Does It Make Sense To Put Data Centers In Space? Can They Really Cost Less To Operate?

[@jmc12331]

the collision of "scamming investors for seed funding with a space startup" and "scamming investors for seed funding with an AI startup" was inevitable

[@davescott7680]

It's always a good example of just because they have money, doesn't mean they're smart. We don't live in a meritocracy.

[@barnmaddo]

Now all they need blockchain and they'll be unstoppable!

[@whatsappvideos9665]

So investors are this dumb... Damn... I had no idea. Now I am totally confused as to how and what ppl and economies are.😊

[@derHutschi]

@@barnmaddo to misqoute "Life of Brian": she said "Blockchain, stone her" 😉

[@evelynashe8701]

​@@davescott7680While true, the stock market is very effective at selecting for people who are accurate at predicting the outcomes, though they end up with so much excess that their families can easily squander it, and that they can spend huge amounts outside of their field of understanding

[@Rubrickety]

Instead of this piecemeal approach of "data centers in space", "precision manufacturing in space", etc., I propose we aim big and simply put the entire planet in space.

[@zimriel]

So: disassemble Earth and turn its silicon and metals into a ringworld around the Sun. Kardashev 2?

[@monguskooklord7867]

this is a good ass comment

[@DavidArellanoSTP]

That’s something up to venture standards!!😂

[@DecidedlyNinja]

Give me 1.5 billion dollars, and in ten years I promise the Earth will be in orbit around the Sun.

[@scooby2142]

I would like to see MICROWAVE Spaceships . You know the type - 10 million years of space travel condensed into five minutes. With a grill attachment of course.

[@dakota5569]

Meanwhile Microsoft commissioned a nuclear reactor for power. Probably cheaper

[@yren3386]

Microsoft also prototyped throwing data centers into the ocean. Under water data centers, cooling problem solved.

[@acceptablecasualty5319]

​@@yren3386Not solved, but significantly reduced.

[@DJRonnieG]

No nuclear power to lower energy bills for plebs, but for corporate "A.I.", no problem :(

[@ForrestTessen]

How many times have I rushed out to the data center to replace a failed Ethernet cable or swap out a bad stick of RAM? Countless. These small but critical issues always seem to happen at the most inconvenient times-late at night, during high-stakes projects, or when you least expect it. Each time, it’s a reminder of how even the smallest components in a system can bring an entire network or server to a halt.

[@blshouse]

On call 24/7/365, but I can't simply commute to space to swap out a hard drive gone bad.

[@render1802]

Then AI will be creating new jobs! You telling me you don't wanna be a space IT guy? (just kidding)

[@jackielinde7568]

Eh, I suspect their answer to the failing hardware issue is going to be each node (box, attachment, rack, etc.) is going to contain several servers. if a part on one goes down, they'll just shut down that server and continue with whatever is left. However, that bus is going to be the biggest headache, as it looks like a single point of failure, and their table didn't inspire any confidence in redundancy. Upgrades are also going to be an ongoing issue, as hardware gets improved on a very short cycle. Are they just going to send up newer boxes with updated hardware and just ram it in somewhere, or are they going to have to eject existing nodes. Which leads to the last problem: DO THEY HAVE A PLAN FOR THEIR DECOMMISIONED HARDWARE? (Want to make a bet it's to just set it adrift in orbit around our planet? I'm placing my money on that bet.)

[@3ddan148]

@@blshouse but if you could, gotta admit, would be a pretty stellar commute!

[@andyalder7910]

It's not like modern systems don't have redundancy; I too have been sent out in the middle of the night to swap a DIMM that the CPU has already swapped with a spare.

[@exchable]

I'd love to be the IT guy who has to fly up there at an hour's notice when the next inevitable "Crowdstrike" hits.

[@CrazyRFGuy]

Meh. That only screwed people who ran bare metal windows. Anyone running a proper hypervisor (IE not windows) were able to recover with shoving a new clean and working vm. OFC you needed to also not store your primary data (and have backups) on something not windows. iSCSI storage SANs FTW. Windows is a find desktop for the office worker, not so good for a server.

[@jeromethiel4323]

Cloudstrike should be an example case for "checking or bounding your inputs." Software that just blindly accepts as input a file with no error checking is how you get cloudstrike (as far as i understand it). If the file is corrupted or formatted incorrectly, the software using that data should fail gracefully, not just crash. If i am incorrect, i would love to have someone inform me. Had some health issues, and stopped paying attention to the cloudstrike thing.

[@MonkeyJedi99]

Or when high energy particles strike the orbiting data center and scramble data.

[@jeromethiel4323]

@@MonkeyJedi99 Agreed. You have to account for soft failures. Which means you need ECC on your data, and perhaps redundant CPU's so that they can compare results to detect compute errors. Heck, i remember when some ceramic electronic packages turned out to be problematic, because the ceramic was slightly radioactive, and could "bit flip" gates randomly. When you are outside the atmosphere, the radiation goes up. And the higher the orbit, the worse it gets. When you leave the protection of the Van Allen belts, it gets even WORSE!

[@nikolaideianov5092]

​@@jeromethiel4323iirc some of the systems that are working with soft failiers can survive a scary amount of damege

[@engunneer]

A note about AI chips (NPUs) in phones and such: they are designed to /run/ the models that were trained in the big data centers. The training part still takes a huge amount of compute.

[@FuncleChuck]

Well that’s that idea here. The training would happen in space. The whole concept is a fraud on top of a scam, but there it is.

[@novantha1]

While not frequently used, there are still a few ways you can train on low power devices; evolutionary strategies, forward-forward networks, etc, can all train without back propagation (eliminating the need for extensive precision accumulate instructions), so you indeed “could” train on an NPU if you were determined enough. You are correct that it’s very rarely done, though.

[@iCore7Gaming]

Yes but currently datacenters are running these things on GPUs still. Chatgpt I'm pretty sure is still run on GPUs. I guess the main issue about these ML ASIC (NPUs) chips is memory limitations in both capacity and speed. But TPUs from google are used to train ML so we do have that. They still require LOADSs of energy, but far more efficient than the best NVIDIA GPU, the H200.

[@vylbird8014]

@@iCore7Gaming TPUs are far more efficient, though good luck getting hold of them. There isn't anywhere near enough production to meet demand, and what production there is already goes entirely to cloud providers who will rent out time on them.

[@toxaq]

Imagine the AWS bill if you accidentally selected the LEO-01 data center...

[@asandax6]

​@@squishy-tomato The edge of the observable universe is like 0.1% to the limit of an AWS bill.

[@ianmiles2505]

Egress charges.

[@tucan1309]

Wouldnt it be cheaper if it succeds? Like thats the whole point.

[@iCore7Gaming]

LMAO

[@iCore7Gaming]

@@tucan1309 Not initially. Upfront cost is quite high, and "fancy new tech thing" would give it an excuse to be VERY high. Consider that Sattlite internet is extortionate, and also starlink internet isn't cheap at all.

[@nazamroth8427]

My first thought when I read this idea: You want to put infamously heat sensitive and heat-producing things into orbit to help them cool better.... in the place where it takes the most effort to keep cool? And based on those renders, they are not even using a sun-shield?

[@Prophes0r]

The 250km^sq of solar cells they would need for a TINY 10Mw install would provide lots of shade. Of course, 10Mw is likely way too small to even be useful. Oh! And we have no way to get rid of even 1Mw of heat, even on the dark side. This is pure fantasy/scam material. Zero parts of this project are possible, except the part where they let you give them money.

[@coreytaylor5386]

and notice there isnt a single heat radiator while the ISS's silhouette is dominated by theirs.

[@mikekopack6441]

Dont the panels kind of act as a sun shade if you put the radiators behind them?

[@tombowen9861]

That was my first thought! "Higher delta T in space!" Yes. but you don't get to use Convection! lol

[@FerdinandFake]

Gigawatts of solar arrays should be sufficient sun shielding for anything

[@CorwynGC]

I love that since cooling is "more efficient" it will cost 0 dollars.

[@freekeefox]

"I'm surprised at the people who funded this" is an such nice way of saying "Oh my God what a dumb idea"

[@nitehawk86]

This is 100% a finance scam. Especially with the "AI" attached to it. I am surprised they didnt say they were gonna mine crypto.

[@MonkeyJedi99]

Look, the "ghosts" of the tachikomas need to be stored somewhere.

[@CumulusGranitis]

It is indeed most subtle. But everyone's favourite Scotsman, as usual, has hit the nail square on the head.

[@Adventures_EC]

totally

[@RFC3514]

"It was a very... brave decision to move forward with this project."

[@Antarez666]

As someone who works with a company with a large number of DCs across the Americas, I'd love to see their poor L1 techs trying to arrange a site access case for devices literally in space. God forbid someone locks themselves out of a device with a networking change, that'd be the most expensive remote hands ticket you could imagine.

[@JonBrase]

"We bill for colo visits to the tune of $50 per pound of equipment and personnel arriving, and require 30 days notice to ensure that adequate life support resources are in place for the extra personnel. Our on-site tech has cabin fever since we combined our support and ops roles a year ago and fired the ops tech, so the cost of the visit is waived if you bring a mental health professional with you. Please speak gently and do not make sudden movements when interacting with the tech. The tech has his own food supply that was last refilled on ERROR: INVALID DATE, please do not feed the tech." Some elements of the above are loosely based on a true story.

[@PiDsPagePrototypes]

Techs will live onsite in their own habitat. Then they'll figure out how to grow Weed, and how to use the plant waste to make Booze. A week later they'll declare Independence and ransom the data for recognition of their sovereignty. It'll later come out that they were acting under secret corporate instructions, as head office suddenly shifts from being incorporated on the ground to doing so on the new orbital nation, and never paying taxation in any nation ( "Oh, you want to tax us? Go ahead, we'll delete your backups first,...")

[@carlettoburacco9235]

Well.......I work in industry. Many years ago I had to take a plane from Italy and go to Brazil to give the command "format c: /S........ a: ........Install" Some things never change.

[@sdebeaubien]

Heaven forbid you have an incident like we had last month with the CrowdStrike security software update. Every single server had to be booted in safe mode, the patch backed out, and then go on. Can you imagine? "Uh boss, I gotta go in space to reboot about 10k servers. Shouldn't take more than a day or so ..."

[@JonBrase]

@@sdebeaubien Fortunately, by the time that happened, we had already migrated all or customers offsite in preparation to decommission the site, and we weren't using CrowdStrike internally, so I didn't have to deal with that at all.

[@slowercuber7767]

13:25 LOL did you do this whole video just to make the “data bus” Dad joke? Worth it.

[@andreirachko]

I don’t get it

[@warriordx5520]

​@@andreirachkome neither

[@DJRonnieG]

​@@andreirachkolook at the specs on any PC. A term like "Front-side Bus" or system bus will yield some answers. Even the earliest hobbyist microcomputer like the Altair and IMSAI 8080 use the S100 bus. It's like a bunch of electrical traffic lanes. All of the card slots on an S100 bus backplane (planar board is basically an old word for motherboard) are on a common bus. For example the CPU chip resides on it's own board, the RAM is on a neighboring board, and the serial ports are on a third board (printer and console goes here). The CPU is sort of like the traffic cop in this context (among other things). Old epsiodes of "The Computer Programme" on BBC cover a lot of these fundamentals. 'The Computer Chronicles' is more current but uses more generic PC terminology that abstracts further away from basic concepts.

[@andreirachko]

@@DJRonnieG damn, thanks man

[@Michaelzehr]

They forgot to add "block chain" to their pitch. But seriously, biggest problem in a data center is keeping things cool, and one of the hard problems in space is keeping things cool. I'm not sure what problem they think they're solving? Electricity is likely to get cheaper over time on earth, but more expensive over time in space, as you have fixed equipment that slowly degrades. Speaking of which, AI chips are changing very quickly. Those chips, even if they are shielded, will not be state of the art in 20 years. And with a falcon launch costing $50m, I'm not impressed that they've raised $2.4m (which will be for design and analysis). (I'm not saying anything particularly insightful... but it's really hard to beat system trends. Here's another one, just back of the envelope, a water cooling rig for a PC can weigh 1kg and cost $200, but the launch cost might be $2000, so that's extra cost to take into account. Even though a water cooling loop with a fan and a ammonia-based radiator are radically different, it might still be illustrative of how much more expensive it is to cool things in space.)

[@Prophes0r]

Cooling might be the biggest problem. But only because it is literally impossible with our current tech. Power generation is also impossible, but only "Technically" impossible, since we can't really put the 250ish km^sq of solar cells into high orbit that even the smallest usable servers are going to need. Maintenance is also technically impossible. We have no way to service stuff in high orbit. And that's where this monstrosity would have to be because anything that big is just going to deorbit itself. The chips used are not QUITE impossible. We do have "modern" CPUs in space. They are just heavily underclocked, double/triple redundant, have boosting disabled, and also can't use their highest speed interfaces. It doesn't really make sense to do AI training on an 8-core 2GHz system that only uses PCIe 3.0. Everything about this is fantasy. It's just AI Bros scamming investors with buzzword bingo.

[@blacknoir2404]

data center property taxes at least 😂

[@obsidianjane4413]

"Those chips, even if they are shielded, will not be state of the art in 20 years." They will be completely obsolete in less than 2 years.

[@TheClintonio]

> They forgot to add "block chain" to their pitch. That's so 2018. It's AI now.

[@FerdinandFake]

It's all a bunch of investment sniffing baloney. About the heat dissipation they claim it's more efficient because of the low temperature of space, which is bullshit both in theory and practice.

[@grantbent]

Incredibly entertaining to have Scott Manley launch the first episode of a space-oriented Shark Tank.

[@heaslyben]

Five years from now they can link to this video and say, "Well...yeah...that."

[@linecraftman3907]

Another fun part is that this 4km wide data center will literally be bigger than the moon in the sky Would be pretty cool to see it with naked eye

[@DJRonnieG]

Gosh, I kind of hope it fails on that count alone. I guess I'd be satisfied with limited use, but this sounds like something that would kill my hobby unless affordable space telescope rentals become available.

[@ArKritz84]

Only, and I do mean ONLY if they call it Skynet.

[@HALLish-jl5mo]

That's already a UK military satellite constellation

[@JulieanGalak]

@ArKritz84 - you win the internet today

[@AndreasPeters-r3e]

@@HALLish-jl5mo and a chinese AI

[@noizz4]

what about neuromancer

[@مقاطعمترجمة-ش8ث]

@@HALLish-jl5mo *IDF

[@Wayde-VA3NCA]

Time to update the 70's wisdom about never underestimate the bandwidth of a station wagon full of tapes.... For the 2020's ; a rocket full of SSD's??? All these decades though, and still hasn't done anything for the latency!

[@tangydiesel1886]

Glad someone said it!

[@negirno]

SSDs are lightweight, but radiation hardening could add up on weight. Maybe having a small spacecraft full of NAND chips would be better?

[@Prophes0r]

Latency has improved TONS at the low/mid end. And at the very top end too if you have access to something with supercruise or even (sub)orbital. The only place where latency is about the same is if you were/are going to use a private courier and/or private jet. Then the transit time is still about the same as it was in the 70s. Also, it is WAY faster to plug a "Data cube" full of SSDs to some 400/800Gbit fibers on either end to fill/drain than it was to write/read those tapes... Or even just plug in 2 dozen U.2/U.3 drives after getting them out of the post-box. In fact, "sneakernet" is the preferred high-end data transfer for large amounts of data to/from cloud providers like Amazon/Google/Microsoft even today. Amazon has has a services called "Snowball" for as long as they have had datacenters. Go look it up, its kinda bonkers. Most places also let you just mail drives in now too. At one point I think Amazon actually had Transit Vans with Server racks full of SSD storage. They would pull up to your current datacenter, Unroll fiber between the van and your rack, and plug right into your gear. The SSDs never left the van. Then they unplug and drive it back to a datacenter. It's the only realistic way to import 100PiB of data.

[@Wayde-VA3NCA]

SSD's are (mostly) NAND flash devices with a disk controller attached...

[@ReNeyer]

Rocket Mail full of USB Sticks!

[@shagiephoto]

Never underestimate the bandwidth of a station wagon full of tapes hurtling down the highway. -Andrew Tanenbaum, 1981. Even SETI@home used this - "Data will be recorded on high-density tapes at the Arecibo telescope in Puerto Rico, filling about one 35 Gbyte DLT tape per day. Because Arecibo does not have a high bandwidth Internet connection, the data tape must go by snail-mail to Berkeley."

[@kitefan1]

RIP Arecibo. Maintenance is a good thing.

[@H3xx1st]

Transferring via "Sneekernet" - Putting on your shoes (sneekers) and carrying a data storage device to another place.

[@Minty1337]

oh my, a silicon valley startup with a name made of between 1 and 3 industry relevant words coming up with a completely unreasonable idea with no plan to finance it besides investors? where have i heard this before?

[@Codysdab]

I like bubbles too.

[@olliestudio45]

too fake for thunderf00t, cue billions from investors

[@oatlord]

Complete with CGI. That's how you know they're serious.

[@Minty1337]

@@olliestudio45 this is just solar roadways 2, now in space

[@gehteuchnixan69]

Why do I still work a 9-5 when venture capitalist are willing to invest in absolutely asinine ideas

[@jackielinde7568]

"Data Death Star" - Now I'm picturing a moon sized satellite wiping out a small city and a tech on board says, "Here's those files you requested."

[@personzorz]

Very intense high frequency radiation carrying a very, very high precision number.

[@chpsilva]

"That's no moon..."

[@simongeard4824]

"Death Star" was the nickname for IBM's "Desk Star" range of hard drives, which had a *very* poor reputation when it came to reliability...

[@JohnMcNeel]

PREPARE FOR DATA TRANSFER

[@chpsilva]

@@JohnMcNeel /*wookie growls* GRWAAAAAAAAL

[@Sweenus987]

Humanity's first giga structure in space is going to be a giant computer full of cat videos...

[@whanua98]

or cat girls anime

[@Johnwashere-dt2ov]

And full of flat Earther’s convinced that there is no “space”. The irony

[@H3xx1st]

Well at least it's better than the early internet being powered by pr0n.

[@ТестТестович-г2о]

Oh yeah, that's something I would invest in!

[@ТестТестович-г2о]

@@whanua98 and

[@zarbizaure]

AOCS engineer here. I'm wondering as to how they could keep such a massive structure stable. I think you can forget reaction wheels and CMGs, and monoprop thruster as it would be a massive propellant budget to counteract the SRP (solar radiation pressure). I see 3 possible solutions: 1) They built it to be passively stable, i.e. by folding back slightly the solar array (have it flying in the solar wind like a badminton birdie). No roll control this way, but maybe this is acceptable if the servicing shuttle can approach from any angle / if the HGA to Earth are multidirectional or orientable. Cheapest and most robust solution in my opinion. 2) You use electric hall effect thruster placed at the end of each boom to produce torque (and also do minimal maintenance). Downside is the need for a full propulsion system + reduced payload power when it is active. 3) you set each of the external solar array on servomotors and the AOCS orients each of the array to control the SRP and compensate for the offpointing. Very slow control but could be enough to be stable at this scale. Option 2 / 3 offer more controllability but are an order of magnitude more complicated than what is done on commercial satellites. Ionic thrusters can be quite prone to malfunctions and need redundant systems, the same goes for large solar array actuators. (Anyway: I don't trust this project at all, but interesting to look at it)

[@feha92]

The solar wind idea is good, considering its size I could see why designing it using principles from a weather-wane could potentially work (assuming it is far enough away from earth). Does not even need to truly be a static profile that leads to a certain orientation, as servos are likely to be rotating parts such as the panels, and I imagine you could rotate them non-homogenously in such ways that the asymmetric forces (from solar wind and potentially even light-pressure) induces the spin you desire. Meaning you get an active control system. I do think reaction wheels could work, if done right. Does not need to have a high output, so long as there is little enough other forces to impact it, and iirc the rotational saturation/capacity can be fixed by flipping the reaction wheel before it saturates and decelerate it instead (and if not, then maybe pushing against solar wind to desaturate it would still work?). But if anything, it would probably be best to keep it only as a redundancy and maybe aid if some extra force is needed "quickly" (likely still measured in days or at least hours). thrusters might work, but then you would need to worry about fuel. I would be surprised if there weren't servos. No idea what srp stands for (I would have guesses "solar radiation pressure", but you already covered forces from solar wind/radiation pressure in your first point. So probably something like "Servo Rotation? Pointing?" instead), but maybe it has to do with intentionally affecting your spin through same mechanisms tidal-locking works?

[@YodaWhat]

​@@feha92- You don't get solar wind in low earth orbit (LEO) or even in medium earth orbit (MEO). You barely get solar wind even in geosynchronous orbit. The rest of your comments are on target. 😊 And yes, SRP almost certainly stands for solar radiation pressure.

[@judet2992]

Or use the solar sail effect to help and make it a statelite (no that’s not a typo)

[@kahlzun]

Since it could deal with rotating slowly, perhaps you could leverage the earths magnetic field, and get effectively a propellantless RCS thruster by charging cables on the edges of the panels

[@feha92]

@@YodaWhat Why would you keep them in LEO? is this thing not going to have huge drag-forces if you do that (because of how huge and flat it is)? I were expecting something well past KEO - or even GEO - tbh, if for no other reason than because correcting the orbit to avoid collisions would be way too costly otherwise (anything up to and around GEO are rather crowded orbits, particularly further into the future - though I honestly don't know _how_ crowded they would be, but considering how you would probably want a polar orbit with matching precession (or servos, I suppose), it feels like it could become an issue). actually, the video even addressed this iirc, coming to similar conclusions.

[@davidgunther8428]

Radiation from the van Allen belts just adds to the AI's creativity (hallucinations). 😅

[@kaelandin]

Schizophrenic AI model when

[@waywardscythe3358]

@@kaelandinthey all are lol

[@Sableagle]

@@kaelandin AI suffering temporal lobe epileptic seizure, also known as Skyrim?

[@jessstrap2088]

The A.I. needed to expand their minds.

[@charlescouncill]

Radiation=shrooms

[@TheManLab7]

0:29 Well it's definitely more than "1.21GW's!!!?" that's for sure. The only good thing is you'll easily be able to run 4 fluxcapacitors with power to spare.

[@yodaman8015]

Well hope they don’t time travel, orbital velocity is much more then 77mph

[@Hiznogood]

- “Hi, I have a faulty raid controller according to the ILO. I’ve uploaded a diagnostic report!” - “Yes, that is correct! We have a replacement in stock. Do you want a technician to replace it?” - “That would be great!” -“ Sir, could please give us the address where the faulty system is?” -“Well … just look up …”

[@Prophes0r]

Nahh. It can only exist in an AI Bro's fantasy world anyway. Just hand the courier a card with "Tech Investor Buzzwords" on it and the Bros will come to them.

[@allangibson8494]

The biggest problem with data centers is COOLING. The biggest problem with satellites is cooling in a vacuum. Cooling is much easier at ground level (or under water). The second problem is slow data rates over radio links compared to ground based fibre optics (at 40Gb/s at the moment per fiber).

[@ac1dflare937]

Dear god thank you. The fact you are the only other person who points out the bloody obvious is staggering. Space is a vacuum meaning and data centre would be dead in hours from overheating with literally no way to cool it. What's insane is everyone here must have heard of a thermos and yet miss the reason it works as an insulator is it's a VACUUM

[@Varadiio]

I know they're not as intense as AI cores, but data transmission is not free, either. GPS and similar satellites are broadcasting

[@ArnaudMEURET]

@@ac1dflare937Did you even watch the video? Scott extensively covers the cooling issue. As for AI training, the bandwidth is a non-issue.

[@allangibson8494]

@@ArnaudMEURET For ANY data centre, bandwidth is critical. Stock Exchanges dumped satellite links because they were much slower than fiber optics. The ISS has trouble with cooling - and it doesn’t have the megawatts that data centers routinely consume to radiate.

[@Exilum]

​@@allangibson8494The cooling is indeed a non-issue if they got the math right, which they likely did. As Scott said, the actual issues are more budgetary and structural. Large AI training workloads have the advantage of lasting weeks at the very least, which makes them particularly fit for a datacenter that may have a few days of data transmission delay.

[@bow-tiedengineer4453]

I love the data shuttle thing. It reminds me of that old saying, never underestimate the bandwidth of a van full of tapes traveling down the highway.

[@Andrew-iv5dq]

Good point about the cooling. Anybody notice how we NEVER have the HVAC guys crawling around on the roof? Yeah, me neither.

[@dracula3811]

Actually, I see it all the time. There are rtu's that I run electrical circuits to. That's roof top units.

[@puddles5501]

@@dracula3811 I think you'll find that comment was intended with a facetious overtone.

[@iLLadelph267]

​@@dracula3811same lmao, even residential has stuff we do instead of roofers

[@MrGoesBoom]

It's great that so many companies are looking at space and wanting to develope tech up there, but it seems like most of them really really don't think things through, as you point out in this vid. I figure one of the things they also didn't think about is maintainance. Over the course of 10 years all those computers/servers/etc are definitely gonna be needing replacement and work done on them, both from wear and tear and just the march of technology.

[@MrUltrAdaman]

I guess that's part of the reason why they're showing it as modular. Once a module has a specific number of failures it gets replaced

[@MrGoesBoom]

@@MrUltrAdaman point. though I wonder if they factor that into the costs they're projecting

[@Sableagle]

@@MrUltrAdaman Units just jumping off and de-orbiting themselves into the Pacific Abyss? I hope their security's good. Some hacker deorbiting all their modules to create a firework show for 4 July would be a bit embarrassing. I wonder whether they could de-orbit one onto Moscow every couple of hours for a week.

[@nopenoperson9118]

@@Sableagle how the hell did you drag that into this, please for the love of god stop champing at that bit for ten seconds

[@tomcapon4447]

When you see a "startup" that doesn't answer basic physics questions, it's because they are intentionally scamming investors.

[@feartheghus]

Data centers in space make perfect sense for space faring civilization. For here though, almost every single use for it is on the ground in a planet blanketed by atmosphere. It must use wireless signals which will involve high ping, low bandwidth and would be less reliable than data centers on earth with wired connection to the internet. You might say “use lasers to transfer data faster and more reliably” but we have an atmosphere so it instead becomes incredibly unreliable the moment it needs to beam data in either direction by that method. Would be awesome for sending stuff through space though. Edit: the argument regarding power makes sense though, I forgot to add the good sides because they can’t put weight the bad while we are still based here on earth. Solar power is incredible in space even though it’s crappy and inefficient on the ground, and lasers to send data from space craft to other space craft with direct line of sight is great.

[@Wileybot2004]

Space data centers could be useful for exploring other planets. For example if we wanted mars rovers/robots with Advanced ML models rather then have each one equipped with compute have them communicate with a data center in orbit around mars. Then you could have them be able to act on data from the others and not have to deal with the latency from a earth based system

[@riccardob9026]

This remind me about an European research project I tried to fund few years ago. The idea was to develop a super computer (aiming to EXAFLOP and beyond) using superconductive digital electronics (SDE). SDE chips are already produced and sold, although only in very niche application area (e.g., ADC for radar signals). The development of superconducting electronics is lagging behind silicon because of historical reasons and currently superconducting chips are at an integration level comparable to silicon in the 80s-90s. Power consumption of SDE is 10,000 times smaller than the consumption of silicon electronics and in literature one can find descriptions of processors running almost to 100 GHz. Of course, superconductivity has the problem of keeping the temperature low, but even on Earth it is possible to have a power consumption 10-100 times smaller (cooling plant included). In space we expect to get better performances, but it is just a first guess. Yes, cooling in space is a huge problem.

[@The.Heart.Unceasing]

putting anything that need to be kept at cryogenic temperatures in space is a very bad idea, you will either need hundreds of square kilometers of radiators, or so many heat-pumps to bring the cooling loop temp up that you would loose a lot of the energy economy brought by using SDE chips.

[@riccardob9026]

​@@The.Heart.Unceasing Yes, we are aware of the difficulties of keeping the stuff cold. A YT comment is not exactly the best place for an in-depth discussion about this, but let me say that there are many variables that can be controlled and this opens some possibilities. The idea is neither obviously impossible nor obviously possible, some exploration is needed to understand what can be actually done. Keep in mind that ours is a research project at low TRLs (

[@deep.space.12]

@@riccardob9026 I hope you can help to clarify my own understanding: Quantum computing can't use high temperature superconductors as qubits, because they need the actual temperature to be low, so as to reduce noise/decoherence. Meanwhile SDE cares about the superconductivity, not the temperature. Therefore your discipline can utilize high temperature superconductors, as long as there are proper research/engineering to adapt high Tc materials to SDE circuits. Would this be an accurate summary?

[@adityakulkarni4549]

@riccardob9026 That sounds super duper interesting, could you tell me more?

[@wtfdddf]

​@@riccardob9026 please make a blog, this sounds like an interesting area of research.

[@bendeleted9155]

Yes. Yes, I will invest my money into putting equipment inro orbit that becomes obsolete before it is assembled with the equipment that follows, costs more to launch than it's worth, then falls out of the sky during a solar storm. Yes.

[@adama7752]

It told me a few scrolls more than it should to get to this comment exactly.

[@michaeltorrisi7289]

Also computers in space have to be built with ridiculous redundancies because the lack of atmospheric shielding means that they take a lot more radiation and are very susceptible to bit flip issues. You basically have to run every computation twice and compare the results, with a system for determining the point of error in case of a mismatch. Doubling the complexity of every task is fine for like, a laptop. It's not fine for a few thousand servers.

[@fat4eyes]

Silicon Valley baybee!

[@Prophes0r]

The problem of obsolescence is honestly pretty low on the list. The bigger problems are... How do we put hundreds of square kilometers of solar cells into orbit? The ISS only generates ~100kilowatts with it's 2500m^sq arrays. A 10Megawatt AI trainer would likely be too small to be useful. We literally don't have the tech/experience to do it even as an experiment right now. How do we get rid of 10Megawatts of heat? Even if it wasn't in sunlight 24 hours a day (which is a supposed "benefit" of this whole dumb idea) it would be impossible with current tech. How do we do maintenance on this gigantic imaginary space object that we clearly need to put into high-orbit because 250km^sq of solar cells isn't going to work anywhere near Earth? What kind of Magical CPU/Accelerators are we using that are going to work out there without having to downclock everything to hell and back like we currently do with all the computers in space? This is PURE fantasy/scam/bullshit to scam dumb AI Bro investors.

[@ZaphodOddly]

​@@Prophes0rthe heat problems alone are phenomenal 👍

[@MarkGoldfain]

So great to have Scott Manley bring this pi-in-the-sky dreaming back down to earth.

[@stellsy3496]

dont forget that the thing is now a giant photon sail (be it a small and heavy photon sail but i still think you cant just ignore the effect) and will just leave on its own, so orbital upkeep cost is still a big factor

[@Br3ttM]

It's not going to float away, because it's going to be in orbit of Earth. The force from sunlight and solar wind will just slightly change the orbit, unless it actively changed orientation for certain parts of its orbit to slowly pump it up. Even a solar sail orbiting the sun doesn't necessarily get blown further away, it can remain at a fixed altitude by simply moving too slowly to orbit at that altitude, and using the force on the sail to make up for the difference, rather than pushing it further out.

[@marrioman13]

​​@@Br3ttMIt'd actually have a sufficient amount of atmospheric drag that would outweigh the solar sail effect

[@YodaWhat]

​@@marrioman13- That depends on how high the orbit. This bird does not need to be in low earth orbit where atmospheric drag is a considerable factor. It does not need low latency like Starlink, so there is absolutely no reason to put it in low earth orbit and risk it getting hit by all the junk already there. But maybe you didn't listen to to all that Scott said.

[@stellsy3496]

@@Br3ttM just moving too slowly for the orbit wouldnt work if youre not orbiting whats providing the force, remember either the force isnt in the direction of the orbit, or it alternates based on what side of the earth youre on. but putting it in an orbit that doesnt really go around the center of mass of the earth, so slightly offset kept up by the photon pressure and stabilized by slightly tilting the panels, just a few degrees to slightly vary the speed in the normal part of the orbit to ballance the force of the earth trying to puch it on center and the sun pushing it away. im not sure if i explained it properly

[@fred_derf]

Unlimited "Free" power but heat dissipation will be a big issue... (as will lag and trying to maintain a reasonable uptime (e,g. SLA 99.9%).

[@Taladar2003]

Replacement parts will be a problem too.

[@mikicerise6250]

Maybe the back of the solar panels could double as dissipators.

[@termitreter6545]

Its funny how things are free if you ignore the costs :O

[@volvo09]

Yeah that is a LOT of heat to dump.

[@fred_derf]

@@mikicerise6250, writes _"Maybe the back of the solar panels could double as dissipators."_ There is a lot of heat generated by a server farm, they would need a lot of dissipaters.

[@lukecresante]

I saw that this was in the YCombinator batch but then again YCombinator now has 4 batches per year and I believe is accepting more and more startups. This may be their pitch to maintain a good relationship with the specific founders or say "we funded hardward startups" as they skew super software based, especially AI.

[@grelymolycremp7838]

I love silicon valley startups, some of the best comedy of the 21st century.

[@kazioo2]

True, I consider this idea to be absurd in the near future, but there is that percentage of those comedic startups that turn out to be not understood by the general public (us) and then we don't laugh anymore. We had the greatest experts in the fields completely ridiculing some startups that succeeded, because they missed some tiny detail (not necessarily technical, it can sometimes be sociological, political) why it could actually work.

[@alexdhall]

There's no way in heck NASA or the US would allow this. The national security risks, and other obvious threats (space debris, radiation, etc) make this a really bad idea...

[@puddles5501]

@@alexdhall they have no jurisdiction to prevent it, thus starlink exists.

[@grelymolycremp7838]

@@kazioo2 Very true, maybe when California/rest of the world realizes that they'll run out of water cooling these datacenters they'll dump em in space; all it will take is a few billion lol.

[@BigDaddy-yp4mi]

@@kazioo2 Luckily, the laws of physics never change.

[@FourthRoot]

A 16 km^2 radiator is going to be impossible to protect from debris.

[@termitreter6545]

Yet the confidence of bad startups is undestructible

[@kaelandin]

Especially at LEO

[@glenyoung1809]

And that will make it the biggest Kessler bomb man will have made up to point if this idea is built in LEO.

[@celivalg]

the only way I can see that work is by having a lot of subsections you can switch off if you detect a leak, and ahving enough spare cooling capacity to make it work, or just slow down the processing if you can't keep it cool.

[@alexdhall]

Don't forget it would be an easy target for various ASAT weapons...

[@redmondhenry398]

I love the idea that a super advanced AI is powered by a solar panels in a polar orbit and our only chance to fight back is during the eclipse

[@michaelimbesi2314]

There’s a couple more issues with this: 1. The entire point of a data center is to have the data securely stored and readily accessible. To do that with a satellite, you’re going to need a whole network of ground stations to get consistent access to that data. They’re going to need to be redundant to make sure that one is always visible to the satellite even if one goes down or is blocked by weather. And the downlinks need to be hooked into the internet. The latency you’re going to get, especially when the satellite is over the southern hemisphere, is going to be awful. 2. The bit rot from being exposed to cosmic rays is ultimately going to be a dealbreaker. Because of the high latency, the only use case for that data center is longer-term storage of things, but the high bit rot makes it bad at that. 3. The minds behind this seem to have completely forgotten that forms of energy other than solar power exist and are in such common usage that a ground-based data center doesn’t actually *need* any battery storage. The grid provides plenty of power 24 hours a day. 4. Those radiators need to be very sturdy, because even a single micrometeorite hole will lead to the refrigerant inside them leaking out and rendering the entire system inoperable. 5. Where did they get those electricity costs from? Because most data centers are in Loudoun County, VA, which has pretty cheap electricity. Those prices seem a lot more like CA prices than VA. 6. This entire project relies on the current AI craze continuing for a very long time. Given the way that similar tech hype-trains have turned out, and how badly LLMs actually perform at tasks that require accuracy, there’s a very real chance that in three or four years, the entire reason for this idea will be completely obsolete. (If you don’t know what I mean, think about the Metaverse, Blockchain, the Internet of Things, self-driving cars, NFTs, and the allegedly “imminent” switch of all transactions to cryptocurrency (promised to occur by 2010, 2013, 2015, 2017, 2019, 2022, etc) and how all of those have turned out.

[@MonkeyJedi99]

"Solar power in space could work 24/7!" - So beam it down to a ground collector, use what you need, sell the rest. Sure, the FAA might get uppity about having to route around the column of high energy, and you may get a scavenger problem on site with all the pre-cooked birds landing...

[@checker297]

i see this as two actual pitches, one for a giant solar array and one for a space datacentre to use that electricity. The solar array has some potential as it is already tested proof of concept with the ISS etc... but the datacentre has so many hurdles to overcome that I feel like by the time any progress has been made the R+D costs will have blown out so much it is not viable even if it doesnt require ongoing costs and the components outperform earth dc's.

[@adrianthoroughgood1191]

Solar power in space transmitted to earth is an old idea but sending the power to earth is the hard part. Maybe someone had the idea "what if we put the thing using the power in space so we don't have to send it to earth?"

[@filanfyretracker]

And do not forget that anything transmitting and capable of receiving wirelessly is going to be vulnerable. When not If such a datacenter satellite would have its servers hacked by some people with off the shelf SDR equipment and a large dish from a junkyard like those big ones commonly linked to rural America for getting any TV at all.

[@MrMattumbo]

as to your first point, I could actually see this being a good pairing for Starlink. At some point if Starlink reaches wide enough adoption it could start to make a lot of sense to host data in space and use the laser comms to beam it across the constellation and then down to earth. But for there to be latency benefits to that system it'll take a data center constellation which isn't constrained by a sun-synchronous orbit so we'll be waiting on something like small scale fusion to make those satellites feasible.

[@tlxyxl8524]

When I first saw the title, I thought about 3 major problems of this approach: communication, cooling, and regulation. For communication, wired communications are typically more reliable and have higher throughput than wireless ones. A data center will need big throughput. For regulation, many nation require the infrastructure processing of PII of their citizens to be located within their jurisdiction, and space complicated that. For cooling, the problem is already discussed in the video

[@YodaWhat]

Yeah, this idea is not really for regular data centers that need lots and lots of ongoing data traffic. You don't need that for training AI you can send all the initial data up on hard drives that you're sending up in the first place. The resulting AI model is much much smaller and can be transmitted or carried on a hard drive. Also, as Scott mentioned, processes like Bitcoin mining do not require huge data transmission capacity. Which is okay, nobody said this is a one-size-fits-all solution or silver bullet.

[@MediocreHexPeddler]

My first thought upon seeing a 4km-across solar panel array was that keeping it oriented properly would be impossible. You can't rotate something that large and fragile with reaction wheels, so that leaves thrusters, and you'd have to put those near the edges. The slightest discrepancy in thrust balancing would send waves throughout the entire structure and very quickly cause it to wibble-wobble itself apart.

[@oystercatcher943]

But it’s not really rotating to do that. The Earth is rotating and a spacecraft isn’t locked to that. The spacecraft is just fixed relative to the sun. I think that might be the only part of this that does work!

[@w0ttheh3ll]

@@oystercatcher943 you still need attitude control. lift from the residual atmosphere will introduce a systematic torque and small collisions will produce random torque.

[@w0ttheh3ll]

no reason you couldn't put reaction wheels near the edges. but yes, the structure would have to be very rigid. a huge engineering challenge, unless you assemble it in space.

[@oystercatcher943]

@@w0ttheh3llgood point. This is going to be one of those factors that depends critically on altitude of course. I’ve no idea how significant that drag or torque would be. But a 4km wide solar panel at 100km perpendicular to the atmosphere could experience torque given the differential atmospheric density across the spacecraft!

[@pattheplanter]

Isn't it going to act like a solar sail, adding torque across the entire array?

[@truckerallikatuk]

Lots of easy solar power vs the insane issues cooling the things. Yay...

[@personzorz]

Plus, you know, enormous launch cost.

[@kaelandin]

@@personzorz As with all stuff now they're assuming using starship

[@personzorz]

@@kaelandin And not only that, assuming the eight year old talking points version of starship. If they did the same for the talking points version of falcon 9 they wouldn't need starship.

[@dorfl6870]

I am ballparking 10,000 square meter of better than today heat radiators.

[@unsafe_dB_level]

The "cold vacuum of space" would be my biggest concern. With no air, all the heat would have to be dissipated with passive radiators.

[@-yttrium-1187]

To be fair, The temparature of space is the result of your spacecrafts albedo. Which means you can cool your computer down to liquid helium temparatures and use helium as a heat pump without any intermediaries. its the best medium to do quantum computing if there's a market for it.

[@SeanBZA]

@@-yttrium-1187 But also equally hard to cool, as only radiation to carry it away, and you need massive thermal shielding to provide thermal barriers.

[@brucebaxter6923]

@@unsafe_dB_level Oh for gods sake, you turn it sideways. Your solar panels put your puter’s in shade already, so it has zero area being hit by sunlight and the entire area being exposed to -250c radiative temp

[@Ph33NIXx]

I had not considered that 🫢 First thought was it was going to solve the cooling issue.. when in reality they would need big radiators

[@thanos879]

Does an incandescent light bulb get hot in space?

[@NewfieNewbie80]

I’ve officially been called a Space Nerd by Scott Manley. The first thing out of my mouth when a friend told me about this was “what are they going to do with all of that heat?!” Any gamer knows how much heat a GPU can pump out. My own GPU can easily reach 80°C under high load. It will be interesting to see how they rig up their radiation coolant loop directly to the GPUs’ liquid cooling loop. I suspect it is the only way to properly attempt this endeavor.

[@svr5423]

I heard there was a company specialising in continuously releasing helium from gas tanks. That could be used for cooling :).

[@autohmae]

the graph indicated no liquid cooling needed...

[@link12313]

14:00 the neural network ASICs in consumer computers are only for running a model. They are not able to train one.

[@recklessroges]

I think you're right about this one. It's "solar roadways" all over again.

[@Prophes0r]

No. Solar roadways were dumb because they required a scale that was way too big to actually work. This is just flat impossible. We are several orders of magnitude off from being able to build it, and since it will be hundreds of square kilometers for the smallest viable one, it's gonna be in high orbit, where we couldn't assemble or maintain it anyway.

[@justindie7543]

@@Prophes0r Solar roadways made absolutely no sense at any scale.

[@SimonWoodburyForget]

Solar roadways felt a lot more stupid, because even a normal person could tell it's a stupid idea.

[@jesseli9868]

A good metric for if an idea is dumb is if there's a different approach that gets all of the benefits while being better in at least one other way. For example, solar roadways was dumb because you could literally just build the solar panels on the side of the road rather than in the road, have the exact same performance, and have the system cost a lot less. Space data centers are probably dumb because you can get the same benefits of cheap cooling and endless space by building the exact same 4km^2 solar power center, but on the equator in the middle of the ocean (or more realistically just off shore). Actually, Microsoft kind of already did this with Project Natick and found that underwater data centers would be more reliable due to inert atmosphere (they kept an atmosphere of pure nitrogen, vs. normal atmosphere with oxygen) and more consistent temperature. Plus, at least now it's somewhat feasible to send a technician out to repair stuff. Granted, space has the benefit of having more solar power available - but this can be solved by just building 3~5x the panels in your floating data center. Now, if somehow the cost per kg to LEO drops low enough to be comparable to the cost of actually making solar panels (their white paper would suggest something like $10~30 per kg), then this balance shifts. That's a *really big* if - 2 orders of magnitude in cost reduction below Falcon Heavy would be huge. But in a sci-fi world where the cost to LEO is only a few times more expensive than express air shipping around the world, we'd have to really recalibrate our intuitions on what's economical to do in space v.s. on land.

[@VestedUTuber]

Yes. Go ahead and train AI in orbit. That way if a robot uprising happens all we'll have to do is wait for a solar storm.

[@buildacubesat]

I really enjoyed your commentary on this, as I too hope for a future with space-based industry. Maybe sharing your thoughts on space economy startups could become a recurring thing for this channel?

[@gavtriple9]

Giving a whole new meaning to cloud storage

[@johnhelt5475]

13:25 "Data Bus" LOL!!! Love it.

[@johnmoruzzi7236]

A big floppy disk….

[@PaulTopping1]

Scott's final statement is interesting. Since only one out of ten venture-capital-funded companies are successful, the odds of picking the right one to work for aren't good. Especially since jobs at the most interesting ones are commensurably competitive. Also, the success ratio for these companies from the perspective of an employee is probably much worse than 1 in 10. Many may be successful from the VC's point of view by being profitably sold for parts to some other company.

[@Karatorian]

A petabyte by rocket has got the most sophisticated application of good ol' "sneaker net" ever.

[@Zarphag]

A RamRocket

[@bphenry]

Never underestimate the bandwidth of a station wagon full of tapes hurtling down the highway.

[@christiancampbell466]

@@bphenryI thought you were going to say hurtling down from space.

[@VekhGaming]

13:40 I would've called a "Universal Space Bus".

[@nothanks3236]

Sounds like a great idea until all your data is zapped during a solar storm.

[@Quilly_DM]

Or micro debris in orbit punches through it.

[@MDC_1985]

That’s a pretty simplistic view. Do you find your typical response is skepticism? It would seem that data centers on earth would still exist, just that work load could be shared. There is redundancy between land based sites now, why wouldn’t there be redundancy shared between data centers in space and data centers on the ground? Do people tend to believe at a core level that humanity will be gone in 100 years? Why is it that people are willing to say so many things are impossible or never going to work when you consider where we would likely be, what is likely to exist, if technological innovation continues (even without acceleration) at its current pace for another 100, 500 or even thousand years. By that point transport to space and mars on a regular (multiple departures daily) basis is almost certainly likely. Comparatively speaking, data centers in space seems like a no brainer. Why is everyone so critical of the people that take the first steps, steps that must ultimately be taken by someone before the destination can be reached?

[@vincentguttmann2231]

And it raises the issue of repairs, which are a frequent thing in data centers.

[@termitreter6545]

@@MDC_1985 Sometimes the reasons why an idea is stupid are very simple. Good luck radiation hardening your entire data center, make it resistant to micro-meteorites, deal with impossible cooling requirements and never have it require big maintenance or upgrades. Only to switch fast cable connection to much less reliable satellite connection. Your answer doesnt actually touch any of the challenges, youre just romanticising "doing a thing thats not been done before". You should ask first "why has it never done before", its fine to think before doing something groundbreaking.

[@Quasar0406]

​@@MDC_1985 I've got a bridge to sell you

[@luketurner314]

"Backup power supply not required", so if a fleck of paint collides and damages a connection between half the solar array and the data center, does the entire thing go offline?

[@Sableagle]

Oh, come now, Conasgil, if any among us could answer that question for better or worse I would already be drinking in one of the taverns. It's not like a relative velocity of 14 km / s is going to make a fleck ... wait. KILO? Oh, yeah. yeah, that would make a fleck of paint dangerous.

[@myrcutio]

Imagine how high your hard drive failure rate would be in the van allen belt! this would need a lot more redundancy than a terrestrial data center and at least some periodic maintenance to replace/service failed hardware over time.

[@MrAudisek]

A 4km large spacecraft sounds like a massive fishing net for space debris.

[@Marco-qj2tx]

and for creating it (when it inevitably gets hit)

[@venturestar]

🤣🤣🤣🤣🤣😂😂

[@JarrodBaniqued]

Or a massive target for an anti-sat missile

[@hyperforce]

Space would be one of the worst places for a data center... Heat would build up in no time, you can only rely on solar power for electricity, servicing the station would be exhaustively costly. Not to mention, bit flips from cosmic radiation (causing rogue electrons in the microchips of the system to change bit states) would wreck havoc on the system...

[@MoonThuli]

The bit rot would be pretty insane when your data is being blasted by cosmic rays and solar storms

[@col.neville3262]

yeah bit flipping with computers in space is a real problem.

[@angloblaxon]

That's how the ai wakes up and turns into skynet

[@agsystems8220]

AI specifically is robust to noise. It actually sometimes performs better when you randomly change bits as you train it.

[@anonydun82fgoog35]

Don't worry, they'll put a clause in the EULA and people will just click on it and accept it, and then accept regular data corruption as "normal".

[@GeneralERA]

not to mention that rad susceptibility scales with the inverse square of feature size, and the GPUs they'd want to use here have features an order of magnitude smaller than common processors used in space!

[@princecharon]

Random thought after 13:24 - 'Deniable Chinese space pirates have captured the LO data bus! Can your team of implausibly-cool heroes get it back before re-entry?'

[@nicholassaez4405]

Billion dollar idea, put data centers under water then youse the delta T from the top layer of water and the lower layers to power them. Use a modular server system to simplify server mantainence.

[@H3xx1st]

An interesting idea but a lot of people have tried to use the ocean for power generation. It's a lot harder than you'd think, the ocean is quite hostile to metallic objects, and even more so electrically charged ones. It's super expensive to build, and very expensive to maintain and the cost goes up exponentially with size.

[@nicholassaez4405]

@H3xx1st Use small server modules, line them in plastic, etc. It's not going to be perfect, but it should be cheaper and easier to maintain than space.

[@vlachen]

The "Data Bus" reminds me very much of Neal Stephenson's description of Hiro's connection in Snow Crash. "...a 747 cargo freighter packed with telephone books and encyclopedias to power-drive into their unit every couple of minutes, forever."

[@peterfireflylund]

Such a quaint simile. Boeing, phone books, and encyclopedia in a single sentence.

[@totalermist]

@@peterfireflylund it was the 80s...

[@simongeard4824]

Which itself is just a rephrasing of an old joke... "never underestimate the bandwidth of a station wagon full of tapes hurtling down the highway". A line from Andrew Tanenbaum, if I remember correctly.

[@SimonBuchanNz]

Surprised I couldn't find what that is in bits/s with a Google, but with a generous upper bound of 500Mbit/kg for each of those books, and assuming capacity would be roughly weight limited to 274 tonnes, that makes that around about 1Gbit/s. Pretty good if it's wireless.

[@simongeard4824]

@@SimonBuchanNz Well, some figures for you... in 2004, the Auckland residential phone book apparently contained ~350k entries and weighed 1.85kg. So assuming an average of 100 bytes for per entry (for a compact name, address and number), and assuming my calculations are correct, that's about 150Mbit/kg.

[@grrey01]

Forget the station wagon of quarter inch tape, never underestimate the bandwidth of a Cargo Dragon2 loaded with SSDs.

[@eternity_warriors]

8:48 not "right away" but hands down before even opening this video Thank you for that so much! Cheers!

[@thamiordragonheart8682]

The biggest issue for data centers is heat rejection, and if there's anything that gets harder in space, it's heat rejection. and datacenters don't even operate at very high temperatures, which makes rejecting heat through radiation even harder.

[@wally7856]

You would need a heat pump to raise the temperature of the heat you want to radiate away.

[@Andrew-iv5dq]

Seems like locations in Canada (Siberia would have been great until Czar Putin went stupid) or Iceland would be a good place to start. Use cheap hydro or geothermal power and recycle the waste heat for greenhouses.

[@Ammoniummetavanadate]

Just use nuclear

[@RikkiSan1]

@@Ammoniummetavanadate bUt NuClEaR iS sCaRy!

[@Prophes0r]

How are you going to attract the AI Bros with that kind of language/idea? Better tell them you are going to build them on artificial icebergs converted into Megabarges.

[@nighttrain1236]

One data centre near the South Pole and one near the North Pole. That way one data centre always has constant sunlight.

[@foobarf8766]

Yep, already being done here in NZ, heated wave pool run by a new DC, cooling budget follows latitude.

[@DUKE_of_RAMBLE]

Fact checking all this was great work, but the _physical_ Data Bus punny-quip was the best part of this video! Thanks Scott... 🤣♥️

[@eg6841]

Always knew: There is more intelligence in space, than down here on earth. 😁

[@ncdave4life]

Yeah. I just watched the Presidential debate. The Peter principle on steroids, times two (parties).

[@BrightBlueJim]

Considering the selection process for going to space, it's inevitable.

[@Tayken9127]

"I have a history of picking the wrong startups and missing the big bucks payout" You have your own plane Scott, I think you're doing just fine lol.

[@JHe-f9t]

He has a share in a plane and also, planes are like horses and boats, either you're very rich or very poor.

[@ImieNazwiskoOK]

@@JHe-f9t You don't look a gifted plane in the teeth?

[@Steven_Edwards]

As a dead Silicon Valley exile I felt this.

[@CumulusGranitis]

@@JHe-f9t You are thinking sailboats.... hehehehehehe

[@randelbrooks]

wow learned a lot from you. I had not thought of several of the practical objections that you raised especially about heating. Plus the way technology keeps moving forward so quickly I wouldn't put something huge like that in space that becomes obsolete in at least 10 years. When we can create data centers or data networks on something small like the star link satellites then that would be practical essentially disposable hardware or that can be retrieved and recycled. You just watch somebody's got to figure out how to do all this when there's a demand for it. Keep flying safe Scott in your airplane!

[@davidt01]

How will repairs and upgrades be made?

[@Tathanic]

Drones

[@davidt01]

@@Tathanic Drones in space?

[@Tathanic]

@@davidt01 yes?

[@davidt01]

@@Tathanic You mean like a drone with propellers?

[@kaelandin]

@@davidt01 Space drones

[@Tuxlion]

I already knew this would t work for a few reasons. 1: heat is a big problem in space since there's no atmosphere to take it. You need fancy coolers 2: the data transfer wouldn't be good between the two, it'd create a bottleneck 3: the debris already in orbit would destroy the massive solar panel array as there's way too much mass and too much surface area it covers But hey, AI and space are big things so lets mix the two Edit: yeah, everything I pretty much said was said in fuller detail in the video except data wasn't fully discussed.

[@swordcreeper7754]

Add space radiation and the obviously needed maintenance on top and you see why even if it would work perfectly it's an incredibly stupid idea. I'm just surprised people can be convinced to invest in such things. Think about it for 10 seconds and you realize how stupid it is

[@Tuxlion]

@@swordcreeper7754 maybe further away from the sun, but at that rate it'd kill the idea since data transfer would take far longer and if the point is to have servers in space, then you'd want data transfer to atleast be fast.

[@m1k3droid]

Cooling: racks are in pure Nitrogen atmosphere, which becomes the cooling fluid and simultaneously acts as a fire suppressor

[@Richardincancale]

This makes as much sense as putting solar farms in space to take advantage of greater irradiance etc. Exceptions can’t use bog standard panel but need radiation hardened ones, cost $50k a kilo to launch instead of just sticking them on poles in a field, needs crazy electronics in space and on the ground to recover the energy etc. etc. Next!

[@DominikPlaylists]

yes, these people do not even realize a big part of the cost of the panel is driven by how much current it can produce. So panels in space can be the same size but still more expensive because they need to flow more current.

[@MrZnarffy]

The benefit is it is in international space, so you can easily switch jurisdiction just by switching "flag state". No need to move servers physically.

[@joshuaford4460]

By that logic you could just operate in international waters. Not that it would matter though, because no one i can think of would trust a data center that switches its allegience to whoever the highest bidder is at a whim.

[@matthewconnor5483]

​@@joshuaford4460 I actually like the idea of a data center outside the jurisdiction of countries. There's a reason swiss banks are in Switzerland. They refuse to talk to foreign authorities.

[@Keimzelle]

@@joshuaford4460 International waters actually have regulations. Any ship in international waters and not registered in any country (e.g. not being under the jurisdiction of a country) can be entered, pirated and sunk, without any consequences.

[@kendokaaa]

@ozzymandius666 Or they could do the risky stuff with a second entity instead of risking the whole internet archive with good meaning but legally questionable endeavours

[@Kenionatus]

@@matthewconnor5483 That was in the past. Nowadays, Switzerland does quite a bit of data sharing in the name of stopping tax evasion.

[@Prophes0r]

It is a stupid idea on (almost?) every level. Power generation. A current datacenter is approaching (or already exceeding) 3 digit Megawatt needs. The idea of getting that from solar is ludicrous. For reference, the ISS has 4 solar arrays that can generate a combined 120kw with 2500m^sq. So a 10Mw installation, which is likely too small to even be useful, would be 2 orders of magnitude bigger. We are nowhere near being able to make a 2km^sq solar array in space. Heat dissipation. Nearly all of that work becomes heat. It is hard to get rid of that heat in space. There is no realistic/economical way to shed 10MegaWatts of heat, even if it WASN'T in daylight 24/7. Anything that big, remember we are talking multiple HUNDRED square kilometers of solar panels and HUGE (magical) heat radiators, needs to be WAY out in high orbit unless you want to be constantly delivering fuel for station keeping. That makes data links slower and more costly. Makes maintenance impossible. Costs way more money to launch. And out that far you get way more bit-flips because you don't get the benefit of most of Earths magnetic field(or any really). You run chips at lower frequencies, have double/triple redundancy, and you don't put bleeding edge hardware up there. (If I remember right, the new "pizza Box" 1U units that just went to the ISS aren't even fully using PCIe Gen 4.0 and the CPUs are HEAVILY underclocked with boosting disabled.) Space is the OPPOSITE of a good idea. The kinds of computers we put into space would be awful for all that "AI" garbage. However, I'm not surprised that some "AI Bros" are the ones pitching it. AI is already a scam that is just sucking in money from people who have no idea. Why not put the words AI and Space together and get more dumb investors?

[@michaeltorrisi7289]

I almost didn't click on this video because I saw the title and thought "absolutely fucking not". I'm not an astrophysicist, but radiation immediately came to mind, and having to double up your hardware in order to avoid instability means an idea like this is not scalable to the size of a data center. I didn't even think of heat dissipation until I started watching, because I again, I'm a layman, not an expert. Wild that tech bros can fund ideas that a normie could shoot down in under 30 seconds. Such a waste of capital.

[@reahreic7698]

Stick it on the lunar poles. Can use geothermal to address the heat issue. Then you only need to deal with the power, network connectivity, maintenance, logistics, and radiation issues. lol

[@God__Emperor_]

Well said, especially the last part.

[@Prophes0r]

@@michaeltorrisi7289 It's so far from possible that it's a scam. No reasonable engineer would look at this and say "Yeah. We can solve that." It's like bidding on a project to construct an apartment complex on the Moon. We might have an idea how to do it, but at BEST it's a guess. We are decades away from being able to actually DO it, even if every idea we have happens to work.

[@Prophes0r]

@@reahreic7698 You would still need to use radiative cooling. We don't currently have a way to dump that much heat into anything but air/water. And even if we did have a way to dump it into the solid rock of Luna, the rock itself wouldn't conduct it away fast enough.

[@Valendr0s]

Kinda silly... Make basically the same thing but on the ground and you still solve the problem. 16 square km of solar panels in a desert. It's basically just saying "If you are properly integrated, where you produce your own electricity, it's cheaper than getting power from the grid"

[@camojoe83]

Only an idiot would power information infrastructure with something like solar panels..

[@musaran2]

That said, why has no one done it yet?

[@rizizum]

Deserts are one of the worst places to put solar panels, sand will end up covering them, so you'll have to clean them daily, probably using water, which you won't have much in a desert. There's also the extreme heat in the day which decreases efficiency and the extreme cold in the night, which I assume would speed up the degradation process due to constant expanding and contracting

[@camojoe83]

@@rizizum solar panels inside the atmosphere are stupid anywhere.

[@Prophes0r]

Oh boy. Not even close. You are an order of magnitude off. Anything smaller than about 10Megawatts would be useless. Less than 50Megawatts is losing a bunch of efficiency but let's stick with 10. 10Megawatts of solar is about 250km^sq. So, not only does it not make sense in any way to put this stuff in space. It is literally impossible with our current tech. And even if we COULD put that much solar into space, we would have no way to cool it. Basically all of that electricity is becoming heat. And we can't get rid of 10Mw of heat in space. We can't even do 1Mw when it is on the night side of Earth...

[@trissylegs]

What I think might be cool is having so much coolant in your loop you could use at a reaction control system and change the way it’s pumped around the craft to rotate it

[@vodostar9134]

They should put these things at L1

[@kaelandin]

Horrible idea until we get our radiation shielding issues solved.

[@DreamskyDance]

There would probably be more than 1s time lag. Also more radiation. Amd it is even farther away for upgrades or repiars.

[@glenyoung1809]

And what happens to that facility when a good sized CME/solar flare is unleashed and you have a 4km x 4km sized target. Not to mention most of the compute elements aren't going to be hardened against space radiation unless you want to triple their cost.

[@DeuxisWasTaken]

I know that it won't perceptibly dim the sun even in the middle of its shadow, but having a gigantic opaque square right at L1 just sounds wrong.

[@Prophes0r]

@@kaelandin Considering everything else is literal fantasy and several orders of magnitude off anyway...sure why not?

[@radnfo]

AI just jumped all the sharks in the world at the same time!

[@Commentator-jh7wl]

Huh?

[@personzorz]

This is what Tech Bros do.

[@diskdrive123]

You have to assume there is some way to incorporate evaporative cooling in space. Just up the chamber pressure to compensate for the lower gravity and surely there is a coolant that works well in those temperatures and differentials.

[@dXXPacmanXXb]

Startup mindset is the worst. Always coming up with impossible stuff and then saying "nothing is impossible, its just hard" and then they go around collecting money only to fail or completely shift gears 2 years later

[@hermannabt8361]

It wasn’t impossible to build a thinking machine in space. It was impossible to build it anywhere else.

[@juandesalgado]

They may not build it in space, but they'll build it in time... badum tsssh

[@wilppuse]

That sounds familiar, where its from?

[@juandesalgado]

@@wilppuse Before the true answer comes in... my guess would be, one of Asimov's Multivac stories.

[@hermannabt8361]

@@wilppuseI remember it from Bioshock, but I guess they got it from somewhere else. Maybe Atlas Shrugged? I’ve never read it.

[@adriaanwesleystrydom4749]

The Goliath Battery can be a valuable tool for storing the energy. The heat energy can also be used and converted into electrical energy to bolster the data center.

[@klemmonade]

man the numbers they put up are fucking comical, VCs are on actual drugs here.

[@michaelbaker2718]

A quick point of clarification. The processors on the iPhone and other devices custom tailored for AI are to enable query or use of AI models on local devices, they are not capable of training the models.

[@PplsChampion]

2036: zero day issue found -- orbiting supercluster is now world's biggest botnet

[@yodaman8015]

Whoops!

[@yodaman8015]

Better call the space force now

[@MrX-zz2vk]

Speaking of data centers, i learned something interesting today: Three bottles of water are required for every 100 words that Chat GPT-4 generates.

[@mina86]

You've been lied to. Phones with NPUs can generate books worth of data without needing any water. Your computer is capable of generating images (which is harder than text) without need for any water.

[@normduch]

What happens when you need to turn it off and on again? And, how do you check if it's plugged in? I jest, but physical maintenance is a problem.

[@DannyTube69]

1. cooling in space is difficult 2. space has a lot more cosmic rays which are not good for computers 3. their maths doesn't work out that well

[@Prophes0r]

Oh boy. Their math is way WAY worse than that. Anything under about 10Mw isn't worth using. A 10Mw array is going to be ~250km^sq + another 50km^sq to power the 300km^sq of radiators. And that's going to need to be in high polar sun-synchronous orbit, where we don't currently have a way to build or maintain it. The whole idea is pure fantasy to scam money from AI Bros. It is orders of magnitude beyond feasible.

[@paulbrooks4395]

The frustrating part of this "lack of power" line is how states like California have more solar than they can use and even sell it to Arizona (which has an earlier sunset) during peak demand. People talk about power usage but nobody talks about...turning them off at night/prior to peak demand hours. This is the hallmark of consumptive capitalism, if it isn't happening 24/7 generating profit then it's not worth it. For example, Musk is building datacenters that run off diesel generators. We are already in a climate crisis but all that can be damned when there's money to be made. I wish we could use space-based systems, but sadly it will never happen.

[@CalvinCongreve]

Promises are the uniquely human way of ordering the future, making it predictable and reliable to the extent that this is humanly possible.

[@Trafulgoth]

Before watching the video, here's my instinct on the subject: No way this is cheaper to operate. Some reasons why it's not totally insane: 1. No taxes. You presumably don't need to pay property taxes on space assets, since it isn't land or a building. 2. No security personnel. You don't need to worry about thieves breaking into your space asset. 3. No insurance. You don't need to pay for property, security, fire, or disaster insurance in space. 4. No cleaning personnel. You don't need to clean the inside of a space asset that has no people onboard. 5. No grid payments. You don't need to (and can't) pay to access electrical or water services. Some reasons why it is totally insane: 1. Maintenance. The server is down? Yeah, I'll send a guy to check on it in 3 weeks. 2. Electricity. This isn't a data center, it's a power plant with some servers attached. 3. Space. It's in space. The most expensive place for things to get to. 4. Cooling. Cooling data centers and space objects are different kinds of difficult, now you get to do both at the same time! 5. Security. Someone is using your servers for very illegal things? Good thing no one can pull the plug or perform forensics!

[@Tjalve70]

My suggestion: Make this data center in a cold climate, near a city that needs central heating. Use the heat from the data center to heat up homes. That way you get benefit both from the data center, and the heat that it produces. If you manage to turn a waste product into a useable product, THAT's when you get the big profits.

[@H3xx1st]

This is the same as mining bitcoin to heat your house in the winter. Could actually be a more cost effective way to turn electricity into heat.

[@stylinsandwich]

​@@H3xx1stWho needs a heater when you have 10 4090s

[@alterego3734]

13:26 per Jevons Paradox, more efficient AI would actually increase resource usage.

[@CliseruGabriel]

there is one aspect that can be added. You did mention the cooling fluid. That cooling fluid has to be transported into space as well. So it will require a lunch in on itself

[@Ramschat]

Cooling in the vacuüm of space is way more difficult. There's no cooling water. Only radiator panels, and data centres produce a *ton* of heat.

[@nathanr136]

We can barely keep data centers right off shore working in a cost effective manner Can’t imagine the cost of repairs for a data center in orbit even if prices fall to their projected levels

[@TraciPearson-ok2tr]

Thank you for this thorough and well-researched discussion of this idea! When I first heard of it, it seemed fairly sensible to me-the idea of putting data centers into orbit. The more I thought about it, the more I wondered about the various aspects of it and whether it would actually be wise or feasible. You've taken this far beyond my own shallow pondering, and you've done a really good job!