SpaceX: Datacenters in space part 2 — Martin Stellinga

Last week, SpaceX got a gazillion dollars in funding. The main driver: data centers in space. In February I wrote that I don’t think those are viable. Time for a re-evaluation.

What was this about again?

So, back in February, I wrote I didn’t believe data centers in space would work. Too heavy, too hard to power, too hard to cool, and economically nonviable.

Last week, though, SpaceX went public, meaning investors could buy stock, and boy did they. They raked in so much they are now the fourth largest company in the world. Most of that is because of the promises of the massive AI revenue that they will get for putting data centers in space.

Of course, that begs the question: Martin, were you wrong?

Short answer: I could be, yes. But let’s look at the numbers first. Because SpaceX published some information about what they intend.

Power

If you look at the infographic from SpaceX, they’re going to create satellites that have around 120 kilowatts of power for compute. That’s supposedly enough to power an Nvidia GB300 rack (a rack of AI processors). Interestingly, such a rack draws around 150 kWh, so 50% more power than they have.

And I think you need hard discs somewhere in your setup, and maybe some other peripherals. You probably need a transmitter to speak to Earth and probably several to talk to other satellites.

But, okay, if you look at the size of the solar panels in the infographic, and what a solar panel can produce, the power should indeed be around 100 kWh. And you can power an AI setup, if not the exact one mentioned. Maybe. Keep in mind, though, that one rack does not a data center make. Modern data centers have thousands, if not more.

Heat

All the power the satellite produces has to go somewhere. When used in for a processor, that means heat. So, that’s around a 100 kWh of heat to lose, and probably more from other components of the satellite. The infographic shows 100 square meters of radiators to dissipate the heat.

If I read this correctly, the ISS has two 13,6m by 3,12m radiators, so around 80 m2, and can dissipate about 70KwH, so 100kWh with 100 square meters seems feasible. I have no idea how much that costs, though.

Cosmic radiation

So, okay, you could launch a processor rack into space, and you could probably power and cool it. There’s another problem though. One I’d forgotten last time.

Processors do not do well with jolting motion. They also don’t do well with radiation. Unfortunately, a rocket launch is not jolt free. It pulls several g of force. And, when it gets to orbit, it’s past the magnetosphere, and its payload is exposed to all the cosmic radiation flying through the solar system.

But, Musk has a solution. As he says in the article: they’ll build radiation hardened chips! There’s no timeline for those, though. Yeah, that has me brimming with confidence. I bet they’ll be done right after the self-driving Teslas arrive… And I suppose they won’t cost more than regular chips.

Scaling up

Like I said, one rack does not a data center make. You need thousands. Meaning, you need thousands of these ‘small’ satellites, or make them bigger. Musk wants millions.

If I look at the designs, these mini satellites will weigh a ton or two. So, you need to launch millions of tons of crap into space. And since the lifespan of these servers is two to six years, at which point you probably need to replace the entire satellite, you’re looking at a few million tons a year or so.

Oh, look, that’s exactly the projection Musk made in his presentation. Let’s launch ten million tons of expensive stuff into space each year where it will fall back into the atmosphere and burn up a few years later. Good recycling practice.

But, worse, a falcon heavy can lift around 64 tons, meaning ~15.000 flights a year. According to this article, every launch leads to 440 tonnes of fuel use, and roughly 146 tonnes of CO2. For 15k flights, that’s…. 2 million tons of CO2 per year. That’s roughly equivalent to 2 TWh of energy from coal plants. And Co2 in the upper atmosphere from a rocket is worse than close to the ground. That’s not very environmentally friendly.

And as with all environmentally damaging plans: SpaceX might benefit, but humanity as a whole pays the price.

Networking

Scaling up to a massive fleet of satellites sounds fun in theory, but it has one additional downside. It requires communication. Data centers have massive numbers of cables running from every place to every other. That creates a low-latency network of computing power. That doesn’t exist in space.

A SpaceX fleet of satellites will have to communicate through radio, or possible laser, which is slower than fiber optic cable. On top of that, data has to travel up from the Earth’s surface, and back down. That’s a solvable problem, of course, but it will never be as efficient as an Earth-bound data center, unless you can lift a cabled data center satellite into orbit instead of a fleet. And that doesn’t seem likely.

Conclusion, a data center in space will be less efficient and powerful than an Earth-based one. Of course, if the prices are lower, that doesn’t have to be a deal breaker.

Costs

When you start looking at costs, things get even more murky. Sending stuff into space isn’t cheap. As the article I linked pointed out, a SpaceX Falcon Heavy ‘only’ costs $1.300 per ton. So that means it ‘only’ costs around a billion dollars per year to launch all those satellites.

Yeah…

But that number doesn’t say much. The real question is: how do the costs stack up against earth-based data centers. If we really want AI ( I don’t think we should, but fine), and want to pay for it, the question is: which data center is cheaper. If an Earth-based data center costs billions more, space is the way to go. So, a table with cost items (back-of-napkin style, of course):

Earth: Processors Space: Hardened processors (more expensive) and they cannot be recycled

Earth: Wind + Solar panels (possibly nuclear power) once. Space: New solar panels for every satellite, so 1.400 m2 panels per satellite per 2 to 6 years (and there are many satellites).

Earth: One building needed Space: Many rocket launches needed

Earth: Water needed for cooling Space: Cooling done through radiators in space, but you need to build 100m2 of radiators per satellite per 2 to 6 years (and there are many satellites).

If you look at these numbers, the data centers in space are more expensive in most categories. Unless there is no power or water or space left on Earth. And that does seem to be a bottleneck for data centers at the moment.

Hype’s end

But, as much as the AI hype train likes to pretend AI will grow forever, that seems… doubtful. The magical revenue streams of AI, in the gazillions, are not materializing. Basically, Big Tech seems to be investing with borrowed money until the clock strikes twelve and Cinderella’s AI-generated dress evaporates, leaving her in rags.

This means that when the growth dies down, and the bubble bursts, we’ll be left with a reasonably stable set of data centers, instead of the mad near-war-economy driven race to build as many of the things as possible. And with a reasonable pace, I doubt space-based data centers can cut it.

I think it says something about our current system that the fourth most ‘valuable’ company in the world has never turned a profit. The profits will have to go from billions in the negative to tens of billions in the positive, in a few years. And profit is not revenue. Revenue will have to go much higher. No company in the history of man has achieved this feat, but this company — which has never turned a profit — will supposedly do so in a few years.

I call bullshit. This seems like a pyramid scheme, and everybody is now throwing money at it because they hope to get some profit out before the clock strikes twelve. Or they just ‘had to be a part of this historic event’. And if it’s up to Big Tech, our governments and pension funds will be left holding the bag.

As Cory Doctorow wrote earlier this week:

SpaceX is just a continuation of the logic of crypto, in which something is valuable because some people think other people will pay more for it in the future, and not because it does useful things

- Cory Doctorow, pluralistic.net

Conclusion

Most of what I said previously still seems to be the case. I will admit, actually having some working data center in space seems more technically feasible than I thought, but economically feasible… I seriously doubt it.

It feels as if Musk is using his con-man act to lure in AI hype money. And I think all the investors know it’s a scam. They just think they’ll be able to pull up in time, or they just want a bonus this year, and to hell with next year.

I just hope it’s over soon, before Musk starts gleefully killing our future with rockets.

Originally published at https://martinstellinga.com on June 19, 2026.