Interesting enough I had just heard Bezos talking about this same kind of thing at ABF Miami in a fireside.
Meet Project Suncatcher, a research moonshot to scale machine learning compute in space. Artificial intelligence is a foundational technology that could help us tackle humanity's greatest challenges. Now, we're asking where we can go next to unlock its fullest potential. Today we’re announcing Project Suncatcher, our new research moonshot to one day scale machine learning in space. Working backward from this potential future, we’re exploring how an interconnected network of solar-powered satellites, equipped with our Tensor Processing Unit (TPU) AI chips, could harness the full power of the Sun. Next step is a learning mission in partnership with Planet to launch two prototype satellites by early 2027 that will test our hardware in orbit, laying the groundwork for a future era of massively-scaled computation in space.
The research paper goes pretty in depth on how this would be possible.
Between the Bezos talk and this paper, what comes to mind is not just “data centers in space,” it feels like the interesting step toward building artificial rings around Earth.
Inter-satellite links
These power levels can be achieved by drastically reducing the inter-satellite distance. Since for distances larger than the Fresnel limit, received power scales with the inverse square of the distance due to beam divergence, flying the satellites in close formation (hundreds of kilometers, or less) provides ample power to close the link budget for high bandwidth COTS transceivers, as illustrated above in Figure 1. As the distance becomes very short (e.g., ∼10km for a 10 cm telescope), spatial multiplexing emerges as a new opportunity for further scaling.
Orbital dynamics
Figure 2 shows one possible configuration for an illustrative, planar 81-satellite constellation—-all placed in the orbital plane, at a mean cluster altitude of 650 km. The arrangement here is based on a square rather than hexagonal lattice, mostly to simplify its description. Cluster radius is R=1 km, with distance between next-nearest-neighbor satellites oscillating between (approximately) 100 and 200 m, as is shown in Fig. 3. We note that, of course, evolving constraints could change the optimal architecture for our constellation.
My understanding of this is instead of one giant monolithic station, you get a swarm of small bodies in carefully chosen orbits, packed just close enough that the laser links give you insane bandwidth, but still just far enough apart that gravity + a bit of steering keeps them from running into each other.
The piece that really hooks me is the lack of talk online around the type of materials being used. Right now everything still lives in the classic 1950's "satellite mindset: lightweight structures, precision hardware, minimal shielding," all launched from Earth.
The opportunity to think of these swarms as permanent infrastructure, basically described as "synthetic rings of compute and power," a new question pops up. Why are we pretending these are just fancy boxes of electronics instead of treating them like orbital geology? (with rock, regolith, bulk shielding and dumb structural mass that does not care about a few extra tons.)
Play that tape out for a bit and a moon base stops being science fiction and quietly turns into an engineering endpoint. If you really want rings of compute and power around a planet, in the long run it seems a lot more practical to pull stone and metals out of low gravity wells than to keep throwing fragile machinery up from Earth. Here is a fun video explaining it.
z{1..2}.area = length * width80 cm^2is 6400 times the area of1 cm^2?