"If the minimum subnet size is 2^64 , and the complaint is "the routers will fill up if we have millions of routes". how exactly are switches going to cope if you put millions quintillions of hosts on one subnet?"
I'm afraid this merely betrays your lack of understanding.
The large address space allocations reduce the number of routes required by allowing things to be properly routed by subnet. This was how IPv4 worked originally, but it ran out of address space for that scheme decades ago.
Here's an IPV4 example for simplicity. Say you have the following ip addresses:
Under the present overcrowded IPv4 scheme, a router might well need separate routing table entries to reach all those addresses.
Under the original scheme, it would have required a single entry to reach the gateway for the 134.*.*.* subnet.
The gateway for the 134.*.*.* subnet would then have required 2 entries for the gateways for 134.240.*.* and 134.116.*.*
The gateway for the 134.240.*.* subnet would then have required 2 entries for the gateways for 134.240.73.* and 134.116.56.*
The huge address space of IPv6 allows a return to a similar addressing scheme.