Love this perspective on energy becoming the ultimate constraint for AI, it's such a brilliant way to frame the problem beyond just software. It makes me wonder about the practicallities of orbital data centers for real-time applications; your ability to connect the physics to the future of AI is realy insightful.
Love the frame of moving compute to the energy source instead of the other way around. The formation flying to get Tbps optical links is clever, though thermal in vacuum and latncy back to Earth feel like the next big blockers. If launch costs really brush $200 per kg, the econmics start to rhyme with a power bill, which is wild. Biggest open qs to me are repairability and yield in orbit, so lots of redundancy and fault tolerant sw will matter more than shiny TPUs.
Love this perspective on energy becoming the ultimate constraint for AI, it's such a brilliant way to frame the problem beyond just software. It makes me wonder about the practicallities of orbital data centers for real-time applications; your ability to connect the physics to the future of AI is realy insightful.
Love the frame of moving compute to the energy source instead of the other way around. The formation flying to get Tbps optical links is clever, though thermal in vacuum and latncy back to Earth feel like the next big blockers. If launch costs really brush $200 per kg, the econmics start to rhyme with a power bill, which is wild. Biggest open qs to me are repairability and yield in orbit, so lots of redundancy and fault tolerant sw will matter more than shiny TPUs.
Very interesting! Now I'm wondering what's the best solution for my startup servers, and I'll think about something similar (send money pls)