web page content

There was some waffle on the news today, ahead of the SpaceX IPO, about how one of its major earners would be by putting AI data centres into space. Maybe, but how are they going to cool these things? Their energy consumption is enormous.

Assuming they'll be powered by solar panels, will the side of the panel facing away from the sun be used as a giant "black body" radiative heatsink?

That would be sensible. It's looking at the universal back-ground radiation at about 4 degrees absolute, and the power emitted is proportional to the fourth power of its absolute temperature.

It's about 500 watts per square meter at 300 degrees absolute. At that temperature you can use water vapour to carry the heat out to the remote end of the radiating area, and pump the water back after it has condensed.

You might want to spin the area slowly to make it easier to collect water after it has condensed, before you pump it back to the areas where the heat is being generated.

Running everything a bit hotter will reduce the radiating area you need, but will increase the water vapour pressure you have to contend with in the vapour transfer channels.

That's the plan. Scott Manley on YouTube did a video a couple of weeks ago with a back-of-the-envelope estimation to see if it was feasible, and he concluded that it should work, barely.

AI commenting on AI:

"While traditional data centers might require a few megawatts, hyperscale AI data centers can require 1 to 5 Gigawatts (GW) of power—roughly the equivalent of the total energy consumed by a massive metropolitan area"

The radiating surface would have to be very hot to get rid of much heat, far hotter than semiconductors can stand. So there would have to be powered heat pumps between the semis and the radiators.

It ain't gonna happen, and the AI bubble will mostly pop soon too.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

I don't usually watch YouTube videos (TL/DW...), but that one at

My original thinking was that the "dark side" of the solar panels would have been the radiator, but I forgot about the wasted heat from the sunlight which can only go to heat up the panels, as even the best panels are, at present, only 30% or so efficient. So for every 1kW they can produce, about 2.3kW is going to waste and is heating up the panel. Well, that's unless you can add an efficient radiator to get rid of that heat. And he didn't even mention the degradation in conversion efficiency of solar panels over time. It all adds up.

So, as the video points out, the radiators will have to be separate and at right-angles to the solar panels (so edge-on to the sun). Now we've got extra weight and more cooling fluid required. Sheesh! That's going to take a lot of rocket power to get into space, and vast amounts of propellants.

The Mississippi river could sink a fair amount of heat. And Louisiana has a lot of natural gas. Good place for data centers.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

I guess I conflated side with back side; the radiator goes in the shade of the solar panels but yes, are at 90 degrees and aren't simply the back side of the solar panels, sigh :-). He also based his calculations on the largest complete assembly that could fit in the largest available launch vehicle, so no in-orbit assembly from multiple launches. I don't remember his specific concerns but he definitely felt that an array of that sized satellites was better than assembling one giant satellite in orbit. Scott does good regular 20 min vids covering the latest news, launches and failures so makes it easy for a casually interested person to keep up with the field.

Radiation increases as the fourth power of temperature. You don't have to get the radiators all that much hotter to get rid of a lot more heat.

And you keep on thinking that the data centers will use conventional semiconductors, while everybody is trying to get quantum computers to work. The one we've got in Sydney runs in liquid He-3 at at 0.1K.

It may not happen the way you imagine it might, but your imagination isn't up to much.

So your "original thinking" wasn't so much "original" as hopelessly incompetent.

The back side of the solar panels would be radiating that heat anyway. There's nothing magical that obliges you to add separate radiators to radiate extra heat.

Why?

The waste heat dissipated by the solar cells doesn't have to be moved around by cooling fluid. The back side of the solar cells has always had to dissipate that waste heat. Piping in some extra coolant so the radiators run a bit hotter just changes the operating temperature you chose to run at.

Putting data centers into orbit was always going to be expensive. Thinking about exactly how expensive is part of the design process. Mindless commentators telling us that it is going to be more expensive than the number they first thought of is just pointless wittering.

If your thinking about computing elements they might use is confined to what you can buy off the shelf right now, that might be a useful commnet.

Back when computers were built out of thermionic valves you could have generated some even more extravagant sets of requirements.

People invest in the technology that will solve the problem they have to solve right now. As the problems change, the investment get redirected.

How many bits?

Imagine square kilometers of solar panels, and square kilometers of cooling panels with kilometers of water pipes. The solar panels need cooling too, obviously more watts than the computers.

And an orbiting hotel for the maintenance crew.

Now imagine maneuvering that to avoid being whacked by all the junk in orbit.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

Don't know. The academics involved have set up a development company, and I haven't kept track of what they tell potential investors.

Maintenance robots don't need hotels.

There's a lot of junk in low orbit. An orbiting data center isn't going to be in a low orbit that ever puts it in the shadow of the earth, and it is going to to whack the junk that comes close rather than dodge it.

A working solar cell in the same orbit would be a bit cooler, because some 27% of the incident energy would be converted to electric power and shipped off somewhere else. Running it hotter than 255C and using it as a radiator to get rid of excess energy seems perfectly sensible, but where would you get the excess energy from?

The problem with whacking space junk is that there's *more* of it afterwards.

Jeroen Belleman

Square kilometers are a lot of target, even for natural meteors.

I don't think that orbiting data centers are going to happen.

It's cheap to transport stuff to Louisiana on trucks or barges.

The up/down data links are another issue.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

Not if you do it right. Work out the orbit of the incoming junk, and hit it with a sticky blob that moves it into a more circular higher orbit.

You avoid breaking it up - that does generate more space junk. A laser hit that boiled off just enough of the junk to move the rest of it into a higher orbit wouldn't need the supply of sticky blobs. Finely dispersed nickel iron vapour probably doesn't count as space junk - it's just nanometeorites, and photon pressure should carry it away.

There aren't all that many of them. There is a whole asteroid belt that has been catching them for millions of years.

Elon Musk begs to differ, and he's just collected a trillion dollars from people who ought to know better.

If anybody would want to bother. Sea level rise is going to submerge the whole state real soon now.

For years international telephone calls went that way. That problem got solved quite a while ago. Fibre-optic cables are a better solution, and if we ever got around to putting up a space elevator we could use them for stuff in orbit.

I was thinking the same thing It is likely just Elon Musk trying to get more mony Seems he has forgotton his Mars plans. One nuclear power station on earth and ever more nano-nano faster hardware and all that stuff fits in a small building.. Like tubes in the past versus chips these days Or in evereybody's 'smart' phone.

With speeds approaching a terabit per fiber, maybe the world only needs a few dozen giant data centers, with their associated power plants and cooling.

Speed-of-light makes relatively nearby hubs better than ones way out in space.

Can RF ever ship petabytes per second to and from a satellite?

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

Yes, but only if the various regulatory organizations allocate the necessary bandwidth on a global scale. Like a bundle of optical fibers, the gross total bandwidth is the sum of the bandwidths of each optical fiber.

The real problem is available bandwidth. To get petabyte per second speeds, the system will probably use mm (millimeter) or sub-mm wave frequencies. Higher frequencies offer more bandwidth, but also offer more headaches. For example, atmospheric attenuation absorption requires more RF power to be usable:

"Clear air atmospheric attenuation"

The frequencies ruined by oxygen, nitrogen, ozone, water vapor, rain, fog, cloud, etc attenuation can't be used. Optical frequencies might be usable for a LEO (low earth orbit) system.

"United States Frequency Allocation Chart"

Another RF problem is the lack of a 100% reliable path. The orbital data-center will probably be launched into a LEO (low earth orbit). That makes the satellite a moving target, which requires a tracking antenna like the Starlink system. It also means there will be times when there is no signal to or from any satellite. That can be fixed by with aggressive buffering, but that makes real time data impossible. At this time, AI does not need to be real-time. However, I expect AI to evolve into a glorified "chat" program, which works best in a low latency and real time environment.

However, I don't think it's going to be a bureaucratic nightmare. I would expect the various governments to declare that optical frequencies (colors) need not be licensed, coordinated, or sold at auction. 2nd best would be something like internet domain name registration, where the early adopters and domain name entrepreneurs staked their claims on the best domain names leaving everyone else with whatever is left. I have no idea how to deal with optical interference problems. One should not expect order from anarchists.

Actually, I'm an optimist and suspect that all the above problems can eventually be solved. However, there's on problem with building an orbital data-center that might kill the idea before it leaves the ground. Space is full of very energetic particles, which do bad things to sensitive components and circuits. On earth, we have the protection of the atmosphere and the earth's magnetic field. We have the option of adding shielding but probably can't tolerate the added weight. Space semiconductors are designed to minimize the effects of energetic particles, but that drives up their cost.

"Why Radiation Hardening Matters for Satellite Processors and What It Costs"

If we disassemble a working terrestrial data center computer, launch it piecemeal, assemble the pieces in orbit, and flip the on switch, it would dead on arrival because of radiation damage. To fix that will likely require better semiconductors that might cost 1300 times the cost.

If a data center in orbit is such a great idea, I believe it could be done cheaper, easier, and more reliably by locating the data-center(s) at the north and south poles.