I don't think the atoms point is right. Silicon semiconductors are made of atoms. Also, we get Moore's law like behavior in lots of very physical industries, e.g., the cost of solar panels.
A pretty simple model that accounts for the data is that Moore's law, and many other exponential growth examples, require ever larger capital expenditures. This worked for Moore's law because at ever step of improvement the devices produced were highly economically valuable. For fusion, on the other hand, you can have an exponentially improving triple product, but it has zero economic value until you cross the net-positive threshold. That basically means that the exponentially increasing development funding needs to be provided by the government, philanthropy, or some other non-profit source. If you're exponentially improving, with exponential costs, and you hit the ceiling of what the government and philanthropists are willing to provide, your progress can come to an abrupt halt without it necessarily meaning the basic exponential engineering curve you were following stops.
I think the parallel between cost of semiconductor fabs increasing and costs of fusion reactors increasing is quite apt.
But we don't actually have exponential improvement in any physical object, that's not to do with information processing - a solar panel or battery made today is not 10x better than one made 10 years ago.
It's not even true of all semiconductors - power electronics, radio, etc.
> But we don't actually have exponential improvement in any physical object, that's not to do with information processing - a solar panel or battery made today is not 10x better than one made 10 years ago.
Do you understand how differently that reads from your original comment?
> Fusion has advanced faster than Moore's law - and unlike the holy grail of computing, true AI, it's now clearly within reach.
You made the claim it's advancing at a rapid rate and almost here, and when someone pulled up the data it wildly disagreed with you. Now you're just moving the goalposts.
A pretty simple model that accounts for the data is that Moore's law, and many other exponential growth examples, require ever larger capital expenditures. This worked for Moore's law because at ever step of improvement the devices produced were highly economically valuable. For fusion, on the other hand, you can have an exponentially improving triple product, but it has zero economic value until you cross the net-positive threshold. That basically means that the exponentially increasing development funding needs to be provided by the government, philanthropy, or some other non-profit source. If you're exponentially improving, with exponential costs, and you hit the ceiling of what the government and philanthropists are willing to provide, your progress can come to an abrupt halt without it necessarily meaning the basic exponential engineering curve you were following stops.