The mathematical physicist Freeman Dyson is famous for not having won a Nobel Prize for Quantum Electrodynamics (QED). This is not because he lacked for insight or contribution. Only three can share.
I used Ai to apprehend the gist. The laws of constraint are still in play. And even though I cannot foresee each and every law and bottleneck as well as you do it helps to know how feasible space data centers is in the near term...
From Gemini: *The author would likely conclude that mass-scale commercial data centers in space are economically unfeasible. While a "Bird" can imagine a beautiful, borderless orbital cloud, the "Frog's" unyielding laws of physics and economics—specifically the brutal realities of cooling constraints, lack of direct fibre access, and severe I/O bottlenecks—prove that keeping data centers on the ground, plugged straight into Internet Exchanges, remains the only truly worthwhile play. *
I think that, in the final analysis, somebody will figure out that it takes a LOT less power to build a workable AI. The second thing they will figure out is that you do not need to embed all of the knowledge of human history in the neural network. Developments in AI are here to stay and are impressive, but this industry is approaching peak stupid. It reminds me of Global Crossing, in which I was an investor, whose mission was to roll out optic fiber networks. It went to zero carrying massive debt leaving huge dark fiber capacity. A similar thing will happen here, in my view. It is not bad, there will be massive excess capacity in supercomputing. That is valuable if you know what to do with it.
I used Ai to apprehend the gist. The laws of constraint are still in play. And even though I cannot foresee each and every law and bottleneck as well as you do it helps to know how feasible space data centers is in the near term...
From Gemini: *The author would likely conclude that mass-scale commercial data centers in space are economically unfeasible. While a "Bird" can imagine a beautiful, borderless orbital cloud, the "Frog's" unyielding laws of physics and economics—specifically the brutal realities of cooling constraints, lack of direct fibre access, and severe I/O bottlenecks—prove that keeping data centers on the ground, plugged straight into Internet Exchanges, remains the only truly worthwhile play. *
I think that, in the final analysis, somebody will figure out that it takes a LOT less power to build a workable AI. The second thing they will figure out is that you do not need to embed all of the knowledge of human history in the neural network. Developments in AI are here to stay and are impressive, but this industry is approaching peak stupid. It reminds me of Global Crossing, in which I was an investor, whose mission was to roll out optic fiber networks. It went to zero carrying massive debt leaving huge dark fiber capacity. A similar thing will happen here, in my view. It is not bad, there will be massive excess capacity in supercomputing. That is valuable if you know what to do with it.