Forgot your password?
Close
typodupeerror
Space Google

Google Tests AI Data Center In Space (npr.org) 118

Posted by BeauHD from the data-center-in-the-sky dept.
Longtime Slashdot reader Geoffrey.landis shares a report from NPR: Google just put a refrigerator-sized satellite into space, part of a research project the company hopes can pave the way for orbiting AI data centers powered by the sun. The prototype satellite has four specialty chips in it called Tensor Processing Units, or TPUs, that Google designed for machine learning and has already deployed in data centers on Earth. The chips will run a version of the company's Gemma AI model for 15 minutes at a time due to heat management constraints, using the open weight model to answer simple queries.

In a blog post, Google said the mission will gauge how the TPUs handle "the physical stress of spaceflight and the radiation and thermal extremes of space." The satellite launched October 1 is a prototype that Travis Beals, senior director and lead of Project Suncatcher, called "a very minimal test" to make sure the chips can run in space. A SpaceX rocket carried it into orbit, and Planet will get it up and running. Beal said Google will then fire up and test the TPUs. The goal is for the mission to be operational for a year.

Google Tests AI Data Center In Space More | Reply

Google Tests AI Data Center In Space

Comments Filter:

  • for space... then we have zero problem using them here on earth. Space is cold but is also a vacuum so you lose all of your ability to convectively cool and are reliant on radiative cooling alone.

    • Re: (Score:2)

      by AmiMoJo ( 196126 )

      The problem on Earth is mostly NIMBYs, and also a bit the time it takes to set up new electricity supplies.

      It looks like the cost to orbit is going to decrease substantially, and solar up there can cover all the power needs 24/7, and so far Space NIMBYs have not had much effect on those mega constellations.

      • Re:If they are thermally efficient enough (Score:5, Informative)

        by gweihir ( 88907 ) on Saturday October 03, 2026 @06:11AM (#66360918)

        Cooling, communication, debris, cost of shipping to orbit, cost of station-keeping, power-generation (no, solar in space is not magic), cosmic radiation, needed size of the satellites, very easy to destroy, etc., etc.. The list of why this is a REALLY stupid idea is endless.

        There is not a single aspect of this were the idea is actually good. All it does is tickle the imagination of the clueless. And that is exactly why it gets pushed.

        • Re: (Score:2)

          by dfghjk ( 711126 )

          "There is not a single aspect of this were the idea is actually good."

          Well, it's good for convincing morons like Marc Andreessen that Elon Musk is a visionary. That's all it's about, the actual tactics are to simply take whatever they want and the country can FO. Of course it's a terrible idea, but investors don't know that.

          • Re: (Score:2)

            by gweihir ( 88907 )

            Of course it's a terrible idea, but investors don't know that.

            What I do not quite get in this context is that there are still investors around with the fundamental lack of insight these people constantly demonstrate. Is this a case of "there is a sucker born every minute"?

            • What I do not quite get in this context is that there are still investors around with the fundamental lack of insight

              Well, you know, D-K... they don't know what survivor bias is.

              Is this a case of "there is a sucker born every minute"?

              They are sure they are smart because they got money, and don't understand that even if they did make an intelligent investment, there was a luck component.

              • Not necessarily... I bet a lot of those investors just want their name on a satellite, or think they'll make tons from the companies stocks when the orbital DC idea works.

                Sure, massive solar panels can give the DC enough power, at least until a meteorite smacks it. Sure, it's cold out there, but exchanging heat gets a _bit_ more difficult in a vacuum.

                • I bet a lot of those investors just want their name on a satellite

                  This seems essentially sane if they have money to burn...

                  or think they'll make tons from the companies stocks when the orbital DC idea works.

                  They read too much sci-fi, not enough science fact, or both.

                  • I do agree.

                    I have to wonder what the AI company that "owns satellite 4850" is gonna do when a newer CPU or NPU comes out.
                    Bet the latency is bad enough that it'd be useless.

                    • Re: (Score:2)

                      by gweihir ( 88907 )

                      Well, at this time we do not even have CPUs of reasonable power that can survive in orbit for a longer time. AFAIK, radiation-hardened 386-class chips is the best thing available, i.e. about tech-level 30 years back or so.

                    • Latency is about a 1/10th of a second if it is a geo stationary satellite, and much lower if it is in a lower orbit.
                      I suggest to put a half assed complicated request into what ever AI you are using and cout: 21, 22, 23 until the first result pops up.

                      An additional 1/10th of a second, you won't notice.

                    • It's more complicated than that. The plan is to put them in sun synchronous orbits [wikipedia.org] so the solar panels always have light on them. Unfortunately, that's an orbit that passes over every point on earth.

                      This isn't like a Starlink satellite that just acts as a network connection. For those, you can use whatever satellite happens to be passing overhead. An orbital data center satellite is unique. If you start a calculation running on one satellite, you have to keep communicating with that satellite. When it

                    • I guess the plan is to let it communicate with the starlink satellites.

                      Thanx for the info, I did not really dig into the space data center idea.

                      I guess as soon as you have enough up, your request will go to one that is close and not on the other side of the planet.

                    • I guess as soon as you have enough up, your request will go to one that is close and not on the other side of the planet.

                      Yes, but that's only useful for short jobs. A satellite that's overhead now will be on the other side of the planet in about 45 minutes. Anything that runs for a long time has to deal with high latency and breaks in connectivity.

                      And if you only send jobs to nearby satellites, that creates other problems. The earth's surface is 2/3 ocean, and large parts of the rest are mostly uninhabited (Antarctica, Siberia, etc.). Your satellites will spend most of their time over uninhabited places, meaning they spen

                    • Actually, scratch what I just said because the situation is much worse. A sun synchronous satellite only passes over at sunrise and sunset. If it's noon where you are, the nearest satellite is 90 degrees of longitude away from you. Sending your jobs to a nearby satellite just isn't an option.

                    • No idea what you want to say.

                      This satellite is just one.

                      Here is how the orbits look like: https://en.wikipedia.org/wiki/... [wikipedia.org]

                • Meteor smacks are rare.

                  The oldest currently operating satellite that can still transmit a communication signal is AMSAT-OSCAR 7 (AO-7).
                  While it is an amateur radio satellite rather than a commercial television broadcasting satellite, it remains the absolute oldest active piece of communications hardware in space.

                  That satellite is only a small amateur radio "relay", it is operational since 52 years, on solar power.

              • Re: (Score:2)

                by gweihir ( 88907 )

                They are sure they are smart because they got money, and don't understand that even if they did make an intelligent investment, there was a luck component.

                Case in point: The fascinatingly incompetent Elon Musk.

        • Re: (Score:2)

          by AmiMoJo ( 196126 )

          Scott Manley did a video about this, and it's not quite so half baked.

        • Well if people are successfully blocking data center construction, their objections are a 'factual' problem for big tech. But people are just going to switch to complaining about blocking out the stars.

          • Re: (Score:2)

            by Junta ( 36770 )

            Finding a single place willing to put up with a datacenter is as much success as deploying thousands of these hypothetical space datacenters.

            For the cost of a *single* SpaceX launch, they could basically bribe an entire small town's population to be happy with them. $50K a person and they'll *probably* pass public comment.

            • Re: (Score:2)

              by gweihir ( 88907 )

              A central problem is that these "high energy density" datacenters guzzle drinking water like crazy. And where do you find drinking water in substantive quantity? Bingo, in places where quite a few people live. $50k a person will only do it if that is enough for them to move to some other place.

        • If one spends a lot of money on something, it is kind of save to assume he has done his math.

          A single star link satellite for example makes Space X 450million earnings.

          • If one spends a lot of money on something, it is kind of save to assume he has done his math.

            No, it's reasonable. It's not safe.

            The other part of that is that you're not necessarily doing the same equation, they may be solving for something something different.

        • There is not a single aspect of this were the idea is actually good. All it does is tickle the imagination of the clueless. And that is exactly why it gets pushed.

          In space, the enraged proletariat will not be able to smash the satellite with pitchforks and rocks. That is why compute in space is so desirable: YOU can't stop it.

          AI will be your personal government enforced monitor. Say, "hi" to Big Brother, in space.

      • Re: (Score:2)

        by dfghjk ( 711126 )

        "It looks like the cost to orbit is going to decrease substantially..."
        Does it though? As long as you ignore the costs that won't.

        "...and solar up there can cover all the power needs 24/7..."
        Definitionally, as solar would be what is available. And if you can't cool it you can't power it, so power isn't really a constraint.

        "...and so far Space NIMBYs have not had much effect on those mega constellations."
        What mega constellations? You are better than that. And the government is already talking about China

      • Re: (Score:3)

        by MrKaos ( 858439 )

        solar up there can cover all the power needs 24/7, and so far Space NIMBYs have not had much effect on those mega constellations.

        By far, with all these new mega constellations the biggest risk we face is from the Kessler syndrome [wikipedia.org] from which there will be no way back and end any ambitions for human space exploration for thousands of years.

        It would be worthwhile to have tight international regulation as it is the one interest that all countries share.

        • You do know that even the Kessler syndrom is something that can be fixed with progress in new materials and progress in understanding. It won't be a problem for long.

          • Re: (Score:2)

            by MrKaos ( 858439 )

            You do know that even the Kessler syndrom is something that can be fixed with progress in new materials and progress in understanding.

            How does materials change orbital phenomenon?

      • Re: (Score:3, Insightful)

        by drinkypoo ( 153816 )

        The problem on Earth is mostly NIMBYs

        NIMBYs aren't people who don't want things to be made. They're people who want them made somewhere else. But only the wealthy are clamoring for all of us to be replaced with computers, so that's not apt.

        • Re: (Score:2)

          by AmiMoJo ( 196126 )

          I suppose to be fair when it comes to data centres the NIMBYs are probably right.

          • If we had reasonable regulations then a data center could go anywhere and not be a problem, but since we don't, they can only go into unpopulated places with lots of water and power without causing harm. Normal DCs can just go into ordinary commercial zones because they use, while elevated, not completely obscene amounts of power and therefore water. And to return to the topic, this illustrates why AI DCs don't make sense in space, because cooling is harder in space than it is here.

      • Re: (Score:2)

        by ceoyoyo ( 59147 )

        There's lots of desert where nobody really cares what you do, and even more ocean. We don't like building in those places because they're far away from where we want to use the services, difficult and expensive to access, and have issues with cooling, power supply, etc.

        But all those problems are worse in space, and you generally need to rebuild your entire infrastructure, from power plant to GPU, every five years or so.

        • Re: (Score:2)

          by AmiMoJo ( 196126 )

          These satellites would be in LEO and fall back to Earth after 5 years anyway. They are thinking like Starlink where they need to launch thousands a year to maintain the constellation. The server hardware will be obsolete by then anyway. What sucks is that then we can't buy it used for cheap, and the fact that all that crap is burning up in the atmosphere and releasing who knows what.

          If the cost to orbit is low enough, and there are no land taxes or electricity costs in space, and it solves the locality and

          • Re: (Score:3)

            by ceoyoyo ( 59147 )

            These satellites would be in LEO

            Sort of. They'd be in probably lowish sun synchronous dawn-dusk orbits, a narrow band following the terminator where the sun always shines.

            and fall back to Earth after 5 years anyway.

            Yeah. Thus "you generally need to rebuild your entire infrastructure... every five years or so."

            The server hardware will be obsolete by then anyway.

            The server hardware maybe. Not the power plant, (extensive) cooling hardware, infrastructure. These aren't GPUs floating in space, they're the entire

            • Re: (Score:2)

              by AmiMoJo ( 196126 )

              You might be right. I wonder about the bandwidth too. Starlink is already over subscribed and starting to enforce limits on its unlimited service.

              • Bandwidth with Starlink is a region, not an absolute.

              • Re: (Score:2)

                by ceoyoyo ( 59147 )

                Not just downlink, but the inter-satellite bandwidth and latency is also likely to be a problem. I saw an analysis of what it would take to build a big datacentre, and there the cooling really does become a problem. Not the radiators so much as the plumbing to move the heat around. So everyone is talking about lots of little satellites, but that means you're restricted to a handful of cards locally. It's going to restrict the kinds of jobs you can do when the next rack over is a thousand kilometres away.

    • for space... then we have zero problem using them here on earth. Space is cold but is also a vacuum so you lose all of your ability to convectively cool and are reliant on radiative cooling alone.

      I expect the AI work done in orbit will be much more specialized. Smaller scale and less cooling demands.
      There is radiative cooling in the sense of IR emissions. It more a lack of conduction, heat transfer through physical contact.

    • Re:If they are thermally efficient enough (Score:4, Informative)

      by pixelpusher220 ( 529617 ) on Saturday October 03, 2026 @01:51PM (#66361442)

      good video by a physicist on the physics

      https://www.youtube.com/watch?... [youtube.com]

  • The promise of "new space" startups like SpaceX is, that they will somehow be able to reduce the costs of bringing up something into space by multiple orders of magnitude. That's what they promise, if people stop believing that they can do that eventually, they will just implode.

    Bringing electronics into space has it's problems, but it's essentially proven technology. We have had space stations converting many kilowatts for decades now. We can do that, we know how to do that. We know where to get the energy

    • Re: (Score:3)

      by dvice ( 6309704 )

      I think that SpaceX is going to build the satellites on the Moon using robots and launch them using rails to minimize launch cost. In theory is sounds like a good plan.

      Personally, I think we will invent more efficient AI faster than the factory is up and running, but I have to admit that SpaceX makes the world more interesting so I have nothing against them.

      • Re: (Score:2, Insightful)

        by gweihir ( 88907 )

        Yes, in a few 100 years maybe. People tend to grossly overestimate what technology can do.

        • Re: (Score:2)

          by ceoyoyo ( 59147 )

          Technology isn't really the problem. We've got most or all the technology to do what the GP suggested. What we lack are engineering experience, infrastructure and good economic reasons to do it.

          Europe had all the technology required to build a casino in the Nevada desert in the 16th century, but it took four hundred years to get around to building the first one in Las Vegas.

    • The promise of "new space" startups like SpaceX is, that they will somehow be able to reduce the costs of bringing up something into space by multiple orders of magnitude.

      Like this? https://ourworldindata.org/gra... [ourworldindata.org]

      Well that's one order of magnitude down, depending on where you place the goalpost. Starship only needs to halve the per kg cost of Falcon Heavy to add another one, again, depending on where you put the goalpost. If the goalpost started with the space shuttle, for example, then without question, it's doable.

      The answer is that putting anything into orbit is prohibitively expensive.

      You sound just like this guy:

      https://x.com/Tazerface16 [x.com]

      TL;DR claims to be an aerospace engineer and, for years, even as recently as April this year, was very lo

      • Re: (Score:2)

        by Khyber ( 864651 )

        That's one part of the equation! Satellite bus cost is another part of the 'magnitudes of order' reduction equation. Where I work right now is addressing that.

    • The promise of "new space" startups like SpaceX is, that they will somehow be able to reduce the costs of bringing up something into space by multiple orders of magnitude.

      With respect to AI processing in space its more about local processing, local decision making, faster response times. It for much more focused and limited work. Not the high volume general purpose work a ground based data center would perform.

      • Yeah, but you could just put everything you'd put into a satellite into a small shed and plop that somewhere. There's extemely little to be gained by going into space for such things, particularly considering the effort.

  • What they've launched is a data center the same way David Hahn built a nuclear reactor in his shed. (If he was alive today, he could IPO his reactor startup and make billions.)

  • The will know this makes no sense with regards top Physics. But they are engaged in the same deception most LLM providers are: Pretending that "AI" will give us a golden age, so they can reap insane profits before it all comes crashing down.

    • Can we launch stuff into space? Yes
      Can we use solar panels in space? Yes
      Can we run processors in space? Yes
      Can we cool processors in space? Yes
      Can we send and receive data in space? Yes
      What exactly do think the problem is that Physics says makes no sense?

      Finance, profitability, return on investment, depreciation are the things you want to look at

  • ...disk space, the new frontier!

  • And news for nerds. So of course /. is ... (looks around) ... excited?

  • This is exciting research. It's going to turn a lot of theory into fact. I find it fascinating.

    It's humorous to read all these comments talking about how stupid the very idea of space data centers is. So many Slashdot experts in data centers, AI, and orbital space. Who knew? But despite the cocksure naysayers, the people who are actually in the business of data centers, AI, and space, think it can be done and can be cost effective. This experiment will eliminate much of the guesswork.

    If it turns out that t

    • Re:Interesting (Score:4, Insightful)

      by dfghjk ( 711126 ) on Saturday October 03, 2026 @07:19AM (#66360984)

      This plan is predicated on people as stupid as you, but with lots of money. That exists, so this con exists. We're all impressed, though, with your willingness to tell us how stupid you are.

    • the people who are actually in the business of data centers, AI, and space, think it can be done and can be cost effective

      That's what they say.

      Do you believe everything people say?

      If it turns out that they can do it

      They can't, and only people who know how nothing whatsoever works thinks they can.

      • We've been putting evermore powerful computing into space for 65(?) years. This is just an AI specific computer.

        It seems to me that it has been proven that it is certainly possible to put a data center, lots of computers together, into orbit. It's just a matter, or question, of scale.

        It is odd how determined you and the other commenter are to refuse the reality of its possibility. But, your combined derision... that's a powerful argument. It has me very nearly convinced that you're right and I'm a stupid kn

        • It's just a matter, or question, of scale.

          If you can recognize that, then you should be able to do the math and figure out why it won't work. A quick napkin-based computation is sufficient. If I can do it, you can do it.

  • There's a very different reason for this (Score:3)

    by Arrogant-Bastard ( 141720 ) on Saturday October 03, 2026 @08:28AM (#66361050)
    Nobody can march down the street and burn them down.

    The fascist oligarchs behind the AI/data center scam know that the public hates these and that eventually, it will occur to someone to torch one of them. That's not going to be so easy if they're in orbit.

    This has nothing to do with research or innovation or anything else: it's pure self-preservation.

    • Re: (Score:2)

      by PPH ( 736903 )

      The reason some people hate data centers is the increased electrical demand that they produce and the resulting draw on limited resources and price increases that may result.

      Data centers in space represent no such risk.

      Leave it to the crazies who hate something, want to burn it down, but don't understand why. Useful idiots.

    • yeah, but it's easy enough to jam their communications.. might cost you a penny... but you can render them useless... and you can always burn down the control stations.

      Also, the data centers aren't so easy to torch, same with control stations/com stations... mainly concrete.. hardened perimiter, usually a small onsite security force... first one or two might be easier.. then they'll be on guard and even more secure... better off with an EMP, or substation disrupution... theoretically... but then you'd like

    • This is the first justification for this I've seen that makes sense. The "free power all the time" is such utter bullshit when the cost will probably be at least 10x what it would cost to build a solar array on Earth that can produce 10x the needed power and push it to batteries so it can run 24/7.

    • Nobody can march down the street and burn them down.

      Aha! Someone else who understands.

  • Shoot it out of the sky and let whoever finds the wreckage keep the scrap.

    • free loot (Score:1)

      by Anonymous Coward
      Like hitting the piñata in the sky and collecting all the ram that spills out.

  • This is a stock price ploy (Score:5, Insightful)

    by memory_register ( 6248354 ) on Saturday October 03, 2026 @09:53AM (#66361130)

    Google knows full well this makes no sense, but Elon Musk said we would put data centers in space so everyone has to act like they will do it too. Heaven forbid Wall Street catches on early to the data center debt bomb.

    • When I first read the post I thought 'Google wants to demonstrate to Wall Street and potential investors that data centers in space don't work. You should invest in us!

  • "Test it on local hardware."

Slashdot Top Deals

"Don't try to outweird me, three-eyes. I get stranger things than you free with my breakfast cereal." - Zaphod Beeblebrox in "Hithiker's Guide to the Galaxy"

Close