The reality is that we already have the most primitive version of this today. PhiSat already uses onboard processing for cloud detection and other EO applications. OroraTech is flying small GPUs on-orbit for fire detection right now. That’s “orbital compute” in the weakest, most practical sense.
Now, for the rest, there are certain parts that are easier than others...
I particulary like the idea of machine-to-machine tip-and-cue capabilities! Tip-and-cue has been revisited and demonstrated the last few years. And actually, it was first demonstrated back in 2009-2010 (tasking the UKDMC-I to acquire imagery).
Nowadays, with optical links and sat-to-sat comms, machine-to-machine tip-and-cue seems very feasible within-fleet (or at least within legal / export control frameworks, etc)
The hards part... As you say, power budget is the contraining factor here. The challenge is radiating massive amounts of heat! We're decades away from proper MWatt-powered orbital data centers (a 2MW facility would require approx 4000m2 radiator area)
You'd need some very fancy engineering or new tech to work around that!
Didn't expect this take on orbital compute, but wow. Logic for processing near sensor is solid. What about the power budget for those AI models?
The reality is that we already have the most primitive version of this today. PhiSat already uses onboard processing for cloud detection and other EO applications. OroraTech is flying small GPUs on-orbit for fire detection right now. That’s “orbital compute” in the weakest, most practical sense.
Now, for the rest, there are certain parts that are easier than others...
I particulary like the idea of machine-to-machine tip-and-cue capabilities! Tip-and-cue has been revisited and demonstrated the last few years. And actually, it was first demonstrated back in 2009-2010 (tasking the UKDMC-I to acquire imagery).
Nowadays, with optical links and sat-to-sat comms, machine-to-machine tip-and-cue seems very feasible within-fleet (or at least within legal / export control frameworks, etc)
Similarly, I'd love to see near-real-time downlinks enabled by sat-to-sat comms, and Starlink-like networking: https://www.muonspace.com/press/muon-space-to-integrate-spacexs-starlink-mini-space-lasers-into-its-halo-tm-satellite-platform
The hards part... As you say, power budget is the contraining factor here. The challenge is radiating massive amounts of heat! We're decades away from proper MWatt-powered orbital data centers (a 2MW facility would require approx 4000m2 radiator area)
You'd need some very fancy engineering or new tech to work around that!