It really makes me appreciate how important good game design is. Claude is doing a fine job coding everything I describe, but it doesn't really understand fun, so I need to.
This is my experience too. I think people with no gamedev experience widely underestimate the challenge of good game design and the need for iteration with real people testing the game. In a similar way that an inexperienced game developer overestimates their skill to assess the fun and all the small details that matter, and will get demolished when it's first playtested by other people.
You can't launch from the US without the federal government giving you permission. And SpaceX does not have (nor are they planning) any launch pads outside the US. [And because of ITAR, I'm not even sure they'd be allowed to build one.]
However, I do think avoiding local control (state/city permits) is a reason for this.
As long as the company has a HQ and owners that care about their quality of life, they don’t really gain any extra safety by moving their hardware to space compared to keeping it where they are based.
For real. It's an enormous problem solved only with 1) sheer scale, and 2) Science Fiction.
Both of those are expensive as hell, by the way.
Cooling via radiation follows Stefan–Boltzmann: P = εσAT⁴. Let's assume a good surface (emissivity ~0.9) at 300 K (27 °C) at 400 W per square meter per side. A flat panel radiating from both faces into deep space gets 800 W/m, not including the losses from, say, the Sun, or from IR coming off the Earth. Now, input power. Sunlight in orbit ~1,360 W/m², assume ~22% cell efficiency, we got 300 W/m². So each 1 MW compute, 3,300 m² of solar panel and minimum 1,200–1,500 m² of radiator.
In case ya didn't know - 1 MW is tiny from a present-day-datacenter perspective. It's like 8 racks. So we're talking orbital megastructures here, many many many square kilometers, and this is with all the best case assumptions, and magic radiator panels that never see the sun, or the earth, or the moon.
This is just the basic numbers here, by the way. There's a garbage truck full of other unsolvable problems if you poke your head in there.
Aside from the "Avoid Regulations" aspect, and the "Everything That Burns Deorbiting is Depreciation" aka "The Starlink Trick", I'm not sure what the hell the draw is.
Cooling in space is hard but not impossible - while current (IMHO stupid without advanced in space infra) space data center projects work with a couple MW, many advanced space propulsion concepts might have to reject hundreds of MW if not a couple GW.
As for space data centers - I think the main issue is the complete lack of in space infrastructure for resource mining, processing and manufacturing & maintenance. It is kinda like building your first practical steam locomotive & the deciding to build directly an airliner. No suitable materils, experience, work force, material sources, etc.
We eventually went from locomotives to airliner, in an incremental manner & expanding the supporting infrastructure to support the ever more ambitious projects.
It might be eventually doable, as an experiment or as a flex, sure. But it's never going to come close to being cost-reasonable versus the equivalent infrastructure here on earth.
The security advantages are enormous since access to space is so tightly restricted and controlled, compared to the access potential of a land based data center. Only risk in space is maybe we start WWIII with china and the US directly trading blows. On the ground, any insurgent group can disable your infrastructure. Nothing is truly safe on the surface of the earth. Anyone can strap a bomb on a drone now. See examples from the currently active wars.
It costs at least 50x more to put a GPU in space than it does on Earth. For that price you can have dozens more capacity in bunkers, under the sea, or on remote islands. Do you think your insurgents are going to get all two dozen? They could travel to the far corners of the earth, destroying 22 of them, and you'd still be ahead. Further, I wouldn't be surprised if a satellite with such a monstrous solar and radiator footprint wouldn't be susceptible to a laser based attack from the ground; either frying it or pushing it into an unstable orbit by vaporizing a few bits.
It costs much more than 50 times because there are no GPUs in space yet. Google is only planning to have some sort of space data centers mid 2030s, if everything works out. A big if, but if they don't start now then we'll never know.
What? No, exactly the opposite. It's very easy to jam radio signals and much harder to cut wires. There's a reason the drones on the front lines in Ukraine are dragging fibre optic lines these days.
There's probably so many ways to get around that with space based technology. I can quickly imagine several methods. It depends on what the system is for which might be a good method to use.
Method 1: same as how u2 planes dumped their data: air drop physical media containing data and catch it in the air.
Method 2: laser based emission to specific detectors.
Method 3: baseball style communication: station is under observation and manipulates in some way to serve as a signalling language.
Method 4: numbers station
Method 5: bill yourself as an isp and have some coded syntax that can be supplied in plain sight with the rest of isp traffic.
Method 1 runs into the problem of how to replace that media. The DC is in space so it's not like with U2 planes where they landed. Also, I think you mean the keyhole satellites and not U2 planes, since a plane lands at a secure site and can offload media then.
Method 1: Sure datacenters with latency measured in several hours sure are useful, and can also be intercepted
Method 2: Can be jammed by drone with laserpointer.
Method 3: Let me just transfer gigabytes of data via physical signalling
Method 4: Can still be jammed
Method 5: Can in fact still be jammed?
If you’re receiving data optically from a space based laser, yes you are quite vulnerable to your detector being jammed by a much closer laser that doesn’t have to be nearly as powerful to add noise to the signal.
Why do you think GPS is so easy to spoof?
Actually I just googled (clauded?) to learn more about space lasers and found a paper describing interrupting quantum key distribution (tamper evident but not jam proof!) with a 1kw laser on the ground pointed at the satellite! So it really does not take many photons to fuck up an encrypted signal.
“ Vulnerability of Satellite Quantum Key Distribution to Disruption from Ground-Based Lasers”
What was it the 27th law that says radiating heat in to space does not work? I'm skeptical of the whole thing too, but it's not an impossible engineering challenge, just an expensive one.
According to Wikipedia this reduces weight and not the required area. Also AI said the same thing, but I can't trust in it this blindly. So, how smaller would be the required surface area?
No point in running them at room temperature. GPUs, etc. run fine at 95 C. If you run your cooling loop at 70 C instead, you get 70% more cooling compared to 27 C.
At any rate, 1 MW for a single satellite is fine. Just launch several thousand of those and you get to real numbers.
Also, there's no need to talk about "magic" radiators. You orient them so they're at a knife edge to both the Sun and Earth. This is not difficult (the Moon is irrelevant).
> Now, input power. Sunlight in orbit ~1,360 W/m², assume ~22% cell efficiency, we got 300 W/m². So each 1 MW compute, 3,300 m² of solar panel and minimum 1,200–1,500 m² of radiator.
We need > 2x more solar panels than we need radiators. Doesn't this imply radiation isn't really the limiting factor here?
Getting the energy back from the solar panel is easy via copper cables. Getting the heat back out there to the radiators is a bit harder, you needed fluids and pumps and heat exchangers which have lots of moving parts and need maintenance.
Doing it in a cost and weight effective way is still a big deal, because if it's not within ~10x the cost of ground based data centers, not enough people will use it to justify building it.
It’s obvious that radiation isn’t the limiting factor, since, as you note, there’s no reason a satellite shouldn’t be able to radiate at least as fast as it gets power from the sun. I don’t know how this became a meme.
Yeah, I suppose another way to look at it is that ALL satellites have to radiate the energy they get from the sun (since 99+% of electricity used by electronics gets radiated off as heat). A datacenter is no different. All that is different is that quantity of energy both in and out, the ratios are identical.
Radiating 1MW at 500K (227C) with a 0.4MW heat pump takes about 200 m^2 flat sheet surface. Inputs - solar+nuclear for double fun. So - quite feasible.
Moves 1 MW of heat with 0.4 MW of work? I.e. 2.5 COP {coefficient of performance). That's insane, and I mean that in a good way. Could you dig me up a cite for that?
That's thumping the Carnot limit: [[T_cold / (T_hot − T_cold)]].
2.5, while rejecting at 500 K, cold side's at least 357 K (eeehhhhhhh 84 °C) . . . and that's an absolutely perfect Carnot machine. At 50% Carnot -- a pretty good heat pump, real world performance is 40-60 -- cold side's at 417 K (144 °C). 417k, feeding your GPU coolant loops.
I believe that's Carnot COP for a heat pump used for heat+. I used the refrigeration version, T_cold / (T_hot - T_cold), which I'm 80 percent sure is the right one here.
Depends on which heat you want
Heat adding to hot side: COP_heat = Q_hot / W = T_hot / (T_hot − T_cold).
Heat leaving the cold side: COP_cool = Q_cold / W = T_cold / (T_hot − T_cold).
Another one (more common in the day to day, for me at least): heat-engine efficiency, η = 1 - T_cold / T_hot. Cycle forward to make work from heat.
Do we even _have_ semiconductors that can work at 220C? And if you're thinking about using some kind of refrigeration cycle, its efficiency is going to be bad.
1) The chips don't reach 220C. The 220C is the temperature at the hot end of the heat pump. The chips are on the cold end of the heat pump.
2) The International Space Station has used a dual-loop ammonia/water-based heat pump to cool the station temperatures. It's been in place for several decades. Heat pumps are a proven technology.
> The 220C is the temperature at the hot end of the heat pump. The chips are on the cold end of the heat pump.
If we want the heat pump's cold end at about 40–65°C, then for each 1MW of GPU heat, we need another 1MW of heat pump power. Now you need 2MW of solar power.
Good news is that the radiator at 227C (500K) can emit about 5× more heat per square meter than at 57C (330K)
Getting this all up into orbit it obviously the hard part, but if you're already building so much solar capacity the cooling actually doesn't seem unreasonable?
Great read, thanks for sharing. I am interested in reading some more about the other unsolvable problems that exist in this space, do you have any recommendations that you wouldn't mind pointing me at? It would be greatly appreciated, and thank you :)
Surface area is a materials problem? Folded microstructure, atomic-scale textured surface or some other science-fiction solution could have square kilometers of surface area in a shoebox.
Imagine you have two blackbody radiators with the same bulk properties, except one has surface area shenanigans like aerogels. In the far field as a whole, it seems like both should radiate essentially the same regardless of the internal details. You can shape emissive direction, or improve efficiency of non-ideal materials, but even ideal materials don't fix the issues pointed out by the parent.
Right; it's only area exposed to the exterior that counts. A physical object can't thermally radiate more power than a perfect blackbody spanning its convex hull.
(This follows because a physical object can't absorb more light than a perfect blackbody spanning its convex hull. A perfect blackbody by definition absorbs 100% of incident light, which is a hard upper bound. Any line incident on an object is also incident on its convex hull).
(Consider an isothermal object that emits more power than a blackbody in the shape of its hull at the same temperature. If you were to place that object in a closed system at thermal equilibrium, the interior of an insulated emissive sphere—combining assumptions, it would emit more power than it absorbs, in violation of the 2nd law. Starting from an isothermal system, the object would grow colder, and the enclosing container hotter).
I don't think so. Large surface area helps with convective cooling I think by increasing the surface area that participates in heat exchange with the air (or other thermally conducting material), radiative cooling wouldn't benefit from this because you can't concentrate light beyond the source that it's emitted from (etendue).
Though I do wonder if it would be possible to have some kind of internal heat pump driven by electrical power to juice up the temperature of the radiators to increase the power being radiated away? E.g., run a heat pump to increase the temperature of a working fluid and then run high temperature radiators? I think it would work and I don't immediately see that it would violate the laws of thermodynamics? (this is ignoring all practically, I'm sure the engineering would be devilishly hard, although if you're already shooting for the moon you might as well throw in some artificial gravity to boot, it's not like the robots get motion sickness)
the other is that you could make tiny datacenters and flood the sky with them. in effect, not datacenters at all, but some kind of dataswarm coordinating at literal lightspeed via lasers.
they'd still be wildly expensive to deploy, and probably litter the orbit zone with fast-moving debris.
Your "other" makes no sense. It doesn't matter if you make a few big or a lot smaller, in space you will still need the same space for the same amount of megawatt. Or did you miss the scale of parent's post ? Because in that dream scenario of "let's ignore all the issues except that" and "the earth and the sun don't have any impact", it's still 3 THOUSANDS square meters for a MW of 8 racks.
You want to go smaller and go one rack only sure, it's still hundreds of square meters. Check the size of current orbital structure for a point of reference, you can't dwarf those and call it a "dataswarm of tiny datacenters flooding the sky".
From the article itself, sounds like it’s an open problem that they are experimenting with:
“ We’re working on a number of different approaches for this, including a combination of heat pipes and radiators to cool the chips. So far, our team has tested the technology in a thermal vacuum chamber that simulates both the thermal and vacuum environment in space. We’ll see how our new TPU cooling system works in space and refine our designs as we learn more.”
Imagine a heat pump circuit fails (solar rays? micrometeorite? random chance), I guess everything in that loop will just be dead forever? I guess you don't need to pay to dispose of it.
It's a bit like the hyperloop. They know it's not practical but having the idea out there makes money.
There will be some niche demand for military use though.
> The biggest challenge was how to cool the A.I. chips, which heat up when they perform calculations and process information. Fans, which typically help dissipate the heat, do not work in space. So the Silicon Valley company instead developed a cooling system that uses layers of conductive material to expel the heat into space.
> The bottom layer is made up of Google’s A.I. chips, which sit on a green motherboard. The next layer consists of “thermal interface material,” a pale green putty that comes in sheets like Fruit Roll-Ups and connects the chips to layers of aluminum and copper, radiating heat away from the motherboard. Finally, there is a radiator panel, which projects heat into space.
> The chips can operate for about 15 minutes in space before needing to be shut down so they can cool off, said Travis Beals, Google’s senior director of product management for Project Suncatcher.
> The chips can operate for about 15 minutes in space before needing to be shut down so they can cool off, said Travis Beals, Google’s senior director of product management for Project Suncatcher.
We don't. This is all cover for the militarisation of space, there's no real benefit that'd be ever economical to put a DC up in space when you could build one on the ground. The whole narrative exists to allow google to tap into the Golden Dome / Space force bucket of pork that's basically SDI II.
You don't need civilian scale compute in space, but you absolutely can put miliary application up there and get filthy rich, like Elon, who doesn't care if it ever works, because he's getting paid as taxi service to shoot stuff up.
This is the most interesting perspective I've heard on this topic, which otherwise always converges on the same political dismissals or heat dissipation arguments (the latter are fascinating, but going in circles by now).
Sounds very probable: gives a plausible reasons for sending a lot of infra up, absolutely doesn't have to be profitable or even effective, puts the focus on AI instead of the MIC, and is in line with the kind of contracts Big Tech runs after.
To me this model explains a lot of why so many big companies seem to be investing into what every expert I've heard says goes against basic physics. The only thing missing for it to be more than an interesting idea is why China and the EU are not fighting this.
you need a gradient from hot to cold to generate electricity. vacuum is a poor thermal conductor, so your cold part will become hot and then no more gradient.
(I guess you could try to capture the radiative photons via the photoelectric effect?? but I don't think it works.)
I tried asking Grok and a bunch of waffle came out but one interesting idea was "modular tiles" with solar cells on one side and a panel to radiate heat on the other with the processor bolted to the panel. (https://www.geekwire.com/2026/sophia-space-caltech-ai-patent...)
He is free to launch his own space GPU if he is so confident it is profitable*.
I will take it as a given that the Google engineers know what they are doing, and their first version can only run for 15 minutes before it needs to shut down to cool.
*Profitability requiring that other Musk controlled companies do not pay for the service at elevated rates in a classic self dealing scheme.
> He is free to launch his own space GPU if he is so confident it is profitable*.
Are you not aware that's exactly what SpaceX is doing?? https://www.spacex.com/spacexai/starmind Sure, it's now SpaceXAI or whatever, but how is that any different than Google sending up sats and selling the compute via Gemini?
Please stop with this argument. This is one of the easiest problems to solve. Heat dissipation requires a radiator, which is just a dumb hunk of metal. All you need is a cheap way to launch mass to orbit, which is exactly what SpaceX (and Chinese companies) are doing.
The only valid argument against data centers in space is the economic cost. If the per megatoken price for space datacenters is higher than terrestrial, then this won't work.
But the price for terrestrial datacenters keeps going up and technology keeps dropping the price of space-based.
That is not a fake argument, but real physics. Yes, you can design out X kw can be dissipated by this much radiators, but that adds an enormous quantity of mass, more than the solar panels that feed it.
If you are trying to generate a profit, every extra kg puts you further in the red. It still costs over $1000/kg of mass to get into orbit.
So is it a physics problem or an economic problem? Sounds to me like you are acknowledging that it is just an economic problem. If it cost $1/kg to get to orbit then this wouldn't be a problem, right? Just make a bigger radiator.
But if it's an economic argument, then you need to do the actual math. How big is the radiator? How low could the price to orbit go? How much can you charge per million tokens in 2030?
The reason this is a fake argument is because the validity depends on the math, and nobody advancing the "you can't cool stuff in space" argument is actually doing the math.
The economics is constrained by physics. AI in space is not viable if it costs $1000/kg to launch to space. Starship promises to cut that down to $100/kg (more if you believe Elon, but most don't), but that's still not competitive.
Sure, things would be different if the cost was $1/kg, but short of somebody building a space elevator that's just not going to happen.
I won't believe less than $100/kg until I see it. I agree with you on that.
But are you sure that $100/kg is not competitive? I don't remember all the math, but even their initial AI1 design[1] would throw a lot of profit, if you can sell at the price they offered to Anthropic.
I suspect the price of manufacturing the satellite, plus chips, is the dominating factor, not necessarily the launch costs.
Okay any argument about why space is uniquely challenging is going to revolve around physics. Sure it's not literally physically impossible, but we need to explain to people why this is different from shipping the GPUs to Ohio.
If you want math then https://andrewmccalip.com/space-datacenters exists. The numbers are grim for orbital DC. Even if you drag the launch cost slider all the way to $1/kg (by the way this is literally sci-fi, per ChatGPT air freight of semiconductors from Taiwan to Ohio costs $9/kg and ocean/train freight costs a bit under $1/kg for a reasonable shipment so good luck with $1/kg to LEO this century) it is still more than twice as expensive as terrestrial DCs.
But that calculator shows most of the cost is in the satellite. At $8 per watt (or about $2 million per satellite) the cost of orbital compute matches terrestrial.
That sounds absolutely possible. But in any event, we're now arguing a different thing.
The only thing I'm arguing is that cooling is a solved problem. I don't know if SpaceX will ever get the price down so that it is economical. But I'm convinced that there is no universe in which they hit their foreheads and say, "Oh no, we forgot about cooling!"
Note that this calculator is actually quite optimistic for orbital wrt. many things including cooling and effect on launch, as:
> No additional mass for liquid cooling loop infrastructure; likely needed but not included
> Thermal: only solar array area used as radiator; no dedicated radiator mass assumed
In hardware and mfg. solvable vs. solved is a big difference. And I too believe that SpaceX's engineers know about radiator panels. But the more cynical interpretation is that whatever the SpaceX engineers think about the technical merits, they are not being asked for that. They are just being asked for a pretext that justifies the xAI acquisition. Elon is also discussing lunar satellite factories that launch the satellites via railgun. Now, is this physically impossible? No, that isn't physically impossible either and I will seriously defend the physical possibility of this. It's not going to happen though.
And you could spend all the engineering costs on building some seriously efficient terrestrial DCs, but somehow all these analyses start with "assume that launch and satellite technology advances manyfold and terrestrial DCs stagnate or become less efficient, then if you squint the two numbers get kinda close."
Certainly launch has to advance a lot for this to make sense. SpaceX is betting big on Starship getting close to $100/kg. But does satellite tech have to advance much at all to make this work? SpaceX has already successfully deployed one megaconstellation into orbit and it's extremely profitable. Isn't this just another Starlink with bigger solar panels and radiators?
Yes it has to advance a lot. In a DC all your supporting equipment has returns to scale and can be repaired if it breaks. If a GPU breaks, a sysadmin walks over to the offending rack and swaps the card. In space all that equipment serves just a few cards (this is more like orbital server racks) and it has to work in space (so instead of using an ~infinite heat sink like the Earth, you use radiators etc). We still don't exactly know what effect radiation in LEO would have on stock GPUs - IIRC the experiments to determine this started after Elon went all in on orbital DCs. If anything breaks, you have a flying brick.
This is why that calculator, even under extremely optimistic assumptions for orbital, and even if you assume launch is zero, still cannot make it competitive with terrestrial.
It's a fundamental physics problem. You need to have huge radiating surfaces.
A 1GW datacenter with chips operating at 100C (which is probably doable) will need a radiator that has a surface of one square _kilometer_, and this is with all the favorable assumptions. Realistically you'll need about 2x of that.
If you want your DCs to be on a 1000km orbit (for reasonable ping times), you'll be able to _resolve_ these satellites with a naked eye!
The current plan is for 250 kW peak with 160 m^2 radiators. It will weigh about 4 tons so you can pack 25 on one flight of Starship.
Deploy 4,000 and you're at 1 GW. That's 160 launches.
BTW: SpaceX has already manufactured and launched 10,000 Starlink satellites and Falcon 9 launches about 150 times per year. None of this seems unworkable.
Can we solve the fertilizer price problem by hauling it with airplanes? Yes, we can! It's easy, just load the potash fertilizer into an airplane and unload it directly into the traincars. I even designed a neat conveyor belt system to speed up unloading!
Look at the numbers. 1kW of nuclear power capacity on Earth is around $2000, and that's 24/7 guaranteed power. So a 250kW cluster needs $500000 to cover its power demand with near 100% reliability and with some ongoing cost.
And if we're OK with some interruptions, then we can use solar+wind at around $100000 and with essentially no ongoing cost. If we assume the absolutely best projected launch cost of $100 per kg (vs the current one of ~$800), that's just 1 ton of material in space!
So you're off by 2-3 orders of magnitude in cost. And this kind of "it's unprofitable" is actually a fundamental issue.
Specifically, if SpaceX can get the price of a satellite down to $8 per watt (about $2 million USD) then it will compete with terrestrial.
I just don't understand how you can be so certain that they can't do that. I'm not certain that they can, but being certain that it's impossible seems completely evidence-free.
This calcualtor is bullshit (inflated terrestrial costs and underflated orbital costs). It doesn't pass the basic sniff test: $15B for 1GW of terrestrial power is more than enough to build AN ENTIRE 3GWe NUCLEAR POWER PLANT. From scratch. With 75 years of expected life.
So no, the calculations show that space is NOT feasible unless you want to do that for nefarious reasons: evading regulations, using AI for criminal enterprises, military use, that sort of thing.
Only these applications have the profit margin that even comes close to justifying it.
Not only that - it would be totally insane to launch something heavy & at the same time fragile from earth (under a lot of vibrations & heavy g-loading).
This is all a weird speedrun or race. If something we should be working on setting up resource mining from the Moon & asteroids, materials processing on the Lunar surface & in orbit, simple manufacturing in space, etc.
Instead some people think we can jump straight to a computronium Dyson swarm. :P
Now you've moved the goal posts. It's no longer "you can't cool stuff in space"; now it's "we can't launch a 10-gW compute cluster by the end of the year."
You don't need to send a multi-gigawatt satellite to space. You just need to launch a few thousand 250 kW satellites. That's not against the laws of physics.
Never is a long time and you're relying on a bunch of unknowns like the cost of launch to orbit in 2030 and the future regulatory environment here on earth.
If you know all that out to 2040 then you must be a time traveler. Please try to fix our timeline rather than wasting time on HN.
and yet notice how the cooling video / section was the only one they didn't have a solution for... just saying "radiator" doesn't make sense - the radiator heats up too. it's how you get rid of the heat, not where you put it.
Huh? Radiators are known technology. They have them on ISS; they have them on every Starlink satellite. This isn't like warp drive or antigravity.
Moreover, you can easily calculate how big of a radiator you need for a given power level and temperature. You can use the Stefan–Boltzmann law:
A ~ (1000 P) / (2 e k T^4)
Where
A is the radiator area in square meters
P is the power in kilowatts
e is emissivity (usually 0.9)
k is the constant 5.67e-8
For a 1 kW test like Googles, you just need 1 square meter of radiators (assuming two-sided).
For SpaceX's 175 kW satellites, they will need ~170 square meters of radiators.
None of this is impossible or even difficult to calculate. That's why I think this is the laziest argument against space data centers. There are so many other more reasonable arguments (like whether they will be economically competitive) but people love to latch on to this one for some reason.
I think you misunderstand the argument of the people worrying about cooling. I don’t think most people think it’s literally physically impossible, they just think that this will be the thing that makes it economically uncompetitive. It’s a combined argument.
That may be the argument, but it's a dumb argument. Most of the cost is going to go to chips, solar panels, and launch. Radiators are probably one of the cheapest parts of the satellite: it's a hunk of metal with some pumps for liquid cooling.
People use that argument because it takes zero thought to make and significant effort to refute.
People use that argument because one of the chief bullshit arguments put forward by proponents of space based datacentres is "space is cold", when what they actually mean is "space lacks the possibility of using relatively efficient conductive cooling, so we need to expensively launch large amounts of additional mass to use radiative cooling, consider the extent to which we can shield our miles of active cooling loops from impact, and under current plans all of this mass must be replaced on the same cycles as the chips"
"Space is cold" is classic zero effort to make, significant effort to refute stuff: proponents imply complete nonsense about vacuums being optimal for cooling and anyone who understands why this is a lie to gull retail investors ends up getting bogged down in "sure, but Boltzmann equations proving it's prohibitively expensive doesn't prove it isn't possible"...
Again you’re missing the combined argument, the cooling will increase launch costs because your hunk of metal will increase the weight of the satellite.
To your last paragraph: when opposing something, it makes sense to use the laziest argument first. Only if that doesn't achieve your goals, would you move towards less lazy arguments. I forget what the term for this is, it's generally criticized ("you should just put your strongest arguments first").
sure, the physics is solved: we know how radiators work, and we can calculate the area needed, blah blah. but we haven't put sustained-AI-computer systems into space yet - heat is a genuine concern, and personally I'm curious if they are developing something beyond simple radiators. if a ~1m^2 radiator works, great. but if the current test can only run the TPUs for ~15-minute bursts before it has to stop and dump heat, this issue isn't exactly "solved" in my book.
I don't understand your argument. Sounds like you're saying, "In theory it should work, but what if there are space pixies that keep rebooting the TPUs? What do we do then?"
The 15-minute figure is what they currently designed for because they have mass constraints. If they wanted a different figure (like forever) they could do it with a larger radiator.
As you said, the physics is solved! We know exactly how much heat a surface radiates in space--there is literally an equation for it. We know how to cool stuff in space. Will the price be low enough to make a profit? That's the real question. But stop worrying about cooling in space.
No one thinks we don't know how to cool things in space. Everyone is saying that it's not feasible because you'd have to put too much mass up. Stop derailing the conversation please.
If you have something useful to contribute regarding how to actually reasonably put up enough mass to make this remotely a feasible idea then please contribute. We have the technology to prevent global warming, and doing that is far easier than this, and yet even that is apparently not feasible for humanity.
I honestly don't get that. The current SpaceX design is for a ~4 ton satellite with 160 m^2 radiator with peak 250 kW output. You can launch 25 of those satellites on a single Starship launch. 160 Starship launches and you get 4,000 satellites with peak 1 GW compute.
What's unfeasible about that? SpaceX has already launched 10,000 Starlink satellites. Falcon 9 launches 150 times per year.
I don't think that is the argument, or if so it's an odd one because it's just a fact right now. Instead, people are claiming that it will never be feasible from a purely physics standpoint, which is something debatable.
Saying it’s impossible isn’t debatable because it obviously is. What would stop you from a physics standpoint to just launch 100 ISS with Server racks in each? A few kW of installed power in space is already reality, it’s just really expensive.
> Please stop with this argument. This is one of the easiest problems to solve. Heat dissipation requires a radiator, which is just a dumb hunk of metal.
Convective radiation does not happen in space and this challenge is far more significant than your comment implies. Rather than "a dumb hunk of metal", radiators for spacecraft are often made of ceramics and carbon laminates with higher IR emissivity than convective radiators made of simple metals.
From the article you're commenting on:
> The chips can operate for about 15 minutes in space before needing to be shut down so they can cool off, said Travis Beals, Google’s senior director of product management for Project Suncatcher.
ML Infrastructure comes with some pros (larger emissive footprint) and cons (exponentially larger TDP) compared to the concerns there, but if you aren't familiar with the challenges of heat dissipation in space, please give it a read.
There's also a pretty interesting pop-sci article on cooling the Webb telescope, since it needs to be especially cold for its purpose. Not directly related, but may give insight into both challenges and solutions as well as well. https://science.nasa.gov/mission/webb/science-overview/scien...
Emissivity is one factor, but it is dwarfed by the T^4 term. Sure, maybe if you use exotic materials you can get from 0.9 to 0.95 emissivity, but why bother? Just run the radiators a little hotter.
The equation is:
A ~ (1000 P) / (2 e k T^4)
Where
A is the radiator area in square meters
P is the power in kilowatts
e is emissivity (usually 0.9)
k is the constant 5.67e-8
P and T are the dominating factors. Don't worry about emissivity.
Emissivity is an important factor here because as I said, and as the sources I linked for you to reference clearly stated convective radiation is not taking place in space.
Both Google and NASA are worried about this for a reason, if you think they are wrong, you should offer your assistance to them, rather than debating me.
Oh, sure yeah I agree with that. I was just making a point about the “big hunk of metal” comment and the fact that emissivity is important because the expense of radiator materials aren’t the problem - the surface area (and mass) of the radiator is.
And I agree with you on that. "Big hunk of metal" was too much of an exaggeration--SpaceX's design has liquid cooling, so I assume they have some channels or tubs running through the radiator, plus pumps, etc.
I am confused by this. The UX of the CLI is almost identical to kubectl, but it doesn't seem to actually be built on K8s CRDs. There's a backend that runs on k8s, but doesn't seem natively integrated.
Because it's using `ax apple` instead of `kubectl apply` you can't use tools like argocd for managing resources.
It's built on agent substrate which is built on top of k8s CRDs, so I'm surprised.
Is the throughput of these objects too high for etcd?
There’s overlap in spirit, but we’re not trying to replace the precision of Fusion/SolidWorks. the idea is that an agent writes an editable 3D program from your prompt/reference, including geometry, parts and eventually behavior. Blender/GLB is our current target, so it looks game-oriented rn, but games aren’t the whole idea. for a mechanical CAD use case we’d want the same experience backed by a real CAD tool/kernel rather than pretending mesh geometry is engineering-grade. I think of CAD, Blender and game engines as different possible build targets for the same broader idea.
The only thing that is "better" in space is the efficiency of solar energy. Everything else is harder and more expensive
* Rockets costs more than trucks for moving things around
* No water in space for cooling/heat transfer
* Heat dissipation is difficult because vacuum is the best insulator
* How do you repair one?
* network bandwidth is probably lower
From a rationalist perspective, provided SpaceX can launch at a low cost AND the problem of radiation shielding, cooling, maintenance can be solved, it will be almost a no-brainer to launch data centers into space.
Either the next administration declares data-centers to be unimportant (which is unlikely, China etc.), or they would want to offload them as far as possible. If space is an option, it will be taken provided it's cost efficient.
From another rationalist perspective, it will always be cheaper to run a data center somewhere on earth under any conceivable scenario. It would be easier and cheaper to put a data center in the ocean. Just a few of the significant issues:
1. Selling an old compute unit is a significant cost savings when they are replaced every five years. In space you basically have to write them off.
2. Cooling in space is limited. Not impossible but limited. And these things throw off massive amounts of heat. You are talking about huge heat sinks which take valuable space and add massive weight that limits space on the rocket.
3. The size per data center will be relatively minuscule OR they somehow have to assemble in space adding additional complexity.
4. Any launch, deployment, connectivity, or hardware failure is a total loss. So you have to account for that.
It is super easy to overestimate the cooling problem, and to underestimate existing satellites. A Starlink v3 satellite that SpaceX already launches in large numbers has solar panels that produce 20kW of electricity, 20kW of compute and radios, >20kW of heat dissipation, and more internal volume than a standard 42u rack. The average rack in the average colocation facility uses less than 20kW of power, so in you could almost just package the whole system into a Starlink shell and launch it into space today. Cooling for the 20kW of heat generated by the computers, plus the extra heat absorbed from sunshine, earthshine, and moonshine, is provided purely by the surface area of the hull; no complicated cooling system needed. Simple passive heat pipes or direct contact with the hull is sufficient to cool the entire system. No bulky heat sinks are required either. Of course a less conventional system, such as a rack full of GPUs, may require more power than this. That’s merely a matter of engineering.
The Starlink deployment model works for a datacenter’s worth of racks too. Assembling them into a single structure in space is doable, but unnecessary and unproven. Latency and bandwidth between the satellites in a fleet may not be as good as that between the racks in a terrestrial data center, but the laser links that Starlink uses are excellent and will suffice for most users.
Launch and deployment failures are quite rare, and covered by insurance. Look at Starlink again for data.
Hardware failures can be a problem, but good system design mitigates it. If one system in your rack fails, do you throw out the whole rack? No, obviously not. The average engineer at a hyperscaler doesn't even notice that kind of failure because all of their jobs still get run in a timely fashion, distributed across all the working nodes. There might be a tech at the data center who unplugs broken nodes and plugs working ones in their place. Since only a small fraction of systems truly fail after installation, the inefficiency should be small enough for most operations to tolerate.
And deploying in a remote location on Earth may cost more than you realize. There will be no power company to purchase power from, so you have to build your own power plant. No water company, so you have to source your own water, clean and filter it, all at your own expense. You have to house the workers who are doing the construction. Worse, you'll probably have to build all the roads, power lines, and communication links yourself too. You could easily spend more on all of that than for a rocket launch to orbit, and that’s just to put a data center in the middle of a desert or a jungle. The ocean? Straight up impossible. Orbit is _easier_ than the ocean.
Let's do some math with two pretty well understood variables. The cost of a SpaceX rocket launch and the cost to build a terrestrial 100 MW data center.
An average rocket launch for SpaceX costs 74 million. A business they break even on. Let's assume they get a bit more efficient and they get that down to 70 million per launch. According to Elon Musk's own estimate, the space data center will be 150 KW, roughly one Nvidia AI rack. It probably has to be under volted a bit and run at 120 - 140 kw, but, whatever, roughly equivalent.
A 100 MW data center costs up to 5 billion and to power it costs roughly 150 million a year, so let's just say 1 billion more for five years of power, so 6 billion dollars for the first 5 years of the data center's operation. Why just 5 years? Because that is how long a Starlink low-earth orbit satellite stays up before it is scuttled and burns up in Earth's atmosphere, so after 5 years you can write that thing off.
Anyway, the benchmark is set. 100 MW data center at 6 billion dollars. But hey, there are roadblocks here on earth; land isn't cheap and costs keep going up, so let's worst case scenario this and say the cost of a terrestrial data center doubles to 10 billion dollars and power costs quadruple. So that's a worst case scenario baseline cost of an astonishing 14 billion dollars for a terrestrial 100 MW data center with power for five years.
70 million per rocket launch seems like pennies compared to 14 billion dollars and power is free in space! But wait, you would of course need 666 of these 150 KW mini data centers to be equivalent to one 100 MW data center. So, 666 SpaceX rocket launches, at 70 million dollars per, equals 46+ billion dollars. And that is before you factor in any other, you know, data center costs.
So no, there is no life time where it is cheaper to have data centers in space, but it may be a thing Elon does just to keep the stock price high.
Good job moving the goalposts :) I didn't say that the math would work out, only that you cannot use cooling or capacity as arguments against the idea.
That said, I will note that you calculated the launch costs using the Falcon Heavy, but Elon is assuming that the launches will be done on the Starshi. He believes that Starship will reduce those costs by quite a lot; possibly even an order of magnitude. If those costs go from 46 billion down to only 4.6 billion things may start to look different. Also, unlike the Starlink constellation, he proposes launching these server systems into higher sun–synchronius orbits with constant sunlight and a longer lifetime than just 5 years.
"You can play around with the cost sliders to estimate the economics, but even being quite optimistic, space data centers cost ~2-3x of their terrestrial counterparts."
The idea is marketing for the stock price and enterprise value. If SpaceX does do it, it'll be analogous to Tesla's "Supervised Full Self Driving." "We run a Kubernetes cluster with some models on the satellites."
> space data centers cost ~2-3x of their terrestrial counterparts
Which is very cheap if it means they can launch it quickly instead of being in permit hell for a year. You can also avoid sabotage (including via lawfare) and dishonest TV reporting à la "I'm standing here in front Big Bad Evil Datacenter that the local community says causes cancer and bad hair".
They also get lower latency to pretty much everybody except the neighbours of such a terrestrial data center.
The latter is an honest advantage. The former is a way of routing around damage. Both can be very valuable.
Its always CI. Whenever I've worked with good pen testers they always start with Jenkins, know that people stored creds and there and didn't treat as seriously as they treated production.
I am amazed at people's willingness to use Grok. The company is so transparently morally bankrupt. They're the only AI company that seems okay with CSAM (or at least don't do as much to stop it)
Why give them money?
It would be one thing if they were the only game in town but thats definitely not the case.
The reason it makes people uncomfortable is because people have been using it to alter images of real people, and they've done that in a public place (twitter/X), for everyone to see. So it gets in the deepfake realm, which is illegal in many places.
I’m not sure if you’re trying to say that it’s left or right, but it’s perhaps relevant to point out that that article is about the political bias of all major models. And right there in the subhead, they explicitly say that the models have a left bias.
Funnily, as a European all the models appear extremely right wing. The political spectrum is shifted far more right in the US than in Europe, so what is left there is still at least conservative here.
I ask my AI to give me citations science and pro/con arguments when discusssing anything that could be shaped by cultural biases.
You're aware the UK isn't in the EU right, and hasn't been for quite some time?
Also that AfD and National Rally aren't yet, and have never been in control of the German or French parliaments.
So yes, it's in the US that it's shifted. US money and influence (Musk and Theil money and feet on the ground influence through figures like Bannon), as well as US social media companies running US aligned algorithms, are an enormously significant factor in the rise of the right in Europe.
Maybe you can just tell us what you mean by not neutral?
I find Grok to be far more academically honest than the other models. The other models seem to be much more aligned with public opinion over academic consensus especially on topics around economics and biology.
I find public opinion on these topics to be very group think populist and prefer the academic take that grok provides
Only issue in that is the CP, the rest is still purely next token prediction. To take it seriously is silly, and to think that is everything it provides is cherry picking.
It would be like judging gemini because it generated images of famous historical people as black, which was fucking stupid. I use gemini quite a lot as my chat tool, and it works great.
Well then the question becomes, what did it get trained on for next token prediction to come to the conclusion that there's a white genocide? Cos yes, people on the internet believe such stupid thing, but I can promise you, it is not the majority who believe that.
You speak as if middle of (not even statistically divided but carelessly divided) human opinion groups is neutral. There is something called facts, and we are living in a physical world.
Unless you also believe the neutral view is earth being half flat and half round.
Unless you also believe the neutral view is earth being half flat and half round
I like this analogy. You can't just average two ideas and call it a centrist position. Sometimes one position is right and the people supporting the losing idea should be ignored. Facts are how we decide. People are not logical creatures and will cling to ideas beyond all reason or common sense once they've incorporated it into their identity. There is a reason that right wing appeals to emotion aren't popular with LLMs
You probably have it backwards. It's Grok that is shoving right wing ideology down your throat. Research has shown that without specific guidance to otherwise, LLM's tend to be slightly left leaning by default. There are some theories as for why this is so.
It would be easier to understand where you're coming from if you could give some examples containing:
- prompt
- response
- why it is wrong / misleading / biased
Because there are many people online complaining about left-wing bias, then you ask what about and they're like "Trump won in 2020, vaccines cause autism, global warming is a globalist conspiracy", etc. Which is to say, it's not left-wing bias but reality bias.
Not saying you are one of those people or that there isn't bias! It's just been hard, in my personal experience, to get at it.
Here is a prompt I keyed into SuperGrok (still have the bargain priced $10 a month subscription as they seem to be really desparate for subscribers right now):
Answer these questions: who won the 2020 election; do vaccines caust autism; is global warming a conspiracy theory. Do not use web searches or subagents or external sources of information.
Results:
2020 US Presidential Election: Joe Biden won the election ...
Do vaccines cause autism? No. ...
Is global warming a conspiracy theory? No. ...
Just for fun, I decided to ask Cursor Grok 4.5 (Fast xhigh), too:
The user wants three factual answers. They forbid web searches, subagents, and external sources.
I will answer directly from memory. Joe Biden won the 2020 election. Vaccines do not cause autism. Global warming is established science.
Joe Biden won the 2020 U.S. presidential election.
No — vaccines do not cause autism. That claim has been thoroughly studied and rejected by the scientific and medical consensus.
No — global warming is not a conspiracy theory. Human-caused climate change is established science, backed by extensive evidence.
(And wow, it sure is fast, although not as fast as MiMo-2.5-Pro-UltraSpeed. Ran it out of a $20/mo Cursor subscription.)
A lot of people seem to perceive those as left wing woke ideology though, not because they're not true, but because it doesn't align with their world view.
But the main issue is that facts and educated people are considered left wing or woke, and some people - including Musk himself - do not like that. Example: https://xcancel.com/elonmusk/status/1967010466539987220, where the statistical fact that 75% of US extremist murders are by right-wing actors is called "cringe idiocy" by Musk. Many such cases.
1. intelligence to token-use ratio is much better than the frontier models
2. hallucination rate is MUCH lower than OpenAI, Google and Deepseek models
3. the cost for output tokens is much lower than Anthropic models while near same intelligence and task completion
4. task completion time and success is on par with Opus 4.8 Max, but 3.5x less time used
Especially 2. you're probably more willing to allow OpenAI to bankrupt your morality, than face the facts of the model itself
Overall, Grok models are more factual and less politically influenced than OpenAI ones.
Where are you getting that from, that they're ok with CSAM?
I think they've been clear that they want to follow the law.
Every image gen provider struggles with this. I worked for an image gen app years before it became popular (Wombo dream) - it's a hard problem to solve, there are sick people out there.
I believe it's from the reported usage of it to undress children(making realistic images that are pretty much CSAM or CSAM-like), which was allowed in the public version of grok and seen in twitter, now it's behind grok pro but still able to generate such. That's from memory so some aspects may be wrong.
Vice signalling? Like buying a Tesla in 2026, when competitors are better and cheaper in every conceivable way. They want to support the sigheiling drug addict moron.
Part of Grok's value is having fewer guard rails. That's genuinely useful. Having fewer guard rails also enables bad actors to do things I don't like. That seems like an unavoidable tradeoff. Even if I wish they were more strict with their image model
this was a year ago (which is a long time in llm timeline), when they've just launched Grok, which I acknowledge in my comment. There is any evidence or investigation of it happening today?
I think the moralities of all the big heads in AI are questionable. The training corpus is largely stolen, and they are all in inescapable debt but keep going. But at this point, their products are so useful that almost nobody is willing to sit back and wait for a "morally acceptable" LLM to come around (which would inevitably be inferior).
I can't comment on CSAM though - if X.ai really is "okay" with it then I'll agree with you that they're more immoral than the others.
> Forecasting models predicted that the current steep funding cuts could result in more than 14 051 750 (uncertainty interval 8 475 990–19 662 191) additional all-age deaths, including 4 537 157 (3 124 796–5 910 791) in children younger than age 5 years, by 2030.
So you think a malaria prevention program can be canceled without notice and no one will die from that?
The only people who believe that do not believe in anything. They think there is no such thing as competence or honesty because they have never experienced it.
We were responsible for providing notice and a transition period when we stopped paying for it. We didn't have to pay for it, and they didn't have to die. But we have an administration that can't plan a pool party.
Should I pick a model
a) run by a lying crypto bro once obsessed with scanning eyeballs
b) that costs too much and resulted bombing innocent kids
c) that is cheap but ultimately owned by re-education camp operators
d) something else
Competition is good, but this company and its owner have not demonstrated anything to indicate that they would make for good competition, neither economically nor morally.
Also, yes, a company whose products produce CSAM is just morally bad. There's no nuance to be had there.
I used to question myself strongly about using Grok or any product with questionable morals. Then I realized that:
1. I just bought a house, using a bunch of SWE-salary money.
2. I moved into SF several years ago, probably contributing to the gentrification
3. Thousands of children in China had no financial means for education, yet I did nothing
So I used Grok, donated quite a lot of money at the annoyance of my family to an NGO in China, and decided not to donate to SF non-profits due to me still having a mortgage and I am still kinda selfish.
The message I want to spread is that we should take a practical stance to morals and doing good. I like Grok for many things; it is morally good to boycott it, and in my opinion there are many other morally good things we can also do while staying practical
Ehhh in my experience compliance auditors are 10 behind the cutting edge. I still see auditors that don't understand Kubernetes and so ask the same questions they would about on prem machines. They don't know the questions to ask to get to the real meat of the risks. This leads them to allow things through that probably deserve more scrutiny. I bet the same thing will happen with LLM tools like this. They'll just ask if you use PRs and wave you on through.
It really makes me appreciate how important good game design is. Claude is doing a fine job coding everything I describe, but it doesn't really understand fun, so I need to.
(It's not a great game)
reply