80C (still very hot for datacenter hardware coolant) is 350K. And there are other challenges, like effects from being in LEO, or proximity to wherever the solar array is.
There’s tons of spam about “solving” this after the Tech Bro boom, but I don’t really buy anything I’ve seen. Nothing but a bunch of lead (or the Earth’s atmosphere) is going to stop fat gamma rays or extremely fast nuclei.
AI doesn’t actually need that much. I’m pretty sure entire models like GLM 5.2 or Deepseek v4 are trained (and served) on a much, much smaller scale than a 100MW cluster.
But if that’s the case… why even invest in orbital data centers in the first place?
Why not desert ones? Why all this cash there instead of actually improving LLM architectures!?
There are so many nested levels of absurdity here. It’s all just total mania, with zero punishment to those doing the funding because they are too rich to feel any consequences.
deleted by creator
800K is 526C! You can’t run a datacenter at that.
80C (still very hot for datacenter hardware coolant) is 350K. And there are other challenges, like effects from being in LEO, or proximity to wherever the solar array is.
And this is just one of MANY ridiclous engineering challenges. Another great example is that GPUs, memory, and SSDs get random bit flips in orbit, and the issue gets worse with smaller lithography: https://www.itpro.com/server-storage/high-performance-computing-hpc/367323/hpes-supercomputer-helps-iss-astronauts
There’s tons of spam about “solving” this after the Tech Bro boom, but I don’t really buy anything I’ve seen. Nothing but a bunch of lead (or the Earth’s atmosphere) is going to stop fat gamma rays or extremely fast nuclei.
deleted by creator
To be fair, 100MW is pretty big.
AI doesn’t actually need that much. I’m pretty sure entire models like GLM 5.2 or Deepseek v4 are trained (and served) on a much, much smaller scale than a 100MW cluster.
But if that’s the case… why even invest in orbital data centers in the first place?
Why not desert ones? Why all this cash there instead of actually improving LLM architectures!?
There are so many nested levels of absurdity here. It’s all just total mania, with zero punishment to those doing the funding because they are too rich to feel any consequences.