Back in 2024 a company called Aetherflux was founded with the intent to put up huge solar power arrays and beam the power back down to Earth. That goal apparently didn't last very long because the California-based company with everything named as if it's Texan, called Cowboy Space, has made it known their intent now is to launch a fleet of orbital data centers, which means to compete directly with SpaceX.
The California-based company, which is run by Robinhood co-founder Baiju Bhatt, intends to operate a constellation of 20,000 AI data centers in low Earth orbit (LEO). It will build those satellites in house and launch them on a homegrown rocket — a vehicle Cowboy Space unveiled here on Monday night (Oct. 26), at an event it called "Big Sky Rising."
"It's a fundamentally novel approach to rocketry in the modern era, in our opinion," Bhatt said during the event. "The idea is that you have a two-stage rocket where the upper stage of the rocket itself transforms into the satellite."
The rocket is named Big Iron, as is the data-center satellite, which I suppose is another way of saying it's a two stage rocket where the upper stage stays in orbit to become that satellite. Founder Baiju Bhatt is an Indian billionaire and absolutely can't be accused of not having visions for developing businesses.
The stainless-steel rocket will be 282 feet (86 meters) tall and 15.4 feet (4.7 m) wide. It will be powered by Cowboy Space's Clydesdale engines — nine in the first stage and one in the upper stage, the same configuration employed by SpaceX's workhorse Falcon 9 rocket. And, like the Raptor engines that power another SpaceX launcher, the giant Starship, Clydesdale will burn liquid oxygen and liquid methane, a combo known as methalox.
The first stage of the Big Iron rocket will be reusable and while they don't have a launch site named yet, they plan to be close enough to an ocean to recover the booster on a barge ship, like the vast majority of Falcon 9 launches. The upper stage, as mentioned, will stay in orbit as a Big Iron satellite and if there are going be tens of thousands of these on orbit, I'm sure they'll have something like a serial number or other Data Center identifier. The spacecraft will tip the scales here on Earth at about 33,000 pounds, and after unfurling its solar panels will be about 433 feet long.
During launch, the Big Iron will have a nose cone to protect the rest of the upper stage from the first minute or two of ascent, but it will not be dropped, as the Falcon 9 drops its payload fairings for recovery and reuse. Instead, the fairing will turn into a radiator, to dump the prodigious heat generated by Big Iron's GPUs (general processing units) into space.
Don't forget there will be 20,000 of these GPU satellites up there. That's a lot of GPUs: each satellite will carry five racks of them and will provide about 1 megawatt of compute payload.
Cowboy Space plans to operate about 20,000 Big Iron satellites in LEO, a constellation it calls Stampede. Each rocket will carry just one of these satellites to orbit, so it will take 20,000 launches to build Stampede — and, likely, several thousand more every year to maintain it at that number.
Cowboy Space therefore intends to launch very often — multiple times per day, from launch complexes it calls Trailheads. The company has not yet announced where these complexes will be, but each one will likely feature four active launch mounts, Bhatt said on Monday night.
So when does all this go live? In their minds, they're going now, but if you're wondering about when we'll see launches starting, I've got to think mid- to late-'29. At best.
Artist's illustration of Cowboy Space's newly announced Big Iron rocket lifting off. Note one on the right 30 seconds or so into its flight curving toward the upper right corner and one on a pad near the left edge that seems to be releasing some cold gas, so possibly in some part of a countdown. (Image credit: Cowboy Space)
An interesting note to end this with is that Cowboy Space is operating one piece of space hardware at the moment. It's a satellite called Reason-1, launched on SpaceX's Transporter-18 rideshare mission back on Oct. 1.
"Its job is to prove out power beaming: using a high-power laser to transfer energy from low Earth orbit to a receiver on the ground," Cowboy Space said via X on Sept. 28. "If you want to build data centers in space, you have to master the optical systems between orbit and Earth. This is where that journey begins."

"bounce a signal off a satellite" contest?
ReplyDeleteThat's pretty ambitious - they need to show some hardware and experience to back it up.
ReplyDeleteAKA: I'll believe it when I see it!
Jonathan
This sounds like a lot more mathing is needed.
ReplyDeletehttps://spectrum.ieee.org/orbital-data-centers-heat
ReplyDelete> At ABI Research, where I work as an aerospace analyst, we did a rough total-cost-of-ownership comparison between a data center on Earth and one in space. It showed that the cost to launch and run a GPU in space for a year is at least an order of magnitude higher than the same feat in a terrestrial data center.
article claims output radiator area equals input solar area.
google "thermodynamics radiating waste heat from satellite ratio of solar panels to heat radiators assuming waste heat is radiated at maximum temperature silicon chips can tolerate" claims output radiator area is 2.1 times input solar area.
On the other hand, if the control for the totalitarian enforcement drones is up out of reach, there isn't a data center on the ground for the citizens to attack.
Thanks for that backup article. It sounds roughly like I'd expect. Not based on calculating numbers just the old, "you can't get something for nothing."
DeleteRe: the last statement, the emerging trend that I think bears watching is the requirement that anything built on the ground has to pay for its own power generation and cooling water so that the area they built the data center in pays nothing for it and there's no reason to attack it. This seems to be emerging around the country.
I can't believe space is as cheap as 10X ground. Each of these factors look like integer multiples of cost to me, and making progress on one doesn't help you achieve others. Multiply all these cost factors and compare it the only advantage of space, free power: $0.15/kWh * 1000 * 24*365 = $1,314,000 for a megawatt/year:
ReplyDeleteradiation hardening
chips which run hotter
vibration and G force of launch
assembled in a cleanroom
enough spares that capacity is still useful in five years
broken stuff out and spares in without manual labor
deliver with a rocket instead of a semi
"Space is the ultimate high ground". What's the collective noun for debt slaves? So instead of "flock" cameras, government will have "peasantry" cameras.
I can't see a way that orbital data centers make sense for computing for earth use.
ReplyDeleteThe primary use I can see for them is to test and develop equipment to use in orbit around the moon or Mars, supporting research or colonization before it makes sense to have them in the ground.
Jonathan