they have this figured out already
>>
>>
why is he trying to combine all his ventures into one? genuine question
he has tesla spacex xai and the boring company
hes done or talked about all these collabs:
>tesla + boring company (failed subway but with cars)
>tesla + spacex (launching a car to space, "cold gas" thrusters allowing a car to fly)
>tesla + xai ("full" self driving that doesnt work)
>spacex plus xai (this thread)
>spacex + boring company (digging tunnels in mars to live in)
he just needs the xai and boring company and hes got them all linked
whats his end goal? surely he knows that everything he says is either bullshit or grossly exaggerated
>>
>>
>>108046368
the electronics are radiation hardened
>>108046375
he is trying to regain voting supremacy over tesla, spacex will buy tesla eventually
>>108046381
what do you mean? the heat will radiate from the chassis
>>
>>108046391
>regain voting supremacy over tesla
he doesnt need this, he hasnt had it since covid and nothing has happened
>spacex will buy tesla
again what is his end goal? you can say that x company will by y company but why?
>heat will radiate from the chassis
how? look up how the iss does this and then compare the numbers
>>
>>108046408
>he doesnt need this, he hasnt had it since covid and nothing has happened
he has a fifth column in the shareholders, if tesla gets bought he gets to do away with the milestone requirement and fully integrate the financials, cull some retards across all three companies, etc.
>>108046408
they only need one loop, there's no requirement for human habitability
I'm not sure what you mean, each starlink is well under a megawatt. there's going to be hundreds of thousands of these things
>>
File: 1653068554106751.jpg (93.1 KB)
93.1 KB JPG
>>108046037
>space cooling is easy
the absolute state of the indians i share this board with
>>
>>108046464
>fifth column in the shareholders
i have no idea what this means so it is probably meanless
>tesla gets bought
tesla is pumped to a trillion dollars, he will need to buy out every shareholder and spend at least a trillion to do it. remember the 50 billion twitter deal? imagine that x20. its not happening
>cull some retards
youre forgetting that he is willingly and enthusiastically hiring hb1 jeets anywhere he can
>no requirement for human habitability
true but you still need to radiate out a megawatt of heat somehow. it doesnt matter how hot the thing gets, it still needs some form of cooling or else it will eventually just melt
>each starlink is well under a megawatt
yes? no shit each are under a megawatt. they are probably under 1% of a megawatt
this helps my argument much more than it helps yours, its easier to cool down 5kw than it is to cool 1mw by (to use his favourite phrase) two orders of magnitude
the starlink satellites have cooling panels on them (not to mention they also "beam" down energy in the form of "internet waves" so the energy doesnt build up as much). they have to dissipate much less energy than a data center up there
>hundreds of thousands maybe ill believe it when he puts a single one up there
>>
>>
>>108046327
>training becomes cheaper than your competitors
how will you train thousands of gpus unless he's building a whole space center up there for a dedicated AI data center. You know nothing about how any of this work when you said training would be cheaper despite fact most of compute AI companies are used for trainings that take months and require lot of bandwtih and connection. Imagine lag of having to pass information between gpus across distances while gorund based dont have to travel that far
>>
>>
>>108046516
they cracked X-ray litho, retard- go research Substrate
none of this is viable without that, not 18A, nothing. I expected more of this board.
the amount of waste heat you think will be generated per satellite is much lower than you seem to be implying, and each satellite is much smaller physically than the ISS
all of the h1b jeets that aren't useful will be sent back
im done replying to you, you're an idiot
>>108046551
efficiency is relative
>>108046517
>i have no idea what this means so it is probably meanless
there are people who own a lot of tesla shares that want to vote him out. spacex buying tesla mitigates this issue
>>108046559
training checkpointing, Cybercabs will train, existing Teslas will train, xAI DCs will train, etc. you don't seem to be grasping the full picture here. new Teslas will have Starlink sats built into them (Cybercab likely already does)
>>
>>108046640
>training checkpointing, Cybercabs will train, existing Teslas will train, xAI DCs will train, etc. you don't seem to be grasping the full picture here. new Teslas will have Starlink sats built into them (Cybercab likely already does)
in order to train you'll need to sync the data and be able to make sure the overall model can be tuned properluy you can't just train one part then try to recombine them like jigsaw puzzle. Why do you think companies are so interested in memory, bandwith and being able to overcome memory bottleneck where gpu computations outmatch speed of which data can be transfered.
>>
File: 1763316153434165.jpg (44.3 KB)
44.3 KB JPG
>Elon just posted
>>
It's actually a shame he did this because SpaceX IPO was probably going to be incredible to get in on (it's a very profitable company and generates billions in profit each year), but now xAI basically means
a) It's going to bleed money at an unprecedented rate
b) It also is taking on all of xAI (and X/Twitter)'s debt which is in the billions
c) You have retarded fan fict like OP's pic where Elon says incredibly dumb shit, that he can't deliver on and ultimately will scare investors off once they realize he can't deliver on it.
>>
>>
File: 1766460836884049.jpg (546.2 KB)
546.2 KB JPG
>>108046322
bruh you just dump it in space, space is cold as fuck
>>
>>108046640
>each satellite is much smaller physically than the ISS
well there we go, useless
if its smaller than the iss it generates less waste heat than the iss, which is about 2kw. are you saying that its economically worthwhile to blast 4 gpus inside an enclosure into space a thousand times and somehow have them all link up?
>>
>>
File: d22.jpg (151.9 KB)
151.9 KB JPG
>>108045553
about as retarded as mars colony, but it does pump his stocks. and spacex will genuinely drive industrialisation of the moon.
>>
File: weirdest duck.jpg (64.8 KB)
64.8 KB JPG
>>108045647
>>108045681
>>108046662
You guys are forgetting the best part about this fucking scam, which is the part where he mentions building stuff on the moon with materials already there, basically more satellites through an entire supply chain that we can't barely build on Earth, fully automatized. We are talking mining, processing/refining, manufacturing and quality control. There are so many reasons why this can't be done before the end of century, without making use of magical technological leaps, that whoever gets excited by this is nothing more than a worm with no right to live on $God's good graces
>>
File: 5e8ecfd234ad2-33-5e8c415082c3f__700.jpg (158.5 KB)
158.5 KB JPG
actually, in space you could do superconductive computing. much more efficiently. requires new engineering ofc.
>>
>>108047260
>genuinely drive industrialisation of the moon
cope
>https://www.youtube.com/watch?v=UYRgrw_oPWw
this is your government openly admitting that china won
>>
>>
>>
File: kessler.jpg (315.1 KB)
315.1 KB JPG
OHHHHH U PUT IN SPACE
OHHHHH WE DO TOO!!
OHHHHH U PUT IT THERE??
OHHHHH WE DO TOO!!
OHHHHH THERE CORRISION?
OHHHHH SO SOLLY!!! ACCIDENT! ACCIDENT!
>>
>>108047279
my govt? bruh i'm russian.
also china has russia tier rockets. spacex is MVP and it's gonna drive the competition to pull up its pants.
what causes you american cuckold to hate your own people thus? we hang traitors like you before our enemies.
>>
>>
File: rad_space_pics4812875628_d2e02a2673_o.jpg (1 MB)
1 MB JPG
>>108047304
nothingburger, you can still launch through it. space is conditional upon industrialisation of moon, at which point humanity has taken off and can trivially clean up orbits.
>>
>>108046674
>space is cool as fuck
>because there is nothing that can contain energy in the first place
>so you first have to put the excess energy into a medium
>and then send it off into space
>and the medium that we use is ... .... ???
I just want to torture people until they laugh themselves to death.
>>
>>108047307
>spacex is MVP
how? they do nothing but launch their own satellites to space. ULA is the mvp and competition doesnt exist when the upfront costs are in the 8 figures
>you american cuckold
im not american, im australian. i have no skin in the game but can easily see that if the american congress is discussing how to thwart china without publicly disclosing it theyre in a rough spot. also your mvp is going to be more than a decade late (if ever) on fulfilling the third artemis stage
>>
>>108046662
AND why shouldn't a government funded rocket company be responsible for the liabilities from Musks (probably drug-fuelled) impulse and overpaid purchases of random social media websites? The synergies are obvious. Twitter can, eh, post launch notices and stuff. Until the Democrats return and tear this fat scamming fuck forcibly from the public teat.
>>
>>
>>108047365
>Until the Democrats return and tear this fat scamming fuck forcibly from the public teat.
Your optimism is a joy to read. NTA
Nothing would make me happier in the near future than watching Musk shit and piss himself in anger as his incestuous merger goes down the drain
>>
>>
>>
File: 778x1024_5983_no_signal_2d_illustration_dead_horror_space_planet_nasa_dead_space_astronaut_picture_image_digit.jpg (125 KB)
125 KB JPG
>>108047351
any company getting ahead is good because that drives competition. thus spacex reusable fanboyism is good because it puts more pressure on other participants to ship their own attention-grabbing techs. while you sound like a seething hapa. china is based because they ship. same as why spacex is based.
>>
>>108047378
holy visceral hate
he really gets you seething doesnt he?
are you jealous or something?
>>108045553
i think its retarded.
because how are you gonna upgrood your hardware?
if he does this to evade earth's legislation then hes gonna get a nasty surprise because his companies are still earth-bound, and are not something irreplaceable, so all his circus can get drowned in a glass of water
on the other hand
this could prompt the creation of a proper spaceport in earths orbit, for servicing this junk
but that, in turn, creates a whole lot of commercial possibilities for him because orbital docks allow to build a lot of interesting things for a lot of wealthy people
the idea in isolation is retarded.
but if he makes it into a stepping stone, and doesnt crash before that, then it makes sense.
theres also the problem of rentability for that project and how it will affect shareholders etc
hopefully he doesnt use his own shares as collateral because that could end him:
>company value drops to zero
>youre left with debt, and now have to pay it from your own funds
thats the recipe to go into deep red. thats way worse than getting btfod back to 0
>>
>>
>>
>>108047427
>because how are you gonna upgrood your hardware?
That's the neat trick, you don't. You just let that shit burn on reentry and call it a day. Then launch another 1000 satellites on the same day over brunch. Kessler who?
>>
>>108047408
>reusable fanboyism
reusable rockets still havent been proven to be economically viable, and its unlikely that they ever will be without a massive reliability boost (and everyone knows that spacex and especially starship isnt the king of reliability)
>>
>>
>>
>>108045553
>>108045553
>>108045553
G E N I U S !!!
Keeps mentally deranged libturds who literally hate AI (and humans/white-people) from being able to physically attack the AI servers!
>>
>>
>>
>>
File: file.png (1.4 MB)
1.4 MB PNG
>>108047510
.
>>
File: 1759205026054494.png (1.5 MB)
1.5 MB PNG
>>108047541
>hyperloop
>what we actually got
>>
File: oldam ron miller salyut type space station.jpg (180 KB)
180 KB JPG
>>108047482
theory is for fags. doing is for chads.
>>
>>108045553
It's part of the Golden Dome that Trump keeps mentioning.
Basically a planet-scale missile shield, with real-time detection capabilities. Able to strike anywhere, at anytime.
A large part of this is detecting anti-american activities and in these scenarios every microsecond counts so rather than "spy satellite sees thing, sends data to Earth for analysis, Earth sends signal to missile to fire" it'll just be "spy satellite sees thing, does local analysis, sends missile immediately". It'll also help significantly with satellite-to-satellite combat since there latency is even more important.
>>
>>108047378
>as his incestuous merger goes down the drain
The merge is complete lol
>>108047365
>a government funded
it's privately held, getting funds from the government or having government contracts doesn't change anything. Practically all of FANG is government funded to an extent.
Also SpaceX actual profit driver is Starlink and Starlink subscribers. The rocket launches, which are largely government contracted, do not make nearly as much money as Starlink.
>>
File: 1767158287444433.jpg (113.7 KB)
113.7 KB JPG
>>108045553
heat is basically impossible to get rid off outside of thermal radiation into the vacuum of space.
they'll start the first rack and cook the station in like two minutes.
brilliant move mr musk, you sure beat them (?)
(also space latency fucking sucks, and you know what needs low latency? user facing services, like AI)
>>
>>108047592
>no pedestrian traffic
surely not, has anyone tried it? i imagine an ebike could make great use of that system
>>108047603
theory gave humanity everything it has. you should look more into the behind the scenes of the apollo (and previous) program
>>
>>
>>
File: 1770069540849.jpg (26.7 KB)
26.7 KB JPG
>>108045553
blocks sunlight since there will be so many that a new ice age will happen. or is happening due to his starlink satelittes?
>>
>>
>>
>>
>>108046037
>they are going to send the whole fucking wafer up there, probably working with Cerebras to make a rad-hard version of the WSE-3 and a compatible grok model quantized in BF16 or with a custom ARM chip to pull the sparse off the wafer and send it back to earth
Man you people are out of touch with SpaceX, Tesla and xAI.
Tesla already builds AI4 chips for their AI computes for their cars. They plan for inhouse fab building in the near future. They have AI5 taped out, AI6 being worked on,AI7 chips, etc. xAI and now SpaceX wants to partner with Tesla to build out their chips in space at megascale. Musk has already talked about not using rad-hard version, they dont use any rad-hardened version for their Starlinks in space, or their rockets in space. Its a non issue for noisy AI chips in the first place, so its unneeded.
>space cooling
Starlinks are already 20kw sats that use space solar and space cooling. They have 9000 of them in space. You dont need to look at another company. SpaceX is the ONLY one with credibility and at scale.
>>
>>
>>108047708
Thats also a low ball figure conservative estimate too.
https://spaceflightnow.com/2023/02/26/spacex-unveils-first-batch-of-la rger-upgraded-starlink-satellites/
The v2 mini Starlinks have 116 square meter of surface area. With 1367 Watt / sq meter in space x 20% solar panel efficiency, you get 273 watt per sq meter x 116 sq meter = ~31.7 kw. Lets say of the 116 sq meter, 16 sq meter is the center "core" thats not solar panel. So thats 273 x 100 = 27 kw of solar panel. Now if you consider that its going to be a space baser solar panel instead of residential one(residentials hit 24% for newer ones), you have to use the standard space based solar panel efficiency, modern space solar panels have ~30-35% efficiency rating. But lets say we use a standard 30%. That gives. Thats 1367 watt / sq meter x 30%, thats 410 watt x 100 sq meter = 41kw. Double my initial conservative low ball.
>>
>>108047656
It's about ambition and funding. You have to present a goal. "I'm gonna make reusable rockets" doesn't sound as exciting and ambitious as "we'll have people on Mars by 2020". Only retarded subhumans would take anything people like Musk say literally and then gloat about it not happening. It's an embarrasing display of utter stupidity.
>>
>>
>>
>>108047766
>16 sq meter for core
Its actually closer to 10 sq meter, but w/e. The center core is roughly same size as a standard car, Tesla Model 3 has ~10 sq meter for l x w. Their larger V3 center core will be same size as a bus or a truck. And their solar panels will be even larger with more capacity probably 3-5X
>>
>>108047766
your link doesnt have any wattage numbers, and your calculations are wrong. you assume every bit of surface area is in the sun and generating solar energy 24/7. obviously this is absurd just because only one face can be in the sun at any one time, not to mention in low earth orbit night effectively exists
the fact you got that much obviously wrong tells me you have no idea what youre talking about and im only going to believe you if you can get a source that makes sense
>>
>>108047807
>you assume every bit of surface area is in the sun
starlinks orient their solar to point towards the sun to get solar power to charge their battery for power when they go dark
>generating solar energy 24/7
no, there was no assumption, this is just raw kw capability, if I wanted 24/7 power output calculation, then i'd give you some kwh number not kw, you're confused
Further, your 2nd arguments are even more irrelevant because they want to launch new sats to SSO, which gets 24/7 solar power, that also dont need batteries, so its just raw solar power 24/7. You dont even know how wrong you are and you're telling others they are wrong. The level of confidence really is staggering
>>
>>
>>108045681
>>108047852
Low IQs dont realize SpaceX already operates 20-40kw solar panels in space that power their 9000 starlinks.
How do you think they cool those? lmao
>>
>>108047837
>orient their solar to point towards the sun
every satellite does this, but there is surface area on both sides of the panel so you need to half your original estimate
>raw kw capability
meaningless term without a definition. a satellite like what a starlink is needs to have juice in it 24/7. if it can only generate power 12/7 it has to store half of what it produces. i thought this would have been obvious
the discussion isnt about how much peak power production the satellite has, its about how much it has to dissipate per day. basically get your average heat in watt and thats your number. that number is no where close to 20kw
>>
>>
>>
>>
File: Screenshot from 2026-02-03 09-43-12.png (9.3 KB)
9.3 KB PNG
>>108047852
you can supercharge this shit and it makes more sense in space than on earth
you use a heat pump
superheat the cooling fluid
and leverage the fact that heat emission as IR increases with an exponent of 4 on temperature
and your stem is gonna get more efficient the more energy your compute will consume
because you can superheat the coolant all you want
but then you need to heat it back up once it expands, to maintain the same efficiency
and you cant go below absolute zero. not that you would want that, or even go too low either
which is no biggie anyways
if you dont max out your cooling capacity, it will run less efficiently
big deal, its not like youre wanting for energy
i like this solution, actually
the more strain your system is put under, the more efficient it becomes
i find it elegant
>>
>>
>>108047879
>data centers are magic that dont obey the law of thermodynamics
You're a retard. Its power in heat out. If you can expand 30kw of power and cool it, you can do the same for 100kw, or 1 gw. black body radiation physics don't work like magic, they are formulaic.
>>
>>108047887
I'm not saying you can't dissipate arbitrary amounts of heat, you just need arbitrary surface area. In the real world, have a 500x radiating surface to computer ratio is utterly retarded when the primary purpose is computing.
>>
>>108047541
>>108047592
>Hyperloop™
>it's a one lane tunnel
>>
>>108047898
They currently have 30-40kw solar panels on their Starlinks. Their V3 Starlinks are ~3-4X larger and their solar panels powers will be roughly 100-150kw. Thats the current near future sats they have. So they can scale up 100kw power in space effectively. Thats like ~100+ GPU cluster node in space per sat for 70kw of GPU compute unit. If you think they can dissapate 70kw of heat on one sat, then you can dissapate 3.5MW of heat from 50 of these sats.
Each of Starlinks currently dissapate ~30-40kw x 9000 = 360 megawatt of solar panels in space and heat being dissappated.
>>
>>108047935
You're still not getting the point.
Let me put it another way. If it was built on the ground it would require 500x less auxiliary material, which could be computers instead. Anyone building an earthbound datacenter would intrinsically be able to have more computers doing more for less with the same amount of money and time. That's the physics of it, there's no way around it.
I'm sure elon will get daddy trump to subsidize the project to the tune of a trillion dollorydoos so that elon can pretend it's more cost effective, of course :)
>>
>>108047953
There are 100 data centers in US that are cancelled or delayed by year due to NIMBY, partisan politics, populist politics and straight up grifting from the city councils. Fundamentals of ground base data centers are currently cheaper today, but that wont always be the case int he next 3-4 years as Musk predicts when more data centers come online and politics become even uglier
>>
>>108047966
And people are desperate so much so that they're willing to build in other countries like Saudi Arabia to get away from delays in US. There's also currently gluts in power plants, hassle in connecting to grid, huge taxes and blackmails being pushed towards people wanting data centers. And they should take advantage as people want to build, and govs want to get in on the grift. But soon, it would become too much and cost wont be sustainable.
>>
>>108047966
lmao you can just took in first person troon
Cope and make excuses as much as you like, it's essentially trivial if you're willing to spend what it's worth. Corpos are cutting every corner possible and then crying about the corners being there in the first place. Even if every corner is repsected, it's still orders of magnitude cheaper and more efficient to build a datacenter on earth.
There of course is a business for some computers in space (which is why there are already computers in space), but if you're looking to generally shift datacenters up there, which are there for the sake of being datacenters/avoiding politics, the outcomes are going to be REALLY, REALLY funny. I hope it happens so I can laugh at the 800% usage markup over normal datacenters and then see the tears on twitter about how everyone is bullying poor old mr musk and illegally not using his epic space datacenters
>>
>>108047998
There's natural affinity between Starlinks in space providing local internet across the globe, connecting and communicating with other sats with space lasers to route traffic. And these being retro fitted with small independent GPU clusters that provide inference from sats. Thats the affinity between xAI and Starlink that exist already. Alone by themselves there may not be sense as the cost may not justify, but together, there is lot of affinity in reduction of cost and efficiency gained. The question is how they will use the GPUs. They will likely offer enterprise usage as they are both ISPs in space and now want to be "space compute" and also X is now a global social media, they want localized AI as much as possible to reduce speed of inference where possible. Other companies have to deal with providing AI at long distances if they are not local themselves. That costs anywhere from 100+ms latency just travelling between distances for requests, where as local space clusters can provide sub 20 ms latency everywhere on earth.
Then the bigger question is what role does Tesla's future chip fab play in expanding and synergizing together in another massive cost reduction and specialization improvement for their future computes.
You would have to bet against people on Earth not wanting any computes, not wanting faster computes, not wanting cheaper computes, not wanting localized computes, not wanting redundant computes, etc to bet against Musk's empire. You'd have to bet against the entire computer industry to bet against Musk at this level of argument. Or be politically motivated to seek any irrational argument.
>>
>>
>>108048077
>And these being retro fitted with small independent GPU clusters that provide inference from sats.
These distinct clusters will have shit-scale data bandwidth communication between them.
NVIDIA's buy of mellanox was about very high speed data comms bandwidth inside the GPU clusters inside a data center, and they are doing it over ethernet-like cable using light particles.
Good luck doing something similar in space. Shit bandwidth between satellite clusters (running, what, 4 GPUs in a satellite? that's fucking pathetic) means your AI is slow to train and slow to infer.
>>
>>108048135
>These distinct clusters will have shit-scale data bandwidth communication between them.
Nah, few companies have already proven they can scale clustered training with minimal loss, one from china, another from US doing global cluster training, etc.
>4 GPUs in a satelite
GB200 NVL72 consumes upward of 130KW (and can reduce power as needed) 32CPU 72GPU unified system. Thats nvidia rackscale. Within the Starlink v3 solar power constraints. Thats just off the shelf nvidia component.
>>
>>
>>108047953
a radiator that sits in 0g doesn't need much material. I agree with the other anons in that something you can immediatly deploy without NIMBY, not need to plan for power plants or transmission lines, not need to plan for water, not need to plan for zoning is a win. You also don't have to pay for power on a continuing basis.
None of it would make sense without starship. In terms of spacex, Elon seems to be building his own demand. He has hundreds of launches per year permitted, iirc, and is building massive manufacturing lines for starship to be cranked out in mars volumes. Those early rockets can be put to work building a new revenue stream for spacex, and test out the designs as they evolve.
elon is being sued for running the gas turbines in memphis, iirc. Fuck that noise; If he does not need to deal with that bullshit, all the better for him.
>>
>>
>>
>>108047772
>calling out lies is an embarrasing display of utter stupidity
You sound like a gaslighting psychopath who should have his skull bashed in in front of a cheering crowd.
>also it's "embarrassing"
>meaning you're also a low-IQ psychopath
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
I don't know about all the factors, but latency won't be an issue at all. Most AI tasks take seconds or minutes to complete, so a few seconds of latency won't be dealbreaking. It's not like twitter will be hosted in space entirely. Plus, for batched industrial scale jobs (which take up a much higher percentage of AI traffic than you might expect) latency of literal hours can be acceptable.
>>
File: 1741809507770.jpg (360.2 KB)
360.2 KB JPG
>>108048764
the idea is fucking retarded. there won't be datacenters in space before we have literal magic technology. why do people nowadays take these retarded attention-grabbing baits seriously? mind uploading will be real before fucking datacenters in space.
>>
>>
>>
>>
>>
>>
File: 20251017_074122.jpg (54.9 KB)
54.9 KB JPG
>>108048815
that's a nigger take. the problem with missile defence isn't detection, it's that you physically cannot intercept massed strike. you'd need to completely industrialise LEO and still ballistic missiles might get through.
>>
File: file.png (194 KB)
194 KB PNG
>>108048846
its literally in the first couple paragraphs. The Pentagon is doing with Spacex the same shit they have done with Oracle, Palantir and Facebook before however this time it will not end the implications if they succeed are terrifying for everyone.
>>
>>
>>
>>
>>
File: f1e130fbb7292c6e4c7554909efdaebc.jpg (107.2 KB)
107.2 KB JPG
>>108048890
are actually retarded? do you understand that launching a counter-missile is incredibly mechanically hard and datacenters aren't gonna help you with it? and america doesn't have enough interceptors to shoot down more than like 2 missiles. are you pretending to be retarded?
>>
>>
>>108048944
Still less hassle, the water for cooling is there, less space constraints (there are already cargo ships holding tons and tons of weight so it is already a tested tech), you can easily onboard people for servicing, far far cheaper to build. and engines are in the 10k horsepowers so easy to power, some 3rd solutions like solars could be used to lower the costs of operating like solar
>>
>>
>>108048993
Each could only hold whatever could be powered by that surface area in solar panels plus battery, any engine power should need to be minimal since they wouldn't go anywhere. There's no reason to add any other fuel source to add to the cost and potential environmental problems that are trying to be avoided.
>>
>>
>>
>>
>>108048927
Only 4 hours of sunlight per day on avg. Where as in space, its 24/7. Also has to worry about environmental degradation, have to build a ship for each of them, have to build giant ships for them which can be bottlenecked with grifting middle man from state, local gov or companies
>>
>>108049287
The belief that Elon Musk is a fat wanker conman ≠ 'unconditional cynicism' fwiw. And I had to undertake (and pass) a multipart visual puzzle IQ test just to inform of you this undeniable fact, so there.
>>
>>
>>
>>108048954
The point isn't to actually intercept the missiles, it's to have enough fully automated firepower completely out of reach that if anyone even so much as thinks about shooting a a missile at America then their little shithole gets sent back to the stone age, regardless of if it's some rogue Iranian general launching a single missile at Israel or China sending their entire nuclear arsenal as a first strike
>>
>>
File: 3aaa7a35ad8a9b92b36c277463b6164a.jpg (1.8 MB)
1.8 MB JPG
>>108049586
that's not the trump's favourite "golden shower" and that capability is already there. the point of any new interceptors is to be able to ZA WARUDO russia's or china's entire first strike and any other strikes they might send. it's not a feasible concept anyway and datacenters in fucking space are fucking irrelevant to it.
>>
>>
>>
>>
>>
>>
>>
>>
>>108045681
Spacex is space and elon has the most talented people on the payroll to tell elon if this is retarded or not.
So it's either this can work or it doesn't work and this a deliberate grift to excite investors who currently have tunnel vision for AI and won't invest in anything else. When the AI bubble pops elon can just drop the program.
>>
>>
>>
>>
File: ad3cd11b9fc50049ef1720479e62cd62.jpg (491.1 KB)
491.1 KB JPG
>>108045553
Just you wait until bits start randomly flipping due to cosmic radiation, and the AI meme tursn into Hitler.
>>
>>108047541
>>108047592
I honestly think the hyperloop is neat. Yeah its disappointing that there isnt an actual vacuum tube thing happening but there's potential. Bore company seems more like a pet project or stepping stone for something else.
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
File: image.png (1.3 MB)
1.3 MB PNG
>>108045553
Elon is also flying to Mars this year for sure!
>>
>>
>>
File: DeMusker.png (586.6 KB)
586.6 KB PNG
>>108053048
>>
>>
>>
>>
>>108053988
If it made more sense, then more people would be doing it. Microsoft tried the ocean data center nonsense. It failed. Everyone with common sense knows why it failed. Ocean is the most hostile place on earth. Salt water erosion is hostile to everything. So you're not thinking right, just like many here who have retarded notion that they are somehow more superior that multi trillion dollar companies and almost trillion dollar man running multiple successful business empires
In space, there's nothing except maybe sun light, which can cause damage if not properly shielded, but with space equipments, everyone knows how to make basic space shielding that its a non factor.
>>
>>108045553
Will never be delivered just like all the other Musk pipe dreams.
>>108054430
>In space, there's nothing except maybe sun light
You people really know nothing, what are micrometeorites? Space is more hostile than the ocean, people on the ISS have to deal with all kinds of problems all the times, but of course dunnin kruger musk and his fanboys look down on actual scientists and engineers of NASA and other space agencies cause they are so smarter than "old space" right?
>>
>>
>>108045596
Not to mention the giant solar panels and heat dissipators to power and cool that thing would increase the amount of space junk particles and also be the main target of those particles. Just random tubes of cooling liquid being broken off by some bullet-speed flying hunk of metal, releasing tons of cooling fluid and creating 20 more flying hunks of metal flying at the speed of bullets.
>>108047262
Also this. But don’t doubt their faith in progress. OpenAI expects AI self improvement/singularity by 2028, so they think that by 2030 we will have AGI that automates everything so worrying about future problems isn’t something they think about.
However in best case scenario I can actually see them just sending people there to live in some “colony” like on the international space station and just have there few extremely highly paid mechanics and engineers doing the actual work being paid crazy money to their families to live isolated, in BEST case scenario in the next 20 years.
>>
>>
>>
>>108045553
How do you cool something in space genius, it's called space, there's nothing there, it's a vacuum.
Current spacecraft can only reliably cool things by either induction (VERY SLOW) or by using a medium that gets ejected into space like water, which would be finite. The entire idea is a literal waste of time.
>>
>>108057150
>The surface needed to radiate the power of a single solar panel away is approximately one solar panel. The back of the solar panel is facing away from the sun.
Solar panels don't make the heat of a graphics card, dipshit. All electronics in space are carefully engineered to only have the heat load that the craft can reliably remove. Which is why they're almost universally low-power ARM applications. Yknow, the things Graphics cards aren't.
>but space is cold :^(
That's too fucking bad, it's called 'Space'. There's nothing there to pump heat into.
>>
>>108047858
>>108047887
> If you can expand 30kw of power and cool it
https://umatechnology.org/how-much-power-does-starlink-use/
>The power consumption of each Starlink satellite is relatively low compared to traditional communication satellites. According to SpaceX, each Starlink satellite consumes about 1-2 kilowatts of power, which is significantly less than the power consumption of larger geostationary satellites. Geostationary satellites, which orbit at a much higher altitude than Starlink satellites, require more power to maintain their position in orbit and communicate with ground stations.
You're right, anon. Who WOULD just go on the internet, and make up lies?
>>
>>108057180
>Solar panels don't make the heat of a graphics card
They make the electrical power which the GPUs turn into heat. High power GPUs need more solar panels, more solar panels give more radiator surface ...
As I said, a lot of people have your naive take. If you persist at this point, it goes from naive to stupid. Which is also common.
>>
>>
>>
I just ran some numbers on a potential radiant heat ejection system that would be obvious to a ME. It seems workable. t.me that specialized in thermal systems.
I would be more worried about cosmic rays flipping bits in the electronics; but that has solutions too. The practicality of this system depends only on economics, imho. NIMBY, regulations, and underinvestment in power generation would tend to make this more practical, i'd think.
>>
>>
>>
>>108057850
Upper LEO solar pressure is enough to offset drag in an appropriate orbit. Turning PV by say +/-30 degrees might be enough to maintain get control in a good enough orbit (LEO polar orbit wit minimal eclipse) without compromising collection too much.
>>
>>
>>
>>108058285
Here's my bet for what an eceleb deboonker will assume :
- Equatorial orbit
- Not even the lightest current space solar panels (Redwire ROSA). Let alone taking into account that huge power hungry satellites might use flexible rollout arrays even lighter than that.
- Not combining the radiators with the PV.
- Not using gas/liquid phase cooling.
- Not contemplating that solar wind could be used for orbit maintenance, but instead only assuming that the large surface area would cause more orbit disturbance in solar storms.
>>
>>108046375
Because its not possible to raid the offices of a defense contractor without the risk of causing an international incident. Likewise, failures of some arms of the business may be masked by earnings and contracts made elsewhere in net calculations. It may also be possible to reclassify opex from a former operation as capex during a merge. Stuff that would be pretty useful during reporting season if you possibly mayhaps perchance needed that kind of "help" for some reason.
>>
>>
>>
>>
>>108047966
>There are 100 data centers in US that are cancelled or delayed by year due to NIMBY, partisan politics, populist politics and straight up grifting from the city councils.
because they're used for worthless endehaviors like AI, why would people want to bring this into their lives? there are enough documentaries on rural america getting assblasted by giant datacenters, look them up yourself or ask your favorite cuckbot to get the links for you.
>Fundamentals of ground base data centers are currently cheaper today
and forever
there is a inshrinkable cost for sending shit into space, unless we have a permanent civilization with advanced lithography and everything on the moon there is no way in hell a DC in space will cost less than on earth.
>as Musk predicts
who gives a fuck, muskrat is wrong basically 99% of the time
>politics become even uglier
the world is bigger than the US and china, there are/will be plenty of spaces where to build datacenter, it'es especially true now that trump renewed europe's will to severe ties with the US...
>>108048077
>connecting and communicating with other sats with space lasers to route traffic
what they need is storage then, the thing that require the most maintenance...
>nd these being retro fitted with small independent GPU clusters that provide inference from sats
what for? seems absolutely worthless unless you're talking about jwst processing raw data in space instead of sending shit to hearh but you're delusional if you think muskrat cares about progress
> localized AI as much as possible to reduce speed of inference where possible
DC on earth is better for this
> You'd have to bet against the entire computer industry to bet against Musk at this level of argument. Or be politically motivated to seek any irrational argument.
lmao how delusional can you be? DCs in space is peak retardation and a net negative for humanity
>>
>>
>>108058882
Look at what it takes to do a ng datacenter on land:
https://www.youtube.com/watch?v=NuJGgmhKqyQ
Elon is saying, fuck that; lets build a bunch of satellites, not need any land, any water, and no extra power. I suspect he has already paid to have some rough designs done, and already knows that it is possible technically.
>>
>>
>>
>>108059039
>Elon is saying, fuck that; lets build a bunch of satellites, not need any land, any water, and no extra power.
That's because he's lying to you, hoping you don't watch that video and become self-aware that this is just like his hyperloop promises, only the hyperloop is far more technically feasible.
>>
>>
>>
>>108059422
I think musk is trying to create some sort of extra revenue stream for spacex. They have a global telecommunications system. They are building massive infrastructure for launching ungodly tonnage to orbit. What can they do with that? (elon hits a bong) Orbital datacenters!
>>
>>108059212
>zero benefit to putting them there
Continuous solar power in a polar orbit.
Lets say putting 1 ton in orbit takes around 100 MWh. The most light weight solar panel arrays for space now around 100 kW per ton (if designed for extreme light weight with panels kept in tension with centrifugal force, that might go 10x higher, a PV film need only be 100 um thick).
In thousand hours currently available space PV has overunity EROI for reliable continuous power. No batteries needed. No nuclear powerplant we can't build any more needed.
>>
>>108051066
Fill in a retard
It seems the midwit take is no radiative cooling in space
What's the real answer? How do u cool better in space? Just the better baseline cold is enough diff to offset the loss?
Crazy badass buster sword sized heatsinks?
>>
>>108059632
Imho, you build a refrigeration cycle with the radiator fins being the high temp reservoir. Radiant cooling goes with T^4 so bumping the temp helps a lot. There are some toxic/carcinogenic refrigerants that look like they can operate at high temperatures (ex: carbon tetrachloride). Poking around suggest > 400dC fins are possible with a multistage but typical vapor compression cycle. 11kW/m^2 with 400dC fins, which is way over what you could get with a similarly sized solar panel. I stopped poking around, but maybe a single stage could work too.
There would be a million gotchas with the practical implementation of this - like refrigerant freezing into a solid within the system - you would build electric heating into it for startup. But maybe you could use helium as a working fluid and do something like a cryocooler.
>>
>Trusting a nonce who couldn't mantain stupid cars on a tunnel with orbital manufacturing, shipping and maintenance
Bro should stick to begging for access to the pedoparties by e-mail and let actual smart people do the work instead
>>
>>108059632
>How do u cool better in space?
- Lots of surface area, which works out well since you need lots of surface area for PV too.
- Heatpump to increase the temperature of your radiator, a condenser also maintains a consistent temperature better than a radiator with liquid, which is important for radiative cooling.
- lightweight heatspreader for your condenser coils (graphite film).
>>
>>
>>108059952
then T^4 kills you and you need such massive radiant coolers it would be impractical.
Its a balance though. End to end system optimization. But I answered my personal question that the thing was possible from a heat transfer perspective, which is what I cared about.
>>
>>108060024
>then T^4 kills you and you need such massive radiant coolers it would be impractical.
No more massive than the area of PV you need to begin with. If you could shield the backside from seeing earth, then at 120 degree condenser/radiator temperature, you are radiating way more power out the back of the PV than you are generating electrical power at the front.
You might not even need to shield the radiator from seeing earth to get enough cooling.
>>
>>
>>
>>108060135
Lol. When I read about the reasoning behind why it was 'impossible' I tried to figure out how much heat I could get out of 1^m^2. I never bothered to calculate if there was a problem at all. I don't fully agree with your analysis, but I do agree that cooling isn't much of a problem at all, that something on the order of the solar panels at lower temps would be fine.
If you stick a black body plate in the sun in a vacuum, that plate would heat until the point it would radiate away the same energy. What would the temperature be? If it is above ~100dC you would need active cooling to create your gpu chamber. I get around 120dC, so I think you do need active heat transfer. However these numbers make it a very un-challenging design.
>>
>>108053988
xAI is prolly running on a massive loss and these losses prolly lowered the valuation of the company as well, in the eyes of the investors or banks at least and these all result in higher cost for credits (i.e. higher apr etc.)
best way to hide this is by buying xAI at a high valuation by another company, which is his company spaceX. he did the same when he "sold" twitter for 44B to xAI about a year ago. The real valuation of twitter was around 10B at the time, but suddenly, it was valued back at 44B, which is what he decided to pay to himself basically.
>>
>>
>>108059039
>Elon is saying, fuck that
because this moron is an adult-toddler, he throws a fit if not everything goes his way while his companies live on gvt handouts that artificially increase their value, it's insane that investors stll believe in him given how he never delivers on his promises, ever.
>lets build a bunch of satellites, not need any land, any water, and no extra power.
the same retard who want to terraform mars with nukes... he says random bullshit all the time, don't take him seriously.
> I suspect he has already paid to have some rough designs done
yeah and he fired every engineer who does not give him the answer he wants until he find one that does, typical moronic muskrat behavior
>and already knows that it is possible technically.
reality does not matter to him, he'll just lie about the cost and bury the project after a few failures and modify the ledger while lying to investors as he does all the time.
dude should be locked up in jail for the multiple frauds he committed and the multiple lies he told, it's insane to me how so many people still trust this dude somehow???
reminder that he has no engineering knowledge, did not create spacex/tesla, was fired from x.com/paypal because he was a fucking moron, will never ever have security clearance as he is not a true american (he seethes A LOT about this one).
thank fucking god tesla is collapsing and xai is losing billions a quarter just like openslop, once we got rid of these two leeches the world will start healing.
>>
>>108060831
>it's insane to me how so many people still trust this dude somehow??
Because he still knows how to manipulate people into forming a cult of personality around him and his brands, which is the one skill every billionaire is best at.
>>
>>
>>108060365
I missed that space solar panels can't really run at 120C for very long. So you'd need to cool the solar panels too, but low emissivity panels don't absorb too much thermal radiation.
So you'd probably cool the GPUs and the solar panels to around 80C, then use a heatpump to make the radiator 120C.
>>
>>