Will we put data centers in space? (forethought.org)

The viability of space DCs seem to depend entirely on the existence of higher latency, lower throughput, single card to single server workloads, which IMO isn't the primary class of workload today (and if one holds the belief that frontier models will maintain their dominant position, this will remain rack/row/facility level, which isn't viable for orbital facilities to handle as currently designed).

If that workload does expand in usage, and the economics of orbital DCs make sense (with high launch costs, zero ability to do maintenance, and weight penalties for shielding and radiators), then surely ocean based data centers (e.g. Panthalassa) make more sense (lower 'launch' costs, high but existant latency on maintenance, no weight penalties), and if those make sense then surely the distributed land based facilities (e.g. Rune Energy) or home based facilities (e.g. Span XFRA) make even more sense (easier maintenance, low/no permitting overhead because they're covered by existing permits, reasonable quality terrestrial fiber internet).

The article makes several big assumptions that are questionable (experimental cooling can scale, permit issues on land will only become worse, etc). But more importantly, their analysis of the risks of placing a data center in orbit is completely off. They mention that it costs tens of millions to launch a missile to destroy a satellite, and it might be cheaper to simply replace it (based on launch costs). But one strike can hit many targets, those targets may contain hardware costing multiple times the value of their launch, and you will be out of compute while you scramble to launch a replacement. We can also assume that placing big fat valuable targets in space would have many actors wondering how to make it cheaper to destroy them.

For those reasons alone, and because compute will become more and more strategic for defense, I find it very hard to believe that data centers in space will represent even 5% of all compute in the next 10 years. I think a higher proportion will rather be underground or in bunkers than in space.

From the article:

Conclusions

Orbital data centers look more viable than we expected going in and the case rests primarily on launch cost. The engineering challenges seem significant but none appear intractable. Commercial off-the-shelf solar, passive radiators, and standard error-correcting codes seem sufficient for a baseline architecture. We expect limited demos and shipments of compute to orbit in the coming years but our best estimate is that orbital data centers are unlikely to represent a meaningful share of compute before 2030. They might scale quickly thereafter however, and are more likely to be bottlenecked by chip production rates rather than by launch cadence. A generally important dynamic is that terrestrial data centers likely face headwinds that worsen as compute scales up while orbital data centers face constraints that relax with scale.

The more I read HN the more I realize that I am more of an empiricist than a rationalist.

So many people here are opining as if they are correct. I have a large amount of doubt and I like the fact that the world is structured so that people can try things out and see if it works.

If the world was run by the naysayers in this thread, almost nothing would ever get done.

This premise always struck me as a solution in search of a problem, which when implemented introduces even more problems.

Even put launch cost aside and assuming the best case for starship, most of the whitepapers I've seen seem to hand-wave the weight, cost, and complexity of cooling. You need large panels with special coating, pumps, and pipes - and that will get hit with micro-meteors so you need redundancy.

Land based or even sea wave-based datacenters are all much more feasible.

Remarkably few comments are engaging with the actual assumptions, arguments, and calculations in the article. Beware vibes, in both directions!

You can put computers in space, but can you really call it a data centre at the scale that it can be done at?

Surely a single AWS/GCP/Azure/etc DC on earth is so big we won't be getting close to a similar amount of compute in space for the foreseeable?

The economic case also needs to be explained. What are people going to do with a DC in space?

The only use case for datacenter in space is to have high communication speed with satellites that have low compute. Though that does beg the question of why not just add more compute to those satellites going forward.

Perhaps companies should consider putting their data centers underground (out of sight and out of mind). Space seems incredibly expensive and risky in comparison.

Outside cost of launch I have understood that scaling it to sufficient power envelope with needed cooling is biggest step which most likely makes it unreasonable at scale. That is being able to use enough power to get reasonable amount of compute. Scale it small and low power enough and it is pretty simple. But at that point you might not be talking about amount of compute you expect.

Especially fun when you consider how often gigawatts are talked about with AI. And what that would mean just for cooling as in area... Or then dividing that to sufficient number of individual satellites...

Wouldn't it be cheaper to build a nuclear plant to power data centers than to launch the data centers into space? Nuclear is expensive, but rockets, orbital infrastructure, and giant cooling radiators aren't exactly budget options.

For this article, they put their thumb on the scale. It's just nonsense conclusions from bad assumptions.

A 100kW Earth based solar array and a 500kWh grid scale battery that will last ~20 years is already a tiny fraction of the cost of even one Starlink satellite. It doesn't matter even if it was free to put in orbit, as only lasts five years once it's up there.

Their radiator assumptions are also laughable. 13x-28x better than ISS with carbon wizardry entirely on paper. TRL level what exactly?

Google's paper on their Suncatcher project (where they recently put their TPU in orbit and received a response from a prompt) is pretty interesting to read:

https://www.cell.com/joule/fulltext/S2542-4351(26)00362-4

Their blogpost is more "marketing-ish" but the videos are worth a watch too:

https://blog.google/innovation-and-ai/models-and-research/go...

We barely even have functioning trains or planes.

There's a lot of long shot assumptions here, but I feel the biggest is $250/kg launch costs (let alone $100 or $50).

Outside of the physical, technical, and maintenance challenges of putting data centers in space, I simply don’t want more trash littering the night sky.

https://youtu.be/yZpbJJLk1ug

The discussion of cooling in this piece is more convincing than I would have expected, but there remains a glaring social problem:

Should we consider orbit to be outside of terrestrial tax jurisdiction?

The majority of the cost of putting these data centres into orbit would be borne by terrestrial jurisdictions. The places that built institutions, funded education, provide medical care, etc. for the people who built these things. The environmental impact of building these data centres and launching them will be paid by those living on Earth's surface.

On Earth, we often fail to recoup many of these costs because of out-sourcing, but at least some other jurisdiction benefits. If a poor country is lifted up by the benefits of out-sourcing, at least some good has been done. If we let billionaires simply blast off and evade taxes, that's a net loss to the societies that made this possible with no benefit to anyone but the billionaires. That's fundamentally wrong.

Space may belong to no one nation, but neither is it true that space belongs to no one at all. The space around Earth belongs to all of humanity. Orbital data centres should pay terrestrial taxes. It's not clear where those taxes should go, but we need to hammer out the laws to make sure they are paid.

What happens to all the rare earth materials once the space DC have been decommissioned? Isn't it a horrible waste of the already hard to come-by resources to send it to a disposable trash heap in space, all in the name of forever chasing profits?

Wouldn't it be far cheaper to put DCs in offshore windmill parks? You have the power, cooling water abundant, fewer "NIMBY" than onshore DCs?

Is this already done?

How feasible would that be? Don't we have to bring down some satellites from the orbit at some point? How would that work if space datacenter has a limited time use?

It's really hard to get electronic "space grade" there is a lot more radiation there.

Also I wonder how long the ODC needs to operate to offset the energy spend on getting it in orbit.

Underground or in skyscrapers make more sense

For people selling the anti-sabotage / anti-datacenter-terrorism angle: the math doesn't work.

The energy needed to achieve orbital altitude is vastly less than the energy needed to reach orbit. About 1kg of rocket fuel will carry a small explosive and some tiny steel spheres to LEO altitude. The projectile will not be in orbit; it won't have escape velocity -- in fact it will briefly hit zero velocity (wrt the earth) at orbital altitude and then fall back to earth.

The very large cloud of shrapnel created will encompass LEO orbital height for several seconds. If an orbiting datacenter passes through the cloud during those seconds, the shrapnel will be moving at orbital velocity wrt the datacenter. In other words, just one piece will take it out.

With relatively simple rocketry (and, these days, an abliterated LLM to aid you) a dedicated luddite would have several opportunities per day to hit one of a constellation of satellites. If they wanted one very specific satellite they'd have one or two chances a year. The targeting and steering is not hard -- the orbital paths are publicly known and easily observable; you can use RF time-of-flight to implement exceptionally precise position/altitude/velocity measurements from the rocket (no need for GPS -- use your own transmitters on the ground). This is not hard. And it will only get easier as AI advances.

*I strongly discourage anybody reading this from doing the above.*

I'm posting it in order to show people that anti-sabotage is absolutely not a selling point for orbital datacenters. If anything, they are far more vulnerable to luddite attacks. The people channelling billions into this tech know this too. So their motive is something else. I don't know what, but it isn't this. The math doesn't work.

No