Summarizer

Yeah it does not make a whole lot of sense as the useful lifespan of the gpus in 4-6 years. Sooo what happens when you need to upgrade or repair?

They can run these things at 100% utilization for 3 years straight? And not burn them out? That's impressive.

Not really. GPUs are stateless so your bounded lifetime regardless of how much you use them is the lifetime of the shitties capacitor on there (essentially). Modulo a design defect or manufacturing defect, I’d expect a usable lifetime of at least 10 years, well beyond the manufacturer’s desire to support the drivers for it (ie the sw should “fail” first).

The silicon itself does wear out. Dopant migration or something, I'm not an expert. Three years is probably too low but they do die. GPUs dying during training runs was a major engineering problem that had to be tackled to build LLMs.

> GPUs dying during training runs was a major engineering problem that had to be tackled to build LLMs. The scale there is a little bit different. If you're training an LLM with 10,000 tightly-coupled GPUs where one failure could kill the entire job, then your mean time to failure drops by that factor of 10,000. What is a trivial risk in a single-GPU home setup would become a daily occurrence at that scale.

> the useful lifespan of the gpus in 4-6 years. Sooo what happens when you need to upgrade or repair? Average life of starlink satellite is around 4-5 years

Starlink yes, at 480 km LEO. But the article says "put AI satellites into deep space". Also if you think about it, LEO orbits have dark periods so not great. A better orbit might be Sun Synchronous (SSO) which is around 705 km, still not "deep space" but reachable for maintenance or short life deorbit if that's the plan. https://science.nasa.gov/earth/earth-observatory/catalog-of-... And of course there are the LaGrange points which have no reason to deorbit, just keep using the old ones and adding newer.

A "fully and rapidly reusable" Starship would bring the cost of launch down orders of magnitude, perhaps to a level where it makes sense to send up satellites to repair/refuel other satellites.

With zero energy cost it will run until it stops working or runs out of fuel, which I'm guessing is between 5-7 years.

> Sooo what happens when you need to upgrade or repair? The satellite deorbits and you launch the next one.

> Bezos has been pushing manufacturing-in-space for a long time, as a ideal candidate for what to do in space that you might prefer to not do on Earth. Robotics, AI automation, manufacturing - combo it in space, let the robots manufacture for us in space. LOL, this seems so far off from the reality of what manufacturing looks like in reality. - sending raw materials up there - service technicians are necessary ALL THE TIME, in fully automated production lines - sending stuff back down Maybe I lack vision, but data centers in space is a 1000x times better idea and that is already a terrible idea.

The show For All Mankind kind-of hinted at how the labor problem would be solved: recruit like the military and promise huge bonuses that will probably not be realized because space is risky business

Well you see, what you do is send a bunch of humanoid robots up there to do all the work. (please don't ask what we do when those break down)

>>sending raw materials up there That's what asteroid mining is for. >>service technicians are necessary ALL THE TIME Optimus is already very well tele-operated. Even though over time it can likely be trained to do specific tasks far better than even humans.

> Optimus is already very well tele-operated It can't even serve popcorn in a diner.

And what happens every time a rack (or node) fails? Does someone go out and try to fix it? Do we just "deorbit" it? How many tons per second of crap would we be burning in the upper atmosphere now? What are the consequences of that? How do the racks (or nodes) talk to eachother? Radios? Lasers? What about the Kessler Syndrome? Not a rocket scientist but 100% agree this sounds like a dead end.

Just have to size radiators correctly. Not a physics problem. Just an economic one. Main physics problem is actually that the math works better at higher GPU temps for efficiency reasons and that might have reliability trade off.

Only people who never interacted with data center reliability think it's doable to maintain servers with no human intervention.

Microsoft did do the experiment (Project Natick) where they had "datacenters" in pods under the sea with really good results. The idea was simply to ship enough extra capacity, but due to the environment, the failure rates where 1/8th of normal. Still, dropping a pod into the sea makes more sense than launching it into space. At least cooling, power, connectivity and eventual maintenance is simpler. The whole thing makes no sense and is seems like it's just Musk doing financial manipulation again. https://news.microsoft.com/source/features/sustainability/pr...

Salt water absolutely murders things, combined with constant movement almost anything will be torn apart in very little time. It's an extremely harsh environment compared to space, which is not anything. If you can get past the solar extremes without earths shield, it's almost perfect for computers. A vacuum, energy source available 24/7 at unlimited capacity, no dust, etc.

You are right. But in the future we'll be refueling the satellites anyway. Might as well maintain the servers using robots all in one go.

I'd assume datacenters built for space would have different reliability standards. I mean, if a communication satellite (which already has a lot of electronic and computing components) can work unattended, then a satellite working as a server could too.

Whoa there, space-faring sysadmin. You really want that off-world contract tho?

Haha, hard pass on the job. I prefer my oxygen at 1 atm. I'm not a data center technician myself, but I have deep respect for those folks and the complexity they manage. It's quite surprising the market still buys Musk's claims day after day.

There are a class of people who may seem smart until they start talking about a subject you know about. Hank Green is a great example of this. For many on HN, Elon buying Twitter was a wake up call because he suddenly started talking about software and servers and data centers and reliability and a ton of people with experience with those things were like "oh... this guy's an idiot". Data centers in space are exactly like this. Your comment (correctly) alludes to this. Companies like Google, Meta, Amazon and Microsoft all have so many servers that parts are failing constantly. They fail so often on large scales that it's expected things like a hard drive will fail while a single job might be running. So all of these companies build systems to detect failures, disable running on that node until it's fixed, alerting someone to what the problem is and then bringing the node back online once the problem it's addressed. Everything will fail. Hard drives, RAM, CPUs, GPUs, SSDs, power supplies, fans, NICs, cables, etc. So all data centers will have a number of technicians who are constantly fixing problems. IIRC Google's ratio tended to be about 10,000 servers per technician. Good technicians could handle higher ratios. When a node goes offline it's not clear why. Techs would take known good parts and basically replacce all of them and then figure out what the problem is later, dispose of any bad parts and put tested good parts into the pool of known good parts for a later incident. Data centers in space lose all of this ability. So if you have a large number of orbital servers, they're going to be failing constantly with no ability to fix them. You can really only deorbit them and replace them and that gets real expensive. Electronics and chips on satellites also aren't consumer grade. They're not even enterprise grade. They're orders of magnitude more reliable than that because they have to deal with error correction terrestial components don't due to cosmic rays and the solar wind. That's why they're a fraction of the power of something you can buy from Amazon but they cost 1000x as much. Because they need to last years and not fail, something no home computer or data center server has to deal with. Put it this way, a hardened satellite or probe CPU is like paying $1 million for a Raspberry Pi. And anybody who has dealt with data centers knows this.

Thanks for putting words to that; the paragraph which most stuck out to me as outlandish is (emphasis mine): The basic math is that launching a million tons per year of satellites generating 100 kW of compute power per ton would add 100 gigawatts of AI compute capacity annually, *with no ongoing operational or maintenance needs*. I'm deeply disillusioned to arrive at this conclusion but the Occam's Razor in me feels this whole acquisition is more likely a play to increase the perceptual value of SpaceX before a planned IPO.

for me trying to apply some liquid TIM on a CPU in a space station in a big ass suit would be a total nightmare, maybe robots could make it bearable but the racks would get greassy fast from many failed attempts

> but they cost 1000x as much Compute power has increased more than 1000x while the cost came down. I recall paying $3000 for my first IBM PC. > they need to last years and not fail Not if they are cheap enough to build and launch. Quantity has a quality all its own.

But … but what if we had solar-powered AI SREs to fix the solar-powered AI satellites… /in space/?

Maintaining modern accelerators requires frequent hands-on intervention -- replacing hardware, reseating chips, and checking cable integrity. Because these platforms are experimental and rapidly evolving, they aren't 'space-ready.' Space-grade hardware must be 'rad-hardened' and proven over years of testing. By the time an accelerator is reliable enough for orbit, it’s several generations obsolete, making it nearly impossible to compete or turn a profit against ground-based clusters.

On the other hand, Tesla vehicles have similar hardware built into them, and don't require such hands-on intervention. (And that's the hardware that will be going up.)

Car-grade inference hardware is fundamentally different from data center-grade inference hardware, let alone the specialized, interconnected hardware used for training (like NVLink or complex optical fabrics). These are different beasts in terms of power density, thermal stress, and signaling sensitivity. Beyond that, we don't actually know the failure rate of the Tesla fleet. I’ve never had a personal computer fail from use in my life, but that’s just anecdotal and holds no weight against the law of large numbers. When you operate at the scale of a massive cluster, "one-in-a-million" failures become a daily statistical certainty. Claiming that because you don't personally see cars failing on the side of the road means they require zero intervention actually proves my original point: people who haven't managed data center reliability underestimate the sheer volume of "rare" failures that occur at scale.

Do they need to be maintained? If one compute node breaks, you just turn it off and don't worry about it. You just assume you'll have some amount of unrecoverable errors and build that into the cost/benefit analysis. As long as failures are in line with projections, it's baked in as a cost of doing business. The idea itself may be sound, though that's unrelated to the question of whether Elon Musk can be relied on to be honest with investors about what their real failure projections and cost estimates are and whether it actually makes financial sense to do this now or in the near future.

AI clusters are heavily interconnected, the blast radius for single component failure is much larger than running single nodes -- you would fragment it beyond recovery to be able to use it meaningfully. I can't get in detail about real numbers but it's not doable with current hardware by a large margin.

>From lunar regolith you would extract: oxygen, iron, aluminum, titanium, silicon, calcium, and magnesium. Do we actually know how to do that? >From the poles From the poles! So the proposal includes building a planetary-scale railway network on bumpy lunar terrain. >The moon can supply mass, oxygen, fuel, and structure. None of those are things we are hurting for down here, though.

> The real constraint is not materials It's solvents, lubricants, cooling, and all the other boring industrial components and feedstocks that people seem to forget exist. Just because raw materials exist in lunar regolith doesn't mean much if you can't actually smelt and refine it into useful forms.

No maintenance either

For one or a few-off expensive satellites that are intended to last 10-20 years, then yes. But in this case the satellites will be more disposable and the game plan is to launch tons of them at the lowest cost per satellite and let the sheer numbers take care of reliability concerns. It is similar to the biological tradeoff of having a few offspring and investing heavily in their safety and growth vs having thousands off offspring and investing nothing in their safety and growth.

After a few decades, you need to start replacing all the solar panels. And the robot army being used to do the construction and resource extraction will likely have a much shorter lifespan. So needs to be self-replicating/repairing/recycling.

Not to mention… how do you repair it when components fail, especially sensitive electronics against cosmic radiation

A former NASA engineer with a PhD in space electronics who later worked at Google for 10 years wrote an article about why datacenters in space are very technically challenging: https://taranis.ie/datacenters-in-space-are-a-terrible-horri... I don't have any specialized knowledge of the physics but I saw an article suggesting the real reason for the push to build them in space is to hedge against political pushback preventing construction on Earth. I can't find the original article but here is one about datacenter pushback: https://www.bloomberg.com/opinion/articles/2025-08-20/ai-and... But even if political pushback on Earth is the real reason, it still seems datacenters in space are extremely technically challenging/impossible to build.

We don’t even have a habitable structure in space when the ISS falls, there is no world in which space datacenters are a thing in the next 10, I’d argue even 30 years. People really need to ground themselves in reality. Edit: okay Tiangong - but that is not a data center.

Good point. Still a long, long way from data centers.

We have 15,000 satellites in orbit that are almost literally the exact same premise currently being proposed - a computer with solar panels attached. We've being doing exactly this for decades.

I don't think any of the companies that say they are working on space data centers intend them to be habitable.

> We don’t even have a habitable structure in space Silicon is way more forgiving than biology. This isn’t an argument for this proposal. But there is no technical connection between humans in space and data centers other than launch-cost synergies.

I don't know, 10 years seems reasonable for development. There's not that much new technology that needs to be developed. Cooling and communications would just require minor changes to existing designs. Other systems may be able to be lifted wholesale with minimal integration. I think if there were obstacles to building data centers on the ground then we might see them in orbit within the next ten years. I don't see those obstacles appearing though.

> A single AI data center server rack takes up the same energy load of 0.3 to 1 international space station. The ISS is powered by eight Solar Array Wings. Each wing weighs about 1,050kg. The station also has two radiator wings with three radiator orbital replacement units weighing about 1,100kg each. That's about 15,000 kg total so if the ISS can power three racks, that's 5,000kg of payload per rack not including the rack or any other support structure, shielding, heat distribution like heat pipes, and so on. Assuming a Falcon Heavy with 60,000 kg payload, that's 12 racks launched for about $100 million. That's basically tripling or quadrupling (at least) the cost of each rack, assuming that's the only extra cost and there's zero maintenance.

> A former NASA engineer with a PhD in space electronics who later worked at Google for 10 years wrote an article about why datacenters in space are very technically challenging It's curious that we live in a world in which I think the majority of people somehow think this ISN'T complicated. Like, have we long since reached the point where technology is suitably advanced to average people that it seems like magic, where people can almost literally propose companies that just "conjure magic" and the average person thinks that's reasonable?

It's just the thought process that comes with shallow understanding: "I can buy a server" "We can put things in space" "What do you mean I can't get a server in space?!"

"Technically challenging", a nice way to say "impossible"

Great, so you seem to agree the technology exists for this and it is a matter of deploying more of it?