4 ms·

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 ac

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.

amanaplanacanal 6mo ago
I don't see how orbital datacenters will ever be able to compete with terrestrial. And assuming that the current mix of electric generation isn't going to become cleaner over time is an interesting take.
- GMoromisato 6mo ago
  Current mix of electric generation is not going to get 37 times cleaner in 10 years. That's the ratio we're talking about right now. But I do agree that the economic math is not a slam dunk. A lot of things have to go perfectly right for it to make sense economically. But that's the irony: Orbital datacenters are much cleaner than terrestrial ones, and if we don't deploy them, it will only be because corporations care more about profit than the environment. You would think environmentalists would be the ones pushing for orbital datacenters.
  - ben_w 6mo ago
    > Current mix of electric generation is not going to get 37 times cleaner in 10 years. That's the ratio we're talking about right now. Doesn't need to: because data centres are already power-constrained, the correct question is marginal new supply in each case. If you build 25 GW of new data centres (which seems to be roughly the scale being proposed here), the options are (1) build more power plants or (2) have brownouts and/or rolling blackouts. What are those new power plants going to be? Even though Trump hates renewables, they're now the cheapest new power source, which was already necessary to be even considering putting 25 GW of PV onto satellites in the first place. This makes it a question of what's cheaper: over-provisioning and storage, or launch costs?
- x-complexity 6mo ago
  > I don't see how orbital datacenters will ever be able to compete with terrestrial. An orbital DC makes fiscal sense when the cost to launch one to space is lower than the cost to build one on Earth. Key point: The cost of building on Earth will inevitably trend upwards as more restrictions & costs are (~80% rightfully) placed onto Earth-bound datacenters.
  - dingaling 6mo ago
    More restrictions and costs are going to have to be placed on satellites, too. At present it's like railroad building across the Wild West; get some notional national 'permission' and then chuck them up there into a globally-shared space. That's not sustainable as the important orbits become crowded.
  - MagicMoonlight 6mo ago
    And that will never be true, so it’s meaningless. It would also be cheaper to build one on the moon, if it was free to build them on the moon.
    - x-complexity 5mo ago
      > And that will never be true, so it’s meaningless. The math is simple: The total cost for building on Earth is X, the total cost for building in space is Y. If X > Y, building in space makes fiscal sense. To say that it never happens, MagicMoonlight, requires X to always be less than Y, which cannot hold true forever. Eventually, X will grow to be more than Y, simply due to (rightfully placed) increased regulations for building on Earth. > It would also be cheaper to build one on the moon, if it was free to build them on the moon. Arguably, that should be the original goal: Build the DC on the moon.
whatisthiseven 6mo ago
The first premise is reasonable. The second premise is so outlandish as to require dozens of other assumptions, optimistic outcomes for launch, and pessimistic views on the cost of earthbound systems. I care about the environment and I think we can keep earthbound systems, and also reduce their impact. Making assumptions about the feasibility of launch and the economic absurdity of orbital compute, but not affording the same assumptions for what could be done for earthbound systems, is confusing? And no, orbital compute is absolutely not far lower than earthbound in co2 cost. Because it doesn't exist at any scale. All orbital compute is solely dedicated to switching where it is best served. If you were to spitball numbers, are we even willing to assume orbital matches earthbound in compute total, dollar cost, uptime, or any beneficial metric? The only metric I see is just slinging silicon into space.
- GMoromisato 6mo ago
  I get skepticism that orbital datacenters will make sense economically. It would not surprise me if they never happen. But if they made economic sense, then they would be much more environmentally friendly. On CO2 emissions, we're talking about a factor of 37. That's not a small amount. Even a factor of 10 would be a major issue multiplied by terawatts of power. Sure, maybe Earth-bound energy gets cleaner, but it's not going to get 10 times cleaner before 2036.
  - jurgenburgen 6mo ago
    > Even a factor of 10 would be a major issue multiplied by terawatts of power. Sure, maybe Earth-bound energy gets cleaner, but it's not going to get 10 times cleaner before 2036. Currently we’re not on that trajectory but if we don’t get there we might be too busy with civilizational collapse to need space data centers. It would require doubling the rate of 2025 renewable deployments on average for the next 11 years so technically and economically it’s feasible.
  - beAbU 6mo ago
    What aspect of an orbital data centre makes it more environmentally friendly than a terrestrial one?
hommelix 6mo ago
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.
- GMoromisato 6mo ago
  Satellites below 500 km do not need energy to reenter. Their orbits will naturally decay from friction with the upper atmosphere. CO2 emissions from reentering satellites is far lower than CO2 emissions from terrestrial datacenters. If you need 1,000 tons of satellite for a 100 MW of compute, you're not going to get more than 1,000 tons of CO2. In fact, you'll get much less since metal doesn't release CO2 when it melts. 1,000 tons of CO2 is negligible compared to 40,000 tons at launch and invisible compared to 1.5 million tons for a terrestrial datacenter.
rasz 6mo ago
orbital datacenters are the new vertical farming
beAbU 6mo ago
> 2. The cost of orbital datacenters will approach the cost of terrestrial datacenters (in $ per token terms) That is one wild, batshit insane assumption. So much so that I did not bother to read the rest of your post.
verzali 6mo ago
https://open.substack.com/pub/thequantumcat/p/data-centres-in-orbit-are-just-as https://open.substack.com/pub/thequantumcat/p/data-centres-i...
wasabi991011 6mo ago
> 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?
ben_w 6mo ago
> 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. 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-ai 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.
tcp_handshaker 6mo ago
>>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 https://youtu.be/trgn7s5-YHc?t=140

amanaplanacanal 6mo ago

I don't see how orbital datacenters will ever be able to compete with terrestrial. And assuming that the current mix of electric generation isn't going to become cleaner over time is an interesting take.

GMoromisato 6mo ago
Current mix of electric generation is not going to get 37 times cleaner in 10 years. That's the ratio we're talking about right now. But I do agree that the economic math is not a slam dunk. A lot of things have to go perfectly right for it to make sense economically. But that's the irony: Orbital datacenters are much cleaner than terrestrial ones, and if we don't deploy them, it will only be because corporations care more about profit than the environment. You would think environmentalists would be the ones pushing for orbital datacenters.
- ben_w 6mo ago
  > Current mix of electric generation is not going to get 37 times cleaner in 10 years. That's the ratio we're talking about right now. Doesn't need to: because data centres are already power-constrained, the correct question is marginal new supply in each case. If you build 25 GW of new data centres (which seems to be roughly the scale being proposed here), the options are (1) build more power plants or (2) have brownouts and/or rolling blackouts. What are those new power plants going to be? Even though Trump hates renewables, they're now the cheapest new power source, which was already necessary to be even considering putting 25 GW of PV onto satellites in the first place. This makes it a question of what's cheaper: over-provisioning and storage, or launch costs?
x-complexity 6mo ago
> I don't see how orbital datacenters will ever be able to compete with terrestrial. An orbital DC makes fiscal sense when the cost to launch one to space is lower than the cost to build one on Earth. Key point: The cost of building on Earth will inevitably trend upwards as more restrictions & costs are (~80% rightfully) placed onto Earth-bound datacenters.
- dingaling 6mo ago
  More restrictions and costs are going to have to be placed on satellites, too. At present it's like railroad building across the Wild West; get some notional national 'permission' and then chuck them up there into a globally-shared space. That's not sustainable as the important orbits become crowded.
- MagicMoonlight 6mo ago
  And that will never be true, so it’s meaningless. It would also be cheaper to build one on the moon, if it was free to build them on the moon.
  - x-complexity 5mo ago
    > And that will never be true, so it’s meaningless. The math is simple: The total cost for building on Earth is X, the total cost for building in space is Y. If X > Y, building in space makes fiscal sense. To say that it never happens, MagicMoonlight, requires X to always be less than Y, which cannot hold true forever. Eventually, X will grow to be more than Y, simply due to (rightfully placed) increased regulations for building on Earth. > It would also be cheaper to build one on the moon, if it was free to build them on the moon. Arguably, that should be the original goal: Build the DC on the moon.

whatisthiseven 6mo ago

The first premise is reasonable. The second premise is so outlandish as to require dozens of other assumptions, optimistic outcomes for launch, and pessimistic views on the cost of earthbound systems. I care about the environment and I think we can keep earthbound systems, and also reduce their impact. Making assumptions about the feasibility of launch and the economic absurdity of orbital compute, but not affording the same assumptions for what could be done for earthbound systems, is confusing? And no, orbital compute is absolutely not far lower than earthbound in co2 cost. Because it doesn't exist at any scale. All orbital compute is solely dedicated to switching where it is best served. If you were to spitball numbers, are we even willing to assume orbital matches earthbound in compute total, dollar cost, uptime, or any beneficial metric? The only metric I see is just slinging silicon into space.

GMoromisato 6mo ago
I get skepticism that orbital datacenters will make sense economically. It would not surprise me if they never happen. But if they made economic sense, then they would be much more environmentally friendly. On CO2 emissions, we're talking about a factor of 37. That's not a small amount. Even a factor of 10 would be a major issue multiplied by terawatts of power. Sure, maybe Earth-bound energy gets cleaner, but it's not going to get 10 times cleaner before 2036.
- jurgenburgen 6mo ago
  > Even a factor of 10 would be a major issue multiplied by terawatts of power. Sure, maybe Earth-bound energy gets cleaner, but it's not going to get 10 times cleaner before 2036. Currently we’re not on that trajectory but if we don’t get there we might be too busy with civilizational collapse to need space data centers. It would require doubling the rate of 2025 renewable deployments on average for the next 11 years so technically and economically it’s feasible.
- beAbU 6mo ago
  What aspect of an orbital data centre makes it more environmentally friendly than a terrestrial one?

hommelix 6mo ago

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.

GMoromisato 6mo ago
Satellites below 500 km do not need energy to reenter. Their orbits will naturally decay from friction with the upper atmosphere. CO2 emissions from reentering satellites is far lower than CO2 emissions from terrestrial datacenters. If you need 1,000 tons of satellite for a 100 MW of compute, you're not going to get more than 1,000 tons of CO2. In fact, you'll get much less since metal doesn't release CO2 when it melts. 1,000 tons of CO2 is negligible compared to 40,000 tons at launch and invisible compared to 1.5 million tons for a terrestrial datacenter.

rasz 6mo ago

orbital datacenters are the new vertical farming

beAbU 6mo ago

> 2. The cost of orbital datacenters will approach the cost of terrestrial datacenters (in $ per token terms) That is one wild, batshit insane assumption. So much so that I did not bother to read the rest of your post.

verzali 6mo ago

https://open.substack.com/pub/thequantumcat/p/data-centres-in-orbit-are-just-as https://open.substack.com/pub/thequantumcat/p/data-centres-i...

wasabi991011 6mo ago

> 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?

ben_w 6mo ago

> 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. 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-ai 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.

tcp_handshaker 6mo ago

>>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 https://youtu.be/trgn7s5-YHc?t=140