Post

here's a fun stat for the orbital data center curious: Google's latest analysis of launch prices for space data centers suggests Starship will have to fly 1,800 times to drive the learning curve for cost reductions down to $200 per kg. techcrunch.com/2026/10/01/g...

techcrunch.comGoogle thinks SpaceX's Starship has to launch 1,800 times before space data centers get off the ground | TechCrunchGoogle launched its first advanced chip into orbit to pave the way for space data centers.

24 reposts138 likes

Data centers and AI are bad and don't belong anywhere, so I'm going to use Google AI data centers to explain all the reasons why.

1800$/kg for Falcon Heavy is a new one; but still a terrible analysis to get there. FH has not decreased prices relative to F9 due to "heavier launch", which is blindingly obvious because F9 regularly launches heavier payloads than Falcon's Heavy current record mass to orbit launch.

The $/kg that matters is F9 internal rate because that's what enables Starlink and that sits at around 1,300$/kg (and is cheaper than FH). In regards to the required N. of Starship launches; SpaceX has been saying this for about 2 decades; so glad everyone could catch up I guess.

Do you have a cite for SpaceX's prediction that 1,800 flights would be required to drive down costs to $200/kg? Would be useful in my reporting

BAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHA!!!

it's the amount of flights they expect it would take SpaceX to improve the design, manufacturability, reuse, etc. to drive down the cost, based on how quickly they did it with the Falcon line of rockets

well kinda, the learning curve analysis seems to take its final price from what Musk said the cost of the FH would be at its first launch, $90M. But in fact most FH flights sold have cost much more than that.

Even if you could instantly transport data centers into orbit for free it wouldn’t make any sense to out them there. Between heat dispersal issues, power issues, data lag, damage from cosmic radiation, and the impossibility of maintenance, space is possibly the WORST place to locate a data center.

1,800 launches over 10 years = 180 launches per year, or about 3.5 launches every week. TechCrunch correctly highlights that this is an enormous operational assumption.