Which is a synonym for impossible. As far as the latest serious analyses have
shown, data centers in space are a pipe dream. Starlink’s total addressable
market was also shown to be way smaller than expected. The IPO in this case
just signals they’re desperate for liquidity and with no clear path to
profitability, if you discount unlikely, major breakthroughs happening very
soon. They’re changing the rules of stock indexes just to shove SpaceX in.
It goes to show how far the establishment is willing to go to save face. Elon’s
company going under would poke an unpatchable hole in the US entrepreneurial mythology. They can’t afford that right now and they rather crash the whole economy.
Could you link to those serious analyses? The ones I've seen don't portray it as a total impossibility? Scott Manley did a runthrough that seemed reasonably positive on the possibility.
> ... but how do you scale this up right 20 kow used to be enough to power a full rack of computer gear but we're seeing predictions now of 100 kilowatts per rack and that's just one rack in a data center racks right the 48U 19in rack which is you know big and it's just dense with computers how does that fit into a flat satellite well it turns out that like 19in rack is about 50 cm wide it's about 1 m deep per unit and if you've got a 24 1/2 square meter satellite and you take all your one U units and stack them ...
One rack per satellite doesn't seem like it is that compelling of a story.
Put a 100 kW rack in my basement in the winter and hook up the power and I'll be happy to deal with the waste heat for several months (and then ship it to somewhere in the southern hemisphere).
Could you build a data center in space? Yes, absolutely I am sure there are no physical barriers. We have computers in space now, and those computers have telecom links to Earth.
Without even going into the numbers, terrestrial data centers have significant cost advantages. They don't have to spend $$$$$$$ to get to orbit. They can upgrade and/or fix components easily (likely safe to assume a hypothetical orbital DC would plan to never replace anything). They don't have to pay for the full capex of their power generation facilities. Lower-latency Internet. Heat dissipation is a (possibly unsolved?) problem. For every input cost to a data center, moving it to orbit massively increases that cost.
From a pure engineering standpoint: orbital data centers are not optimized to solve any common problem faced by data center operators or users. Permitting can get difficult in parts of the US, but at least permitting is a solved problem.
I think you're understating the permitting problem - it's a major reason for the very large/rapid price hikes on power in the PJM region, and the populist backlash against data center construction, including moratoriums on DC construction. The difficulty in getting new electrical generation interconnected in many parts of the US is one of the major marks in favor of the plan.
I'm not understating it. But I'm not buying the line that suddenly it's impossible to build industrial buildings in the US. I am realizing that there are thousands of jurisdictions in the US with wildly different permitting regimes, and then hundreds of other countries in the world that might be more welcoming.
But let's say they need to stay in the US. Are DC operators offering to buy down utility capex costs so that existing residents don't see a spike in rates? If not, obviously that is going to create opposition as nobody wants their utility bills to rise rapidly. It would probably be cheaper & easier to e.g. write a check to Southern Company to prevent rate hikes directly tied to their DC than to put a DC in space.
The math also barely pencils? IF Starship hits its $100/kg, getting a single rack of servers to orbit will cost ~$100k. A 500MW data center might have ~5k racks, so ~$500m to orbit. SpaceX estimates $100/kg - $300/kg so it could be $1.5B - $2B just to put the racks in orbit, plus the cost of the servers, plus the cost of the actual orbital data center itself, plus the cost of getting the orbital data center to orbit. That's getting into the "hand every resident a check for $100k in exchange for their county approving the permit" territory.
One Vera Rubin rack costs $3-7M and eats something like 600 kw, so you’re probably looking at more like 800 racks for that 500 MW DC. $100k launch costs per rack doesn’t seem too terrible if that’s what it works out to. I’m sure there’s a mountain of solar panels that aren’t included, though?
And it’s not that simple, building out power generation is very constrained, the interconnection queue is years long in many places, and the current backlog for new natural gas turbines is multiple years right now. Fixing the permitting isn’t impossible with some political will, but energy permitting reform is something that’s been bandied around for years in Congress and hasn’t made it across the finish line. EPRA almost made it at the end of last Congress but that session ended before it did. Hopefully it makes it this time, everyone should contact their congresspeople and ask them to support energy permitting reform.
You will have a setup working based on solar energy and battery storage before you get spaceship to not explode anymore and to deliver low price for payload.
And we are talking about AI Datacenters, they are a lot less latency dependend than websites.
Alone the idea that Musk would be able to break through any burocrazy for space stuff and sets up a supply chain for everything space is easier than just setting up some energy and fiber, feels ridicoulys
I don't understand how years spent building an orbital DC is better than years spent permitting. I guess maybe they expect to somehow be able to build these faster? (How?)
Anyway, is it technically possible? Yes. My suspicion is it's at best a wash vs building on the ground. Applying a similar price premium & similar engineering resources to a ground-based system is likely to deliver much more predictable results.
Tech obviously can have success at lobbying. The TikTok ban IIRC got 90 votes in the Senate. I'm sure the total cost of the required astroturf campaign was much less than launching a single orbital DC.
As it stands, most if not all institutional and journalistic research around this topic I would consider compromised because they’re in some way or another financially interested in this becoming the next big thing. Aravolta included. That’s why most articles will counter each hard constraint with a handful of hopeful speculations.
As for pure scientific analysis, like the Scott Manley one, they tend to entertain themselves too much with the physics and mathematics and forget the economics behind it all.
Take Google’s own paper (https://arxiv.org/pdf/2511.19468) that estimates that launch costs, just to roughly match data center energy costs on earth, would need to reach 200USD/kg, which requires a 10 fold cost reduction relative to the current launch costs of Falcon 9. And that is to launch a _disposable_ server into orbit, that will disintegrate after a few years and likely have hardware failures well before that.
And these servers are not anything like a “data center”, and they won’t run the applications that we are already scrambling to find demand in earth. No, these would theoretically run some ultra-niche, highly experimental workloads maybe for NASA or the military. That alone can’t possibly justify the investment, at least not for the retail investor that actually expects a positive ROI. Nevertheless the tech elite and their pet journalists are more than happy to sell this fantasy to the average people.
Hell, I’m still waiting for Project Natick to materialize, Microsoft’s data center on the ocean, which makes far, far more sense than data centers on frigging space. Still they didn’t manage to make that one work in any meaningful sense.
Thanks, I took a look, couple things - the inlet temp on the VR is 45 C, but that’s not the radiator operating temp, you can probably run those chips closer to 90-100 C. And they’re building custom silicon for this, presumably that’ll be one of their design targets. Also, most bit flips should be fine when you’re running inference, you’re presumably running with some randomness anyway. If a node fails/becomes too unreliable, it can be detected and shut off.
Idk, building in the ocean seems a lot harder to me than space. Salt water is ridiculously corrosive, extreme pressure, etc. And one of the main justifications for this is massively increasing the output of solar and making it consistently output its nameplate capacity, which space is great for, and ocean is terrible for. The only benefit for that one I can see is some power savings on cooling, and a whole boatload of drawbacks, whereas we might not be able to keep up with demand with terrestrial power. So I can totally see why they never bothered to complete their subsea datacenter.
They’re definitely not aiming to put niche applications up, they want to run models by the bucketful.
I don’t see how the economics make it impossible? To be clear, I’m not saying that it’s something that’s going to end up happening, I have no idea if it will, but I don’t see how it’s structurally impossible, and I can see some things to commend it if token usage volumes grow like I think they will.
