Thread

Google’s Project Suncatcher to put ML infrastructure in space

233 points · 553 comments · xnx

fwlr · 1790297624 · focus · HN ↗
Yes, the physics are worse, and yes, the economics are worse, but data centers in space have the crucial advantage that they are out of range of the molotov-throwing arm of Joe Public (recently unemployed).
1. thelastgallon · 1790304034 · focus · HN ↗
  Businesses run anywhere real money can be made. Middle east is the most volatile place. Yet, the businesses are perfectly okay building infrastructure to extract, refine and ship oil from there.
  If real money can be made, there is no need of datacenters in space. Offer some country some money and they'll let you build whatever you want. The whole idea is entirely unprofitable thats why they are making up things that won't work economically. Keep giving us money indefinitely, we'll keep spending, money can be made after a few decades of spending ... trust us.
  1. throwaway473825 · 1790304698 · focus · HN ↗
    Note that you didn't mention any such country, and for good reason - there is none.
    You'd need a dictatorship where people's voices don't matter, but then there's no right of ownership either.
    thelastgallon · 1790307189 · focus · HN ↗
    > You'd need a dictatorship where people's voices don't matter, but then there's no right of ownership either.
    And there are plenty of those countries. As well as countries that aren't dictatorships. Every leader everywhere is available for sale. In democracies, usually for a small sum. Dictatorships ask for a percentage of profit.
    Everyone knows how the current administration got to power: <a href="https://edition.cnn.com/2025/02/01/politics/elon-musk-2024-election-spending-millions" rel="nofollow">https://edition.cnn.com/2025/02/01/politics/elon-musk-2024-e...
    Trump will gladly allow datacenters to be built on federal lands. Just like he allows Oil/Coal and bans Wind and Solar. The only reason there is new story of Space Datacenters is because even if they are given everything by the Govt, it simply won't work.
    If you see a datacenter in orbit, it means something went wrong on Earth: <a href="https://systemsthinkingcollection.substack.com/p/space-datacenters" rel="nofollow">https://systemsthinkingcollection.substack.com/p/space-datac...
    People believe all kinds of fiction and garbage, like going to Mars. (Why Terraform Mars When We Can Fix The Earth? <a href="https://www.youtube.com/shorts/8mhib97--n4" rel="nofollow">https://www.youtube.com/shorts/8mhib97--n4)
    The same argument holds for AI DCs. If you can build profitable AI datacenters in space, why can't you build profitable AI datacenters on Earth? The only reason is you can't build profitable AI datacenters on earth. Just give us a few more trillions and wait a decade or two, we promise ...
    jazzyjackson · 1790307693 · focus · HN ↗
    This idea that space is a way to get out of the reach of NIMBYs is so dumb.
    a: now you’re in everyone’s back yard, since satellites in orbit can be seen by the naked eye at twilight zipping across the sky, so now the whole earth hates you
    b: you are still launching from earth, which is a pretty significant target if you’re worried about ecoterrorism.
    joquarky · 1790311303 · focus · HN ↗
    I'm curious why a space data center has to orbit Earth. Why not a solar orbit? Batched work can be done anywhere if you can tolerate the latency.
    Dylan16807 · 1790311758 · focus · HN ↗
    It can go further if you want it to be even more expensive, sure.
    hdgvhicv · 1790315819 · focus · HN ↗
    If you’re putting it in space you don’t care about the cost
    cindyllm · 1790316172 · focus · HN ↗
    [dead]
    podocarp · 1790323376 · focus · HN ↗
    Send it to Titan and cool it in the methane lakes
    hnbad · 1790328697 · focus · HN ↗
    There are significant differences in feasibility and cost between putting something in low Earth orbit (or even very low orbit like Starlink) and putting it in any other orbit around anything else - like easily 2-5x depending on where you're going to put it. I'll leave it to the Kerbal Space Program veterans to explain the mechanics that also make it much more difficult to actually launch and deploy successfully unless you're happy with just having the DC arbitrarily orbit around the sun (and drift year-by-year).
    Once you move outside low Earth orbit (i.e. outside the magnetosphere) you also have to deal with much more radiation so using regular commercially available hardware for the DC is no longer an option so you're looking at another 10-100x increase in cost for the chips alone (and at worse performance because radiation-hardened chips are literally generations behind at this point). You'll also need signficantly more (very heavy) shielding and that also adds to your launch costs. Of course OceanGate is a good reminder that billionaires don't always concern themselves with such trivialities as using equipment and materials actually suited for the task if it means they get an opportunity to brag about how they're disrupting the market by jerry-rigging mission critical components.
    Speaking in terms of feasibility again, availability isn't the only problem you'd be facing. Obviously latency is going to be horrendous because of the incredible distances involved but really there's just literally no infrastructure you can piggyback off of. You could potentially achieve data rates of up to tens of Gbit/s even from L4/L5 (stable orbits 1 AU away) using something like NASA's DCOS but this requires optical ground stations or LEO relays, bringing the weakest link much more within reach again.
    Even if you somehow manage to develop sufficiently ruggedized variants of reasonably modern chips, add the necessary shielding and develop and build out the necessary communication infrastructure from scratch (remember that we haven't even considered how we generate enough power to run the actual DC itself) and are also willing to accept the disgusting latency and packet loss ratio, the elephant in the room remains that DCs on the ground require constant maintenance to operate and replace failing parts. DCs in space can only ever be disposable - once any redundancies run out, they're just very expensive space junk.
    But you have a point: putting it in space is inherently ridiculously expensive. Getting it to L4/L5 is only insignificantly more expensive than getting it to L1/L2 but assuming you can fit the entire DC in a single Falcon Heavy (so you don't also have to worry about assembling the entire thing in space across multiple launches) you're limited to a payload of 25 t and the launch itself already costs you $120 million. Of those 25 t only 7 t or so can be spent on actual compute so your space DC would probably be constrained to the equivalent of about 500 GPUs (assuming this is for AI because everything is for AI these days). Most of the weight budget and the actual cost however would be eaten by all the structural and infrastructural parts required - in terms of cost for compute a space DC that fits in a single Falcon Heavy is easily an order of magnitude more expensive than one built in the US today and those don't need to be replaced from scratch when a significant chunk of the initial parts have degraded.
    TL;DR: Building a DC in space is easily 10x as expensive, requires technology that hasn't been developed yet, requires you to also build all infrastructure needed to actually operate and connect to it, and in the end just gives you a high latency low performance Beowulf cluster that is rapidly converting itself into an expensive piece of space junk if random debris doesn't kill it first.
    NetMageSCW · 1790369804 · focus · HN ↗
    You based all your objections on a faulty premise. Orbital data centers are all designed to be distributed across hundreds or thousands of satellites, not one monolithic satellite.
    hnbad · 1790544386 · focus · HN ↗
    I was responding to the suggestion to put them further away than low Earth orbit. Using "hundreds or thousands of satellites" would literally only make every problem I described worse, not easier, so I assume you missed that part and want to argue the merits of running what's essentially a botnet in low Earth orbit and calling that a "data center". Fine by me although I'm not sure why you decided to respond to me in particular.
    You can run compute in low Earth orbit but compute runs hot. We're not talking 500 GPUs a pop but even for an 8 GPU node you need some 40 m^2 surface just for thermal dissipation. I assume by "distributed" you also imply that they're not going to be in sunlight 24/7 and operate more like Starlink so they'll need batteries to keep running for a third or so of their lifetime.
    Low Earth orbit also means you don't have to worry about radiation frying your off-the-shelf hardware without the heavy shielding but you'll still have lots of random bit flipping to deal with and storage is largely going to be unreliable for anything but short-term, especially on SSDs.
    Distributed also means you replace local speeds with network speeds - Google was only able to achieve feasible "data center" data rates by having satellites fly within less than a kilometer of each other (i.e. densely packed in space terms, no longer really "distributed" in terms of proximity) and even that had orders of magnitudes worse latency (obviously) than you'd have in a data center pod (though not quite as bad as the multi-millisecond latency you'd have at current Starlink level distances).
    "Orbital data centers", even if operated as fleets of disposable low-performance pods instead of singular "out of reach" disposable mini DCs, are worse than real data centers on every single metric: they're slower, more expensive, unrepairable, degrade more rapidly and require more redundancy to compensate for random errors.
    There are pretty much just three use cases:
    - Providing space-based processing for space-based data collection to reduce the amount of data that needs be sent to ground (realistically speaking in the current climate this pretty much only describes surveillance).
    - Avoiding regulatory oversight, permit processes and most risks (e.g. earthquakes, missile strikes, protests, sabotage) associated with building and operating a real data center.
    - Avoiding energy price increases that might make operating a real data center more expensive compared to the fixed cost of continuing to operate an already launched satellite.
    I'm surprised nobody seems to have pointed this out but one of the biggest factors in the premises that would make this entire idea somewhat economically viable (assuming energy prices increase in ways that affect data centers without also massovely affecting the supply chain for launching things into space) is Musk's claim that they not only can make the upper stages reusable and recover them but also that inspecting, refurbishing and reusing them will be considerably cheaper than replacing them. If they can't make launches significantly cheaper still (not that they aren't already ridiculously cheap all things considered), using a fleet "designed to be distributed across hundreds or thousands of satellites" would be economically unviable compared to just stuffing a single satellite to the brim (especially if you can avoid the battery).
    But then the single advantage becomes "it's in space" and that is really only interesting for compute running on data already collected in space - and if that is a problem big enough to be able to sustain an entire business model justifying SpaceX's valuation... well, let's just say that would be the kind of world where if it were a movie the major plot point would be the destruction of that company or defeating the regime that enables it.
    stymaar · 1790315254 · focus · HN ↗
    > b: you are still launching from earth, which is a pretty significant target if you’re worried about ecoterrorism.
    This so much. Good luck destroying a DC on earth without a very strong force. On the other hand, destroying a rocket on its launchpad just take one FPV drone.
    defrost · 1790315754 · focus · HN ↗
    > Good luck destroying a DC on earth without a very strong force.
    Leaving aside the DCs already destroyed in the Iran totally not a War by an enemy that can be easily crushed at any moment ...
    DCs in mainland US aren't well defended - a pattern of shrapnel heads landed by a mobile mortar or two with a 2 km radius would render most in need of some upkeep.
    stymaar · 1790330095 · focus · HN ↗
    Sure, but in my view any organized actor with military hardware in their hands count as a “very strong force” when compared to the average Joe with a Molotov cocktail. And Iran, as a state actor with ballistic capabilities is obviously one as well.
    defrost · 1790389342 · focus · HN ↗
    A couple of rednecks in a pickup with a mortar is far from "a very strong force"; consider how you would upgrade their description if they additionally had three helicopters, now describe them if additionally their numbers doubled, now throw in a couple of tanks.
    When you start describing rabble level opposition that high you run out of ceiling very quickly.
    The US is filled with ex service personal who can whip together a mortar, as "military hardware" goes it's on par with what civilian firework techs have access to and can build in a garden shed, it's home grown resistance level guerilla tech.
    If there's growing ground swell opposition to US data centres then the same Y'All-Qaeda militia plotting against dams and power substations will feel empowered to have a shot at a DC or too.
    stymaar · 1790410322 · focus · HN ↗
    If you consider a full fledged insurgency in a civil war setting, you're very far from “one angry dude throwing a molotov cocktail at Sam Altman's DC” that was the original point.
    I don't think the US have the social fabric needed for an insurgency to develop on though.
    defrost · 1790486455 · focus · HN ↗
    > If you consider a full fledged insurgency in a civil war setting
    Who said anything about civil war?
    The US has many splinter loon groups who blow up buildings and like to attack infrastructure (see: US news). This has happened over the past decades with no civil war.
    Data centres are soft targets - softer than banks. Do you remember when armed bank robberies were all the rage in the 1970s and 1980s? Not a time of civil war, just a time when a couple of people took control of buildings while heavily armed. Such a small group with fragmentation grenades can easily render a data centre unfit for use for several months with relative ease and fast ingress and egress.
    stymaar · 1790537107 · focus · HN ↗
    Armed robberies are done with profit motives by professionalized members of the organized crime. By definition these people have access to the kind of organization and weapons you need to mount such attacks.
    Politically motivated armed attacks are a totally different thing.
    Also, I think you massively underestimate the sheer size of an IA data center. If you want to destroy one with grenades you're going to need a literal truck load of grenades.
    defrost · 1790552295 · focus · HN ↗
    > Armed robberies are done with profit motives by professionalized members of the organized crime. By definition these people have access to the kind of organization and weapons you need to mount such attacks.
    Really? Maybe read up some more, OC scale mafia groups weren't doing many of the bank raids of the US 1970s / 1980s - in any case these are still well below my bar on what constitutes a "very strong force".
    > Also, I think you massively underestimate the sheer size of an IA data center.
    You think wrong.
    > If you want to destroy one with grenades you're going to need a literal truck load of grenades.
    Exactly right, or rather several drums of farm grade fertiliser, diesel, ball bearings, a stolen truck, and somebody (or two) prepared to do a ram raid. Groomed idealogues or addicts with debt have all been used for such things.
    Or, as already mentioned, drive parallel and mortar the centre from one end to the other, achievable by small discontent militia types (who abound in the US).
    stymaar · 1791135949 · focus · HN ↗
    > Exactly right, or rather several drums of farm grade fertiliser, diesel, ball bearings, a stolen truck, and somebody (or two) prepared to do a ram raid. […] Serious question time: Are you honestly unaware of Timothy McVeigh?
    Again a great illustration that you have a scale issue in your reasoning, even if you managed to do as much damage as McVeigh you'd be very far from destroying a DC. If you're lucky you'd have 10% of the GPUs being destroyed and only one half of the DC being able to restart working by the next Monday, and if you aren't your trucks blows off outside of the building and only the racks closest to the wall are damaged and the DC doesn't even stop operating on that day.
    user43928 · 1790320890 · focus · HN ↗
    Have you considered that the 'NIMBY' problem here might be about permits and local resource usage rather than terrorism, which makes about no sense?
    XorNot · 1790322609 · focus · HN ↗
    Permits are not costing as much as a space launch and all the extra costs associated with that environment.
    If it's too expensive to outsource to another country, it's definitely too expensive to put in space instead.
    user43928 · 1790323931 · focus · HN ↗
    Cost is one thing, the other is whether it will be permitted at all.
    Then there's the power grid problem.
    I understand expanding the grid takes a long time, thus the gas turbines. Which I understand are also supply-constrained, and environmentalists are not happy either.
    I'm no scientist or financial analyst, but I could imagine that the space thing would be viable especially in a scenario where AI capabilities continue to grow.
    If AI really does replace not only most knowledge work but also other labor via robotics, and this requires enormous compute, we are going to want to scale this up at a much faster rate.
    pferde · 1790330459 · focus · HN ↗
    Are we, though? Going to want to scale this up, I mean.
    Unless economic systems around the world change significantly very soon, this AI replacement you so cavalierly mention would cause significant portion of human population to reach poverty, causing much unrest and mass deaths.
    user43928 · 1790331301 · focus · HN ↗
    No one wants mass poverty, unrest, and deaths.
    I don't buy into demonizing the rich, including Altman, Amodei, Musk and co.
    I am not concerned that this will be the outcome of the productivity surge.
    We will have enough for everyone, and ensuring comfortable living standards for the general population is going to be in the interest of those in power, even if you assume very selfish motivations.
    I'd also like to point out that both Anthropic and OpenAI created frameworks to monitor labor market impact and propose responses.
    notahacker · 1790326357 · focus · HN ↗
    Launch requires permits too, you just don't get to jurisdiction shop...
    user43928 · 1790328736 · focus · HN ↗
    When I looked it up earlier, it seemed to me that to scale this up we would require larger rockets than Starship if we cannot launch hundreds of times per day.
    Admittedly I know very little about spaceships
    thelastgallon · 1790309710 · focus · HN ↗
    > Note that you didn't mention any such country, and for good reason - there is none.
    All flavors of Govt (authoritarian/democratic) are bending over backwards to build new DCs.
    In Ulanqab, China buildout of data centers 1000x size of Colossus: <a href="https://x.com/RnaudBertrand/status/2099319829589295197" rel="nofollow">https://x.com/RnaudBertrand/status/2099319829589295197
    India, a country which has neither power nor water or any measurable infrastructure is giving massive subsidies to hyperscalers to build AI DCs. They get free land (very very costly), nearly free energy, guaranteed supply of water and so on ...
    Examples of one authoritarian govt and one democratic govt. And there are 180+ other countries.
    sourcecodeplz · 1790314566 · focus · HN ↗
    just to add a few more reasons not mentioned on X why they chose this place:
    Cooling is a major factor. Ulanqab's average temperature is about 4.3°C, enabling free air cooling nearly 10 months a year and very low PUE around 1.1.
    Even more decisive is cheap, abundant power—largely wind and solar plus coal—at roughly half the price of eastern China. Add vast open land, dedicated low-latency fiber to Beijing (~2-5 ms),
    seanmcdirmid · 1790317009 · focus · HN ↗
    It actually abuts Hebei which puts it in striking distance of Beijing. I wonder about water though, it’s just grassland so there can’t be much to spare. It would be like building a data center in eastern Oregon.
    killingtime74 · 1790318633 · focus · HN ↗
    I thought water is for cooling. They dont need it if its naturally cooled.
    rcxdude · 1790322514 · focus · HN ↗
    Or use closed-loop cooling.
    seanmcdirmid · 1790350240 · focus · HN ↗
    They would need closed loop cooling for the months that weren’t cold enough, well, they would probably need it anyways just that it would be more efficient in the long bitter winter. The extra clean energy from wind and solar might make that reasonable though.
    azan_ · 1790323779 · focus · HN ↗
    DCs just don’t use that much water.
    p_l · 1790326983 · focus · HN ↗
    The enormous water usage values are generally about powerplants providing the power, and even the better reports on that include water use by hydroelectric plants - which makes sense in context of the report, but that context nor sense isn't understood by journalists and hype-men quoting just "trillions of gallons a year, more than X cities"
    khriss · 1790323938 · focus · HN ↗
    > one democratic govt
    Honestly, India might be more democratic than China, but the current govt is known for it's authoritarian bent [1] and crony capitalism [2] in particular.
    While governments of all types rushed to embrace AI data centers, perhaps driven by fears of being left out, the pushback against data centers seems to be universal. The difference is, countries with less authoritarian govenrments are seeing those pushbacks be generally more successful.
    [1] The Chief Election Commissioner was recently caught manipulating voter rolls over the written objections of the other two election commissioners. <a href="https://www.bbc.com/news/articles/cqkgw68yjg29o" rel="nofollow">https://www.bbc.com/news/articles/cqkgw68yjg29o
    [2] The transport minister came under flak after pushing a rushed transition to higher ethanol fuel blends. It transpired that his sons just happened to have set up ethanol factories just before the policy was announced and their net worth went up by 5000%. <a href="https://www.bhaskarenglish.in/originals/news/whye20-fuel-put-nitin-gadkaris-sons-in-controversy-135843201.html" rel="nofollow">https://www.bhaskarenglish.in/originals/news/whye20-fuel-put...

Open on Hacker News to reply ↗

Unofficial Hacker News client; not affiliated with Y Combinator.