3 ms·

Nuclear power can be an answer, but it's not flexible enough to be the answer. The demand for power is variable, and the difference between base demand and pea

Nuclear power can be an answer, but it's not flexible enough to be the answer. The demand for power is variable, and the difference between base demand and peak demand can be 2x or even 2.5x. While nuclear power can be viable for satisfying the base demand, it's viability for the variable part is another question. Then there are cheap renewables. If you don't constrain the market artificially, it tends to build solar and wind until nobody knows what to do with the excess electricity. That makes power cheap on the average but potentially very expensive when the demand is high and the generation is low. Nuclear power, with its high capital costs but low operating costs, does not seem to be the right answer to this variable demand / variable generation problem.

UncleEntity 4y ago
There are many, many industries that could benefit from electricity being “too cheap to meter” that could suck up the excess production. Maybe they train GPT-25 only when the demand is low enough the power producers basically give it away?
- jltsiren 4y ago
  That is already happening. One particularly promising scheme is using electricity to make hydrogen and then using that hydrogen in steel production. But those industries also have other expenses. Even if you assume that energy is free, the investment may not be viable.
- Brusco_RF 4y ago
  I heard the other day that many Saudi's leave their AC running when they leave SA for the summer months. Power is THAT cheap.
- epistasis 4y ago
  That only makes sense if the idle capital of GPUs is cheaper than the energy powering them. There are very few applications like this, and even fewer that are easily distributed to that you don't have to use expensive transmission lines to power them. For example, most new hydrogen electrolysis facilities are planning to build direct-connected solar without even connecting to the grid, and using the energy directly, because letting electrolyzes sit idle for a week is not cost effective. And the grid cost would be expensive too. So building utility-scale solar without a grid connection makes the most financial sense.

UncleEntity 4y ago

There are many, many industries that could benefit from electricity being “too cheap to meter” that could suck up the excess production. Maybe they train GPT-25 only when the demand is low enough the power producers basically give it away?

jltsiren 4y ago
That is already happening. One particularly promising scheme is using electricity to make hydrogen and then using that hydrogen in steel production. But those industries also have other expenses. Even if you assume that energy is free, the investment may not be viable.
Brusco_RF 4y ago
I heard the other day that many Saudi's leave their AC running when they leave SA for the summer months. Power is THAT cheap.
epistasis 4y ago
That only makes sense if the idle capital of GPUs is cheaper than the energy powering them. There are very few applications like this, and even fewer that are easily distributed to that you don't have to use expensive transmission lines to power them. For example, most new hydrogen electrolysis facilities are planning to build direct-connected solar without even connecting to the grid, and using the energy directly, because letting electrolyzes sit idle for a week is not cost effective. And the grid cost would be expensive too. So building utility-scale solar without a grid connection makes the most financial sense.