[ExI] Power satellites are being developed now

[Adrian Tymes]

On Sat, Dec 20, 2025 at 6:49 AM John Clark <johnkclark at gmail.com> wrote:
> To be economically viable a data satellite would need one of two things:
>
> 1) A system, probably automated, that would provide maintenance and allow for hardware upgrades to be made.

Key problem: how do you get the new hardware to the satellite?  One
can imagine automated servicing satellites that swap in the new
hardware, but the chips have to come from somewhere.  While orbital or
lunar manufacturing is eventually possible, it will take much longer
to set either up at scale (at least 10 years) than the typical time
horizons of those proposing orbital data centers (less than 5 years).
This remains true no matter when one believes the SIngularity might
occur.  (Even if it occurred tomorrow, and resulted in an AI-driven
all-out push to set up off-Earth manufacturing, it might get to scale
in less than 10 years, but still over 5.)

> Or
> 2) The satellite is able to recoup the costs of construction and of launch in just two or three years.

This may be possible, depending on exactly how one finances it, and
thus on from whose perspective the costs are recoped.  Even "the
business" fractures into multiple points of view: did the data center
jack the business's stock price so it only paid itself off from the
point of view of then-current investors, did it help the business
qualify for a loan, or did it actually get a grant or other income
from the data center that the business does not have to pay for later?

> As for latency, I don't think that would be a major problem because it doesn't matter much if it takes 3 seconds for an AI to answer your question instead of 2 seconds, although you wouldn't want to use it to operate a car or a robot that's on the Earth.

It's not primarily the AI-to-user lag, as the AI-to-everything-else
lag.  AIs often have to look up stuff for complex queries, and connect
to offboard resources.  I've seen this get to 10s even in some simple
queries.

And, yes, people are looking to have such AI operate cars or robots on Earth.

> The lower the orbit orbit the cheaper the launch costs, and the less latency you have, but you can't go too low for various reasons, so I asked Claude

As a non-AI with some expertise in the field, I'd recommend no lower
than 400-500 km (and even then, perhaps an ion drive - though I
recommend some form of on-satellite propulsion in most cases anyway).
At 300 km or less, atmospheric drag makes orbital endurance of
spacecraft without stationkeeping propulsion a few months at best.
https://www.lizard-tail.com/isana/lab/orbital_decay/ provides a good
overview.

> Maximum debris concentrations occur at altitudes of 800-1,000 km and near 1,400 km
> Over 70% of debris is distributed between 500 km and 1,000 km
> Below 600 km, debris normally falls back to Earth within several years  due to atmospheric drag, while at 800 km altitude, orbital decay takes centuries
>
> The Optimal Zone: The 600-800 km range offers a compromise: you're high enough to avoid the very low altitude debris that decays quickly, but low enough to stay below the inner Van Allen belt's main concentration. You'd also be just below the 800-1,000 km zone where both debris density and radiation exposure increase significantly.

Yeah, the thing is, low altitude debris does decay quickly - which
means it ceases being a problem relatively quickly.  Even the ISS
stays around 400 km (with occasional reboost to stay up), and it
hasn't had a major problem with debris (though there have been
occasional incidents).

Try that calculator I linked, plugging in numbers for your data
center.  If, as you say, it needs to pay for itself in 2-3 years
anyway, 500 (maybe even 400) km should give more than enough orbital
lifetime, be low enough that you're under any long-lived debris, and
give much lower latency (especially given the accumulated effect of
multiple background interactions per user query) than 10,000 km
orbits.