It would take an enormous number panels to power a meaningfully large data center.
Credit: Aetherflux
Sometimes, there are just words that work. For a while it was “dot com,” later on “blockchain,” and we of course all know the very recent efficacy of the term “AI-powered.” Well, now the bloom is off all those old roses, and investors are increasingly suspicious of pure software, in any case. What’s in vogue now are AI services, and many investment firms are willing to put up the largest investment volumes in history to get access.
Enter the new useful term: data center. Tech startup and overall big-promiser Aetherflux now has a new big promise to add to its old ones: AI data centers in space.
In theory, the concept could help to mitigate the problem of energy generation for data centers, allowing solar panels above the atmosphere to capture all the light coming from the Sun. Since Aetherflux is associated with solar projects, this would seem to be a natural fit.
The Sun as seen from the ISS
Credit: NASA astronaut Terry W. Virts
But let’s be clear. Aetherflux is a startup company that has has yet to launch a single project. Its background is in proposing space solar power projects, with its current-highest level of advancement being a proposed test launch for 2026. It’s gone so far as to purchase a satellite bus, but hasn’t strictly proven it’s capable of anything, just yet.
Aetherflux claims to be building an “American power grid in space,” but that’s hyperbole so grand as to border on fantasy. Now, it’s pushing a Galactic Brain that’s arguably even more ambitious before it’s even tested the promises it has already made.
Why would Aetherflux do that? Well, maybe it has something to do with the fact that the startup’s most recent windfall was $50 million for space-based solar development, while this announcement could theoretically motivate data center-level investment literally hundreds of times larger.
The founder of Aetherflux is the billionaire co-founder of the investment platform Robinhood. Is this a company more focused on physical results, or PowerPoint presentations that have the potential to generate a few billion dollars per slide?
Put simply, over the next decade we are going to see a lot of companies trying to get their mouths up to the money-spigot. Gun to my head, I’d have to predict that this Aetherflux initiative will be “delayed” unto infinity—but I obviously don’t know for sure.
So, do space-based data centers make sense generally? To me, it seems iffy. There’s the need for extensive shielding to protect the GPUs from solar and cosmic radiation, which increases costs dramatically, but GPUs also have a limited lifespan, even under ideal circumstances. Maybe if you could recoup launch costs for 20 years it could make sense, but the hardware just doesn’t last that long.
Beyond taking a lot of energy, data centers also generate a lot of heat. You’ll notice that Aetherflux’s renders don’t seem to include any radiators or other ways of getting rid of heat. Remember that cooling in space is actually much harder than on Earth, since you don’t have nice, viscous air to pull heat out of things.
All those panels around the ISS? Only some of them are solar panels, while the rest are radiators that need to offload heat to avoid cooking the ISS astronauts—and the ISS doesn’t have anything internal that’s nearly as hot as a big box of GPUs running at capacity.
Radiator panels are an essential part of any heated activity in space.
Credit: NASA
Theoretically, all the extra costs of launch and operation in space are to be offset by the savings on power, but space-based solar is only about twice as efficient as terrestrial solar. You’d still need to launch an awful lot of panels to power a meaningfully large data center satellite.
Even this article that purports to defend the idea actually ends up pointing out how little sense it makes.
Elon Musk, reigning king of incorrect predictions, took to Twitter earlier this week to address this issue. “Satellites with localized AI compute, where just the results are beamed back from low-latency, sun-synchronous orbit, will be the lowest cost way to generate AI bitstreams in [less than three] years.” This from the man who claims to believe that AI will make all products and services functionally free.
Maybe we will end up getting our chatbot responses from space. Right now, though, it seems more likely we’re ramping up into a new, annoying era of billion-dollar PowerPoints. At least this data center project will probably never move forward—the bigger problems are the ones that will.