Even on previous gen nodes it takes a year or two to bring a chip to full production.

Thats the kind of process issue I hadn't understood. I guess the dynamics here don't make operating a "2nd gen behind" system to make memory economic in the good times, and so nobody is ready to ramp up to meet market shortages in the short term.

EU is paying to build out "lo fi" chips for car and other needs: They decided the impact on domestic industry of supply chain logistics to TSMC was bad. A lot of people are shouting "why aren't they doing extreme UV 1nm chip design" when the decision was pretty simple: its possible to source the machines, it's probably faster to get to high yield, the exposure to supply chain risk is real, the return is hopefully high in strategic terms, you can improve density as a follow-on.

No matter what ASML is doing well.

Not only that, but such a node is fully booked with existing orders, many which are long term commitments with penalties if they fail to deliver.