Spoiler alert: it could be 40% cheaper, according to some napkin math I did this weekend, prompted by Zuck’s comments on Thursday on the astronomical number of H100s Meta is investing in.
Interesting thought experiment. A few big underlying assumptions (GPU cost in 5y, power cost in 5y, launch cost in 5y, and AI compute on data center vs edge in 5y) but all on the right side of the secular trend. Key qn will be the magnitude of changes
Absolutely. Very, very rough napkin math here. Including price of power, and GPU utilization too –– I was actually surprised that even the high level numbers checked out though. More feasible than I thought originally
Interesting thought experiment. A few big underlying assumptions (GPU cost in 5y, power cost in 5y, launch cost in 5y, and AI compute on data center vs edge in 5y) but all on the right side of the secular trend. Key qn will be the magnitude of changes
Absolutely. Very, very rough napkin math here. Including price of power, and GPU utilization too –– I was actually surprised that even the high level numbers checked out though. More feasible than I thought originally
Very nicely articulated. Curious, why not in Antarctica or Arctic?
Having data centers in space would be awesome! I'm wondering how often the hardware would need replacements.