While I accept of course that there are government-owned systems that have uptime measured in years (decades?), I think that extrapolating that into an argument about the overall reliability of those systems falls into survivorship bias.

But if I had to make an a priori prediction about which systems would have the better reliability over the long term - cloud-based or those run out of a corporate DC - my money would be on the cloud-based systems.

The cloud providers are going to have better management of power/network/hardware than the most mature government agency, simply because it's a core capability for them.