Cooling in space Cooling in space

Posted 2026-06-11<br>With all the talk of data centers in space, I remember my first question being “isn’t cooling this stuff a fundamental limit on whether we can have orbital data centers at all?” Turns out that at least this part is definitely doable and quite possible.

I’d like to give a quick thanks to Saurav for convincing me of this originally and Parth for discussions/encouraging me to write this.

I’ll go a little bit slow on this one, since my guess is readers who are interested in this particular topic probably need a little more background. If you don’t, you can skim this next subsection and head directly to the next one.

Why do we need cooling?

Ok, assume we have a datacenter in space and all of its electrical energy comes from solar cells, pointed at the sun.

In this orbital data center (which, in my head, I picture as a big box with solar panels sticking out of it, facing directly at the sun) we have a bunch of GPUs. It’s possible we also have some communication equipment (to send data up to and down from the satellite) and maybe some basic power equipment, but assume that ~ all of the power goes to the GPUs.1

A GPU, in short, is a machine that takes in electricity and outputs heat (and, as a side effect, performs some hopefully-useful computation). In fact, essentially all of the electricity input is converted to heat, which, if not dumped somewhere, will eventually cause the GPU to melt and become a very expensive paperweight made of copper and silicon.

In other words, we need some way of dumping the heat out of the GPU, ideally to some large reservoir.

On Earth, consumer GPUs do just fine dumping it out onto “the air on Earth” by running room-temperature air over the GPU. This transfers the heat from the GPU onto the room temperature air (by heating it) and the resulting hot air is blown out of the GPU and replaced with room-temperature air again. Do this quickly enough and the GPU stays relatively cool—ideally, not far above room temperature, which is fine for most intents and purposes. For larger GPUs or collections of GPUs, like those used in data centers, the mechanism is roughly the same, except we use slightly fancier coolants to transfer heat away from the GPU more quickly and spread it out over a larger area (which results in more room-temperature air coming into contact with the larger area, which, in turn, allows more heat to be dissipated).

The problem, of course, is that space has no air to exchange heat out to. In fact, since space is mostly a vacuum, there is no other medium to exchange heat!

So: what do?

Blackbody radiation

From some basic physics, we know that a body at any temperature will dissipate heat as radiation.2 For example, humans (and most warm-blooded animals) radiate infrared light pretty strongly which is why “thermal cameras” work in the first place. (“Thermal cameras” are just cameras that, instead of seeing red/blue/green light, see infrared light.)

There is a pretty famous law, called the Stefan–Boltzmann law, which relates the total amount of power radiated to the temperature of the object, and is a very good approximation for most objects. It states that the power radiated Prad satisfies

Prad=σAT4,<br>where T>0 is the temperature of the object in Kelvin, σ is some proportionality constant (derived experimentally or, alternatively, from integrating Planck’s law) and A is the effective area of the object.3 Potentially there’s a factor ε1 called the emissivity in front of the whole thing, but we’ll assume ε∼1 for our rough estimate, since we will generally have something like ε>.9.4

This gives us a way of “cooling” down some piece of equipment in space, without needing air or another effective cooling medium. To do this, first we can conduct the heat from the chips (which we’d like to maintain at some temperature T, which is, say around T∼330 Kelvin or ∼60 Celsius) out onto some wide region with area A which lies under the solar panels, which we will call the radiator as it, well, radiates.

How do we achieve this practically? One idea would be to use some fluid (which can be a liquid or gas) that has relatively high specific heat to transfer heat away from the chips. We can pump the fluid through the top of the chip, which heats up the fluid by removing heat from the chip, and the fluid’s own heat is then dumped into the radiator, which cools by, well, emitting radiation.

Usual choices for this fluid on Earth include water or some glycol water mixture, but water (and/or glycol) is very heavy and each pound that we add substantially adds to the cost of launching our orbital data center into the atmosphere. Some gases such as CO2 or hydrogen (in spite of flammability) can perform a similar function at reduced weight and efficiency, though obviously gas leaks are more annoying to deal with in practice.

This is the whole required loop for cooling the chips in space. (We will add a slight complication in a second to...