Here's an argument why data centers will be built in space instead of on the ground. Projected by 2040, $700B/year of costs due to environmental, regulatory, and delay costs for building data centers on Earth. Compare this to decreasing launch costs (not shown here) to predict
@DrPhiltill Could they also use the outside air temps for cooling?
@LordxEkim Air has a much lower heat capacity and thermal conductivity than water, so it is hard to carry as much heat away from electronics using just air. In space you will need a working fluid. The space station uses ammonia to cool its electronics and radiate the heat away into space.
@DrPhiltill Cooling a massive amount of hot gpus in orbit will require significant radiative panels. The ultimate winner will be asteroid mining for resources + refining to raw materials + manufacturing solar, batteries, radiators and maybe GPUs in orbit.
@DrPhiltill Have you factored in the potential new usage of nuclear waste for fusion fuel?
@DrPhiltill Add to this that the downward crashing cost of launching a rocket means that many more countries will have easy access to space, by then, and this will mean increased pressure and even higher latency, if we continue to depend on ground based networks.
@DrPhiltill Perhaps this could improve the situation in the long term, although in the short term the orbital data center may be the best option. https://x.com/Dr_Singularity/s...
@DrPhiltill Energy demands may change if we (humans) figure out what we are doing wrong with training/inference since it’s wildly inefficient compared to what nature has accomplished. I’d also factor in the costs of mishaps in space (collisions, debris, etc).