Hacker News
points by overfeed 3 days ago | 0 comments
> FWIW, given how many places in the U.S. are passing datacenter bans

If you have the technical chops to set up a data center in space — with no connections to the electrical grid or municipal water, you can use your technology to build your data center with the same specs right here on earth in 90% of places that you think are issuing "bans". You'll also get much better cooling for free via convection.

A remote town with a population of 43 will be happy for a data center to employ 2 locals, but data center operators aren't doing that today because they want tax breaks tiny towns can't give, infrastructure tiny towns don't have, nevermind a well-armed police force to deter would-be GPU thieves. Space is a red herring.

[–] robotresearcher 3 days ago
If all the energy for a space appliance comes from the sun, in steady state they only have to radiate away the same amount that hits them. All long-lived spacecraft manage it.

Whether they can be worth the money it cost to put them in orbit is another story.

[–] overfeed 3 days ago
> All long-lived spacecraft manage it.

Apples to oranges, long-lived satellites have modest power needs. Doing a back-of-the napkin math on required size of solar array and radiative elements for a single-rack's worth of energy is a sobering exercise for those who are a overly exuberant on the topic. The all-in power estimate for a single GB300 chip is 2.1kW

[–] robotresearcher 3 days ago
Well, it can’t use more than it can get from the sun that hits it. So it can’t need to radiate more than that away. So it’s radiator size needs are in proportion to its solar panel size needs, and no worse than that.

Edit: ISS is about 90KW steady state, and the collector to radiator area ratio is about 5:1. Dunno about the mass ratio.

My point is not that space data centers are viable or sensible. Only that the need for cooling is limited. If a terrestrial datacenter was powered only by its rooftop solar panels, cooling it would be trivial. Space is harder, but there just isn’t that much heat to dump.

A simple rock in orbit radiates exactly as much heat as hits it from the sun, on average.

[–] overfeed 3 days ago
> Well, it can’t use more than it can get from the sun that hits it. So it can’t need to radiate more than that away

You are correct that the first law of thermodynamics dictates the energy, but anyone who has parked a car in the sun will tell you, the temperature where the absorption/dissipation equilibrium is achieved is often incompatible with the survival of the cargo.

edit to your edit:

> Edit: ISS is about 90KW steady state, and the collector to radiator area ratio is about 5:1. Dunno about the mass ratio.

Cool; that's about half the wattage of one modern full-height Nvidia AI rack (172kW) - you need two ISSes to service just 1 rack; extrapolate that to a data center's worth of racks and see if the math is mathing. The ISS is about the size of a(n American) football field.

[–] robotresearcher 3 days ago
Again, just to remind you, I didn’t claim space datacenters are economically rational.
[–] overfeed 3 days ago
My original claim, in its entirety, was that on earth "You'll also get much better cooling for free via convection."
[–] robotresearcher 3 days ago
You sure will. My contribution is that you don’t need much better cooling.
[–] robotresearcher 3 days ago
The average temperature of a rock, car, or the ISS in low earth orbit is about -18C.
[–] Sammi 3 days ago
It's this huge misconception that space is super cold. A better mind model is that space doesn't have much of a temperature at all just cause there's nothing there. Vacuum doesn't have a temperature.

You won't immediately freeze if you enter space without a suit. You'll die almost immediately, but it will be because your blood will begin to boil because of the lack of pressure kinda like an extreme form of divers sickness - it's just not the cold that'll get you first.

[–] robotresearcher 2 days ago
That’s all true. But stuff in space has a temperature, and once I’d been dead a while in LEO, mine would be -18C.
[–] adrian_b 2 days ago
There is the additional problem that the GPUs are not in the panels, so to radiate the heat you must first transport it from the GPUs, then distribute it uniformly over the entire surface of the radiators, otherwise some parts will overheat.

Without gravity, it is likely that this can be done only by pumping continuously a lot of liquid, which will increase the power consumption and it will reduce reliability.

Solar panels that have heat radiators on their backside will have a higher temperature than simple solar panels, where half of the absorbed solar energy is converted into electrical energy and taken away. The efficiency of solar panels decreases with temperature, so hotter solar panels will have to be made bigger for the same electrical energy output.

Maintenance would be too expensive, so a space datacenter would have to be overprovisioned, e.g. with a double number of GPUs, to keep the average of the number of still working GPUs at the desired level over the lifetime of the datacenter. This will also increase the power and cooling requirements in the early part of a life cycle, and both the power generation and the cooling will have to be dimensioned for the maximum values, not for the average values.

It is certainly possible to design a datacenter that will work in space, but there is a very large number of difficulties that will increase the cost both in comparison with terrestrial datacenters and in comparison with existing satellites.

Until a detailed design is done by someone, it is not possible to know exactly how many times more expensive will be a space datacenter, but my opinion, as an engineer with experience in industrial design, is that the cost will be too high to make it worthwhile.

[–] verzali 3 days ago
Well yes, but what matters is how hot that rock gets. Some rocks melt or vaporise in space (see comets).
[–] robotresearcher 2 days ago
That’s right, and it’s not very hot: -18C on average in LEO.

You know space rocks where we live don’t melt because the Earth and moon don’t melt.

[–] bryan_w 3 days ago
It's not a technical problem. It's a people problem. Every data center project over the last year has experienced some form of pushback no matter where they're built
[–] overfeed 3 days ago
By definition, a data center in space is remote, self-reliant on electrical power and doesn't use potable water. These features require technical solutions before they become practical, and once one has cracked those challenges, they have disarmed all the push-back arguments people have raised against data centers. It's telling that the data centers want to build-out near population centers, i.e. where someone else has paid for fiber, grid and water infrastructure; it's a good old fashioned raid of the commons. This is why no-one is pushing for building data centers in locales with 0.2 people/sq mile with no one to subsidize the costs.
[–] mensetmanusman 1 day ago
This is first order thinking, of course buildings are built near builders…

building where people live is raiding the commons? Lol

[–] amanaplanacanal 2 days ago
Likely because the AI companies that are going to use those data centers have the absolute worst marketing imaginable. "We're going to take away your job and might accidently kill everybody on earth!" It's no wonder they get pushback.
[–] chasd00 2 days ago
> A remote town with a population of 43 will be happy for a data center to employ 2 locals, but data center operators aren't doing that today

This isn’t true, many of these DCs are in very very rural areas and get pushback from people not even living there. Governors get pressure from people hundreds of miles away of the site and then shut it down or impose moratoriums. It’s FUD on social media trying to slow AI advancement that is the root of all of it. The genie is out of the bottle, it’s like blacksmiths trying to slow industrialization. It’s already happened.

[–] mike_hearn 3 days ago
Not really. In sun synchronous orbit you can use solar power without battery backup. You can't do that on Earth because of night time.

Remote towns with populations of 43 are exactly the sorts of places that are pushing back on having datacenters in their areas.

And space is the ultimate physical protection. In reality a bigger issue than GPU thefts are protests or leftist groups trying to firebomb you. See the problems the Tesla factory in Germany has had with leftist violence, like setting its electricity pylons on fire (because electric cars "consume natural resources and are not sustainable", lol). The left has become very anti-AI and so the risk of attacks on facilities, especially if built outside the USA, is quite high which in turn means you have to overbuild physical security like high fences.

[–] adrian_b 2 days ago
The number of slots in the sun synchronous orbits is limited. More slots can be accommodated if the satellites can maneuver to avoid collisions.

Moreover, a datacenter in a sun synchronous orbit must be able to maneuver anyway, otherwise without adjusting the orbit it will enter in the shadow of the Earth after a month or two, due to the revolution of the Earth around the Sun.

If the datacenter must make periodic adjustments of the orbit, it will need to carry a lot of fuel, especially because a datacenter will be much heavier than a typical satellite, so much more fuel will be needed for orbit changes.

Thus it is not certain that a sun synchronous orbit is the best choice. It is possible that the extra fuel required for it may make this choice worse than using another orbit where the datacenter will not have power for half of the time.

[–] mike_hearn 2 days ago
That's an interesting choice. All the discussion I saw before assumed sun-synchronous, you're the first to suggest otherwise.

Do you know what the capacity of this orbit is?

[–] amanaplanacanal 2 days ago
Antipathy towards AI isn't just on the left, you know. At least in the US it seems widespread, and the remote towns with population 43 are far more likely to have right wing inhabitants than left wing.