Jeff Bezos Predicts Gigawatt Data Centers in Space Within Two Decades (reuters.com)
- Reference: 0179642248
- News link: https://science.slashdot.org/story/25/10/03/1426244/jeff-bezos-predicts-gigawatt-data-centers-in-space-within-two-decades
- Source link: https://www.reuters.com/business/energy/data-centres-space-jeff-bezos-thinks-its-possible-2025-10-03/
Bezos was speaking in a fireside chat with Ferrari and Stellantis Chairman John Elkann. He said the giant training clusters needed for AI would be better built in space because there are no clouds, rain or weather to interrupt power generation. Bezos predicted that space-based data centers would beat the cost of Earth-based ones within a couple of decades. He described the shift as part of a broader pattern that has already occurred with weather satellites and communication satellites. The next steps would be data centers and then other kinds of manufacturing.
[1] https://www.reuters.com/business/energy/data-centres-space-jeff-bezos-thinks-its-possible-2025-10-03/
Cooling (Score:5, Insightful)
How would they handle the enormous cooling requirements? Can they use some sort of passive heat radiation? Do cooling fins work in space?
Re:Cooling (Score:5, Interesting)
Yes and yes.
[1]https://hackaday.com/2025/06/1... [hackaday.com]
"This is where skepticism creeps in. After all, cooling is the greatest challenge with high performance computing hardware here on earth, and heat rejection is the great constraint of space operations. The âoeicy blackness of spaceâ you see in popular culture is as realistic as warp drive; space is a thermos, and shedding heat is no trivial issue. It is also, from an engineering perspective, not a complex issue. Weâ(TM)ve been cooling spacecraft and satellites using radiators to shed heat via infrared emission for decades now. Itâ(TM)s pretty easy to calculate that if you have X watts of heat to reject at Y degrees, you will need a radiator of area Z. The Stephan-Boltzmann Law isnâ(TM)t exactly rocket science."
Devil will be in the details for how much cooling capacity is needed to reject the heat generated by the GPUs, how long the satellites are designed to last, stationkeeping, etc.
The current trend is toward smaller, disposable satellites, so I don't know if what Bezos is envisioning is a massive distributed cluster of smaller satellites, or a return to larger satellites that are docked to a compute pod payload...
I mean, this could be one possible proposal for keeping the ISS - jettison most everything but the solar panels and radiators, and add station keeping and a compute payload module. Keep the cupola, canada arm, docking capability for tourist and maintenance visits.
[1] https://hackaday.com/2025/06/19/space-based-datacenters-take-the-cloud-into-orbit/
Re: (Score:2)
As a point of reference, the current ISS cooling system can reject 75kW of heat, from a combined area of 84.864m^2. Those radiators are only part of the system, the External Active Thermal Control System, and they are continuously rotated to optimise heat loss from them.
Ref: [1]https://en.wikipedia.org/wiki/... [wikipedia.org] and the cited [2]https://www.nasa.gov/pdf/47348... [nasa.gov]
So that ballparks 2GW requiring ~2250m^2 of radiators. Remember, it needs mechanisms for rotating, and to be placed so that they're mostly not radia
[1] https://en.wikipedia.org/wiki/External_Active_Thermal_Control_System
[2] https://www.nasa.gov/pdf/473486main_iss_atcs_overview.pdf
Re: (Score:3)
In space, radiators work by expelling infrared light.
Re:Cooling (Score:5, Informative)
> In space, radiators work by expelling infrared light.
Infrared light emissions depend on temperature to the 4th power so they barely work at all for lower temperatures and only start being efficient at high temperatures. This means your effective size needs to be huge unless you’re OK with temperatures of several hundred degrees. Putting heavy things in space is extremely expensive, and good luck servicing the system if anything goes wrong. Most satellites stay cool by using many layers of reflective/insulating material and not generating much internal heat at all, unlike a data center with hundreds of thousands of watts. The idea all these extra expenses are more than a battery and 2-4x the solar panels makes no sense at all.
Re: (Score:2)
What's wrong with running a secondary coolant loop at hundreds of degrees? Assuming we have all this energy to play with (run a compressor).
The economics, on the other hand, yeah you wouldn't think so. But you would think Bezos would pay somebody to run some numbers before backing the idea in public.
In the old days I would have said the DoD (now hosted at [1]http://war.gov/ [war.gov] - impressive!?) would surely pioneer this. But these days big tech is so much richer than even the military-industrial complex.
[1] http://war.gov/
Amazon is a harsh mistress. (Score:2)
How do you get rid of gigawatts of waste heat every day?
Heat up rocks and drop them?
Space-based deathrays.
Re: Cooling (Score:2)
They can't. Oh technically you could build black body radiators to do it, but they would need to be kilometers in size, maybe many kilometers, though to be fair so would your solar panel arrays so it might not be a total showstopper. There are so many issues with the idea of data centers in space I'm not even sure cooling is the biggest. The idea only makes sense if you simulatenously own a space launch company and don't have the first clue about physics or engineering. So, you know, Bezos. I think even Mus
Re: (Score:1)
> I think even Musk knows enough engineering to not propose an idea this monumentally stupid.
There's videos scattered about the web where Musk answers questions about solar powered this and that in space and he's clearly annoyed about it. He's done the math, likely had many people under his employ triple check the math, and there's no real value in going to orbit to harness solar power for anything.
Certainly there's solar PV panels on all sorts of satellites to get power but this is a matter of getting power to systems for the purpose of having "high ground" for the purpose of relaying communicati
Re: (Score:2)
> How would they handle the enormous cooling requirements? Can they use some sort of passive heat radiation? Do cooling fins work in space?
Dissipating through thermal radiation would basically be the only option but the radiating surface area required to keep temperatures reasonably low would need to be enormous.
TL;DR: Unrealistic.
and what exactly (Score:2)
is the plan for data transfer? I'm fairly certain that transfer rates over physical cabling trounces wireless transfer rates. Does he plan to run really long cables up to those datacenters? Or does he have a brand-new tech for wireless data transfer that matches wired rates?
We all know he's not doing this (Score:2, Troll)
It's pointless to list out the reasons why.
The purpose of this is to get a little bit of positive attention directed his way that will appeal to a certain class of voter and consumer.
Right now every single mega corporation is basically taking all the water and electricity for their data centers and leaving communities with nothing.
This is him telling those communities that they don't need to worry about losing their water and electricity because the data center will be in space soon anyway.
I
Re: (Score:2)
The data transfer part is basically just starlink.
You aren't going to remotely host your gaming PC this way. But for deep neural nets, which have a massive computation-to-IO ratio, that's a different ballgame. E.g. for an LLM the input might be a prompt of a couple dozen bytes, and the output might be a paragraph of a couple hundred bytes. But a lot of computation is needed to get from the input to the output.
(The AI would also need an onboard mirror of any part of the internet it can search for answ
What a crock⦠(Score:1)
Jeff sounds like he's trying to kick off another round of funding for his c-tier rocket company. Weâ(TM)ll have data centers with dedicated SMRs on site long before we have data centers in space. Hell, we might have fusion powered data centers before it's cost effective to built them in space.
Re: (Score:2)
His rocket company is about as old as SpaceX and has accomplished nothing of value.
Re: What a crock⦠(Score:2)
Virgin Galactic is even worse. 100s of millions spent on a crap spaceplane that doesnt even get to 60 miles up and hasnt gone anywhere in the last 15 months.
Bullshit (Score:5, Insightful)
This is the same type of investor hype bullshit Musk has been peddling about sending people to Mars.
Throw out some wildly huge numbers with minuscule small timelines.
Why? Because they both own space agencies. That's it. That's all it is.
They're both already doing really cool and innovative stuff with their space agencies! They don't NEED to be blowing smoke out their asses, but they're constantly in this pissing contest to see who can come up with the most "believable" and yet wildly outlandish idea of their capabilities.
Re:Bullshit (Score:5, Interesting)
Replying to myself since I was curious and had to look it up:
The International Space Station has about 100 kilowatts power generation capability.
The article says "datacenters" - plural, but for now, let's just assume a single gigawatt datacenter in space in 20 years.
That's about 7300 days.
A single gigawatt datacenter would consume about 10,000x the power generation capabilities of the ISS.
Starting today, we'd need to launch more than 1 ISS PER DAY to reach this goal in 20 years, for a single gigawatt datacenter.
I think this brings the absurdity of the situation into a little more crystal clear of a picture.
Re: (Score:2)
If you can get lunar resources and build stuff in space...it might be possible. But I still doubt it. Hmmm...calling Robert Heinlein... I wouldn't mind an O'Neil space colony...and orbiting solar power stations beaming power back to earth by microwave.
Re: (Score:2)
> I think this brings the absurdity of the situation into a little more crystal clear of a picture.
It really doesn't do anything to illuminate the situation. It would if the ISS were built to generate/consume a lot of power, but it's not. Quite the opposite. It's designed to provide a reasonable amount of power to run life support and energy to run some experiments.
It may, of course, be true that orbital gigawatt data centers are a lot more than twenty years away, but the comparison with ISS doesn't tell us anything.
I thought this was a dupe (Score:2)
But it's actually two dopes: [1]https://science.slashdot.org/s... [slashdot.org]
[1] https://science.slashdot.org/story/25/05/02/1839250/eric-schmidt-apparently-bought-relativity-space-to-put-data-centers-in-orbit
This is the Elon Musk playbook (Score:2)
Throw out a completely nonsensical prediction and then let the media cover it and make it sound like you're some sort of super genius. This works because people are very very very dumb and they do not understand that you would never build anything in space you didn't have to because it costs a hell of a lot of money to put stuff in space.
It also tells communities that are having all that water and electricity taken by data centers not to worry because they're going to be in space soon. It's similar to t
Waste heat disposal? (Score:2)
Large computer centers in space seems like a really bad idea, unless you plan on doing something like capturing an asteroid to use as a heat sink. The moon is much more plausible..but you'd still need to connect down to bedrock.
Thats a hell of an IT bill (Score:5, Interesting)
Spoken like a man who has no clue just how much hands on maintenance goes into a datacenter. Raise your hand if youâ(TM)ve ever had to driven across state lines to push a power button?
and millions to ship new parts to them (install fe (Score:2)
and millions to ship new parts to them (install fees just as high)
Suuure (Score:2)
And flying cars, personal humanoid robot assistants and some other crap.
Skynet (Score:1)
I for one will be welcoming our Skynet overlords and will be looking forward to their administration.
Clownshow hype (Score:2)
Aside from direct-to-TV movie-plot scenarios, there is no market for this.
If you disagree, please describe the workloads which require being in space and have an ROI justifying both deployment and ongoing maintenance costs for this stupidity.
Damned physics. (Score:2)
I think he's ignoring the cost per kilogram to get it there. I suspect that heat management strictly by radiation won't be fast enough, so you'll still need to crank up surface area to cool the chips. So you can't even save mass there.
Of course my non-physicist brain might have that wrong..
Why bother building when you can use magic? (Score:1)
I mean, if you tried to build something in space, you'd have to contend with
1. Launch costs
2. Cooling
3. Radiation tolerance
4. Comms access
5. Launch costs for upgrades and replacements
Way easier to wave a magic wand. That's where the real money is going to be. Mark my words.
Space and power (Score:2, Insightful)
> space offers uninterrupted solar power around the clock
So does planet Earth, if you have a worldwide grid powered by solar... :)
Re: (Score:2)
replying to cancel unintentional down-mod
So a Blackwell cabinet is 1360 kgs... (Score:2)
At $1600/kg to get into low-earth-orbit we're talking $2.2m today to get than into space before any infrastructure to support it. When did Jeff project he puts his first space elevator into operation to actually get all this gear up there?
More space junk (Score:2)
"Reaching for the stars, we blind the skies..." Ronnie James Dio
You laugh, but... (Score:2)
...a former colleague of mine is now working with a startup that's all about orbiting compute. Their business model is basically to let other satellite builders / operators offload their computing needs to dedicated satellites. The customers get lighter satellites (less expensive to launch) and it's more efficient from a computing perspective because a lot of satellites are running their existing compute with a relatively low duty-cycle, whereas the dedicated-compute satellites will aggregate that bursty de
Not this Bozo again (Score:2)
This clown can barely get a rocket into space in the span of 20 years. Gigawatt data centers? I don't think so.
Mirrors instead (Score:2)
More practical to launch mirrors to beam sunlight to earth. Mirrors are lighter, and they don't bit-flip by cosmic radiation.
Get those stupid AI data centers on the moon (Score:1)
So we don't have to deal with them, and we can send Sam Altman to look after them.