Starlink has been so successful, it is facing a lot of competition in the next few years. Every major power wants their own, national starlink network.
China's state-backed starlink competitor GuoWang is putting 13,000 satellites in orbit by 2030. They've already started launching satellites.
China's Qianfan plans 15,000 satellites by 2030.
AST SpaceMobile is building their own network.
Amazon Leo plans for 3,000 satellites in orbit, and is already launching satellites.
The EU is building IRIS², explicitly as a Starlink alternative.
Russia, after realizing how critical starlink is on the battlefield, is building its own Rassvet network. They've already launched satellites.
I'm a layman in this field, and I always thought that satellites were supposed to take a very small space in orbit... until I saw a GIF made from Artemis 2 ship. Satellites were big enough that could see the queue of them orbiting Earth in the GIF. Maybe they were the bigger ones and I'm confusing them with Starlink?
The people who want to put data centers in orbit must be either much smarter than me, or much dumber than me, because I just don't get how that makes any technical or economic sense.
Of course, it's possible nobody actually wants to do this, they just want to get funded to do it. (Old joke: "I wish I had enough money to buy an elephant...")
There's economic case (or rather economics is not factor) for space force edge computing, but who knows how big demand that is. But it is 100% going to happen, and market can be fooled into conflating military economics with civilian economics.
The list of people saying they're going to put compute in space is getting long. Unlikely they're all dumb or faking it. It is clearly technically feasible but how the economics will develop is still unknown.
Starcloud
Axiom Space
SpaceX (Starlink orbital data center constellation)
Google (Project Suncatcher)
Lonestar Data Holdings
Blue Origin (TeraWave/Project Sunrise)
Aetherflux (Galactic Brain)
Kepler Communications
Sophia Space
Madari Space
Adaspace / ADA Space / Guoxing Aerospace (China)
Orbital (LA-based AI inference constellation)
Atomic-6
Planet Labs (with Google)
Crusoe (partnership with Starcloud)
Edge Aerospace
Rotonium
Lux Aeterna
Star Catcher
Loft Orbital
Cowboy Space
OrbitsEdge
Galaxia (Canada)
China Aerospace Science and Technology Corporation (CASC)
Beijing Orbital Twilight / Orbital Chenguang
Astro-Future Institute
Space Compass (NTT + SKY Perfect JSAT JV, with JAXA support)
NASA + HPE (Spaceborne Computer)
European Union / ESA / ASCEND (Thales Alenia Space, Airbus, ArianeGroup, HPE, Orange)
UNOOSA (with Madari Space)
Canada’s Department of National Defence (with Galaxia)
I don't buy the opposition to orbital datacenters. I think that's just the perfect being the enemy of the good, which (sadly) is very common in environmental activism.
Let's assume two premises:
1. Demand for AI compute will continue to grow for the next 10 years.
2. The cost of orbital datacenters will approach the cost of terrestrial datacenters (in $ per token terms).
If you don't buy either of these two premises (and I agree that neither is guaranteed) then you don't have to worry. No one is going to waste money on orbital datacenters if they aren't profitable.
But if you buy the two premises above, then what matters is whether orbital or terrestrial datacenters are better for the environment. And it turns out that, given the current energy generation mix, the CO2 emissions from terrestrial datacenters far outstrip the CO2 from launch.
A 100MW datacenter will emit 1.5 million tons of CO2 over a 5 year lifetime (given average USA energy sources). In contrast, 10 Starship launches (~1,000 tons to orbit) will emit no more than 40,000 tons of CO2. Almost all other environmental effects will be proportional.
So, if you care about the environment, and you believe/fear that AI will demand a lot of compute, then you should hope that orbital datacenter work out. If you really care, you might even help to develop them.
There is already some trash flying around in space from decommissioned satellites. Reentering the atmosphere needs some energy to initiate the process and that was/is not always planned for. The combustion during reentry is producing CO2. This CO2 emission will scale up when the number of satellites is scaled up.
The alternative of collecting debris in space is discussed by some space agencies. I really don't think it is a good behavior from SpaceX to put more trash in space and let public money take care of the cleanup later.
> A 100MW datacenter will emit 1.5 million tons of CO2 over a 5 year lifetime (given average USA energy sources). In contrast, 10 Starship launches (~1,000 tons to orbit) will emit no more than 40,000 tons of CO2.
How did you come to 10 launches / 1000 tons as what it takes to launch a 100MW data centre in space?
> 1. Demand for AI compute will continue to grow for the next 10 years. 2. The cost of orbital datacenters will approach the cost of terrestrial datacenters (in $ per token terms).
> If you don't buy either of these two premises (and I agree that neither is guaranteed) then you don't have to worry. No one is going to waste money on orbital datacenters if they aren't profitable.
I'm glad you said the final quoted paragraph; while it has always been difficult to tell which websites are accurate vs. slop, and moreso today with fully automated slop, I see claims the current rate is doubling between 15 and 3 months. Even at the slowest of these, 15 months, this gets 2^(120/15) = 256x growth in 10 years, which would raise it significantly above current total global electrical demand: https://www.iea.org/reports/energy-and-ai/energy-demand-from...
> A 100MW datacenter will emit 1.5 million tons of CO2 over a 5 year lifetime (given average USA energy sources). In contrast, 10 Starship launches (~1,000 tons to orbit) will emit no more than 40,000 tons of CO2. Almost all other environmental effects will be proportional.
Not so: given that existing power capacity is already a constraining factor for new ground-based data centres, the alternative to be considered is not the CO2 emissions of existing average USA energy sources, but the CO2 emissions of new energy sources. This may be many things, but given that renewables are now the cheapest new energy, if politics stops renewables in the US it just means the data centres (especially at this kind of scale) aren't going to be in the US, but it doesn't say that they'll be in space specifically.
>>I don't buy the opposition to orbital datacenters. I think that's just the perfect being the enemy of the good, which (sadly) is very common in environmental activism.
The opposition is from people who have a brain and like to use it. And dont engage on Musk next cheap trick to sustain his chimerical stock valuations. A 100 MW datacenter in vacuum takes from you "put a man on the Moon" to "lets host Kubernetes on the Sun" :-))
Just to start, a 100 MW datacenter is a 100 MW toaster. On Earth, the waste heat would go into into air or water. In orbit, it can only leave by radiation. So you need a massive radiator area, plus pumps, coolant loops, structure, pointing, redundancy, and micrometeorite protection...
Then you need radiation hardened GPUs, DRAM, SSDs, optics, and power electronics and those are not going to magically become space rated because the pitch deck has a Mars picture on it. Also on earth datacenters these parts constantly fail. There was even some recent study of about 20% of GPUs failing within 3 years, and humans constantly have to replace them. In orbit, you would have space mission to swap the bad PSU...
If you make the calculations, at normal electronics temperatures, rejecting the amount of heat needs a radiator field on the order of 500 m x 500 m under ideal assumptions... Then you neede power, and that has the same scale problem. Meaning you need an enormous solar farm, but wait...what about the scenario of a eclipse? For 30 minutes without sunlight you need about 50 MWh of storage.
Musk is a moron, and this video of him explaining it as simply radiative is proof number 256 you can be an idiot, and a billionaire. And shame on Jensen Huang for not calling him out: https://youtu.be/trgn7s5-YHc?t=140
Who is paying for all this? Starlink revenue? I don't think so. To 100x what they've done over the last ten years and also fund it with money they're not being paid by others is implausible. They've got money, but they don't have THAT much money. They'd need to significantly juice up the whole rest of their business to not just be substantially more profitable, but also launch far more payloads. Does that much demand (for SpaceX) to get stuff into space even exist?
So there must be a big story for the IPO….Remember that Tesla is still not selling actual/real self driving cars. I would say stock wise it would worth it to gable b/c Tesla did so well but SpaceX is being IPO-ed as it would have already have datacenters in space and ride hailing to Mars
32 comments
[ 5.8 ms ] story [ 55.1 ms ] threadReally? I wonder how they are going to get them up there without rocket launches?
SpaceX wants investors to think that they will be able to launch millions of satellites.
China's state-backed starlink competitor GuoWang is putting 13,000 satellites in orbit by 2030. They've already started launching satellites.
China's Qianfan plans 15,000 satellites by 2030.
AST SpaceMobile is building their own network.
Amazon Leo plans for 3,000 satellites in orbit, and is already launching satellites.
The EU is building IRIS², explicitly as a Starlink alternative.
Russia, after realizing how critical starlink is on the battlefield, is building its own Rassvet network. They've already launched satellites.
The Artemis gif you are referring to only shows the satellites because the earth is eclipsing the the sun and they did a long exposure.
Those satellites showing up in the gif are briefly catching the sun at the perfect angle.
Spec Priority: ability to attach said laser defense instrument to home telescope ... and enable user to blast those madafakkas out of the sky.
This article seems to confuse Starlink with ordinary cellular communications
https://en.wikipedia.org/wiki/Kessler_syndrome
Sorry Buck Rogers fan bois, should have left this fantasy in the 1950s...
Of course, it's possible nobody actually wants to do this, they just want to get funded to do it. (Old joke: "I wish I had enough money to buy an elephant...")
Some previous discussion:
A million new SpaceX satellites will destroy the night sky
https://news.ycombinator.com/item?id=47598415
Part of this announcement:
xAI joins SpaceX
https://news.ycombinator.com/item?id=46862170
Let's assume two premises:
1. Demand for AI compute will continue to grow for the next 10 years. 2. The cost of orbital datacenters will approach the cost of terrestrial datacenters (in $ per token terms).
If you don't buy either of these two premises (and I agree that neither is guaranteed) then you don't have to worry. No one is going to waste money on orbital datacenters if they aren't profitable.
But if you buy the two premises above, then what matters is whether orbital or terrestrial datacenters are better for the environment. And it turns out that, given the current energy generation mix, the CO2 emissions from terrestrial datacenters far outstrip the CO2 from launch.
A 100MW datacenter will emit 1.5 million tons of CO2 over a 5 year lifetime (given average USA energy sources). In contrast, 10 Starship launches (~1,000 tons to orbit) will emit no more than 40,000 tons of CO2. Almost all other environmental effects will be proportional.
So, if you care about the environment, and you believe/fear that AI will demand a lot of compute, then you should hope that orbital datacenter work out. If you really care, you might even help to develop them.
The alternative of collecting debris in space is discussed by some space agencies. I really don't think it is a good behavior from SpaceX to put more trash in space and let public money take care of the cleanup later.
That is one wild, batshit insane assumption. So much so that I did not bother to read the rest of your post.
How did you come to 10 launches / 1000 tons as what it takes to launch a 100MW data centre in space?
> 1. Demand for AI compute will continue to grow for the next 10 years. 2. The cost of orbital datacenters will approach the cost of terrestrial datacenters (in $ per token terms).
> If you don't buy either of these two premises (and I agree that neither is guaranteed) then you don't have to worry. No one is going to waste money on orbital datacenters if they aren't profitable.
I'm glad you said the final quoted paragraph; while it has always been difficult to tell which websites are accurate vs. slop, and moreso today with fully automated slop, I see claims the current rate is doubling between 15 and 3 months. Even at the slowest of these, 15 months, this gets 2^(120/15) = 256x growth in 10 years, which would raise it significantly above current total global electrical demand: https://www.iea.org/reports/energy-and-ai/energy-demand-from...
> A 100MW datacenter will emit 1.5 million tons of CO2 over a 5 year lifetime (given average USA energy sources). In contrast, 10 Starship launches (~1,000 tons to orbit) will emit no more than 40,000 tons of CO2. Almost all other environmental effects will be proportional.
Not so: given that existing power capacity is already a constraining factor for new ground-based data centres, the alternative to be considered is not the CO2 emissions of existing average USA energy sources, but the CO2 emissions of new energy sources. This may be many things, but given that renewables are now the cheapest new energy, if politics stops renewables in the US it just means the data centres (especially at this kind of scale) aren't going to be in the US, but it doesn't say that they'll be in space specifically.
The opposition is from people who have a brain and like to use it. And dont engage on Musk next cheap trick to sustain his chimerical stock valuations. A 100 MW datacenter in vacuum takes from you "put a man on the Moon" to "lets host Kubernetes on the Sun" :-))
Just to start, a 100 MW datacenter is a 100 MW toaster. On Earth, the waste heat would go into into air or water. In orbit, it can only leave by radiation. So you need a massive radiator area, plus pumps, coolant loops, structure, pointing, redundancy, and micrometeorite protection...
Then you need radiation hardened GPUs, DRAM, SSDs, optics, and power electronics and those are not going to magically become space rated because the pitch deck has a Mars picture on it. Also on earth datacenters these parts constantly fail. There was even some recent study of about 20% of GPUs failing within 3 years, and humans constantly have to replace them. In orbit, you would have space mission to swap the bad PSU...
If you make the calculations, at normal electronics temperatures, rejecting the amount of heat needs a radiator field on the order of 500 m x 500 m under ideal assumptions... Then you neede power, and that has the same scale problem. Meaning you need an enormous solar farm, but wait...what about the scenario of a eclipse? For 30 minutes without sunlight you need about 50 MWh of storage.
Musk is a moron, and this video of him explaining it as simply radiative is proof number 256 you can be an idiot, and a billionaire. And shame on Jensen Huang for not calling him out: https://youtu.be/trgn7s5-YHc?t=140