100% uptime is nearly impossible. 5ESS Telephone switches were designed with 99.999% uptime over 40 or 50 years, which amounts to a few seconds a year of downtime. However there was a lot of redundancy built into those systems to ensure that.
Network reliability in the datacenter can be at 100% if designed correctly, but once your connections are in any kind of range of a backhoe that reliability can go out the window. Even a datacenter that brings in fiber from two different directions who did not do their homework to ensure those connections never share a facility to their final destination are at risk of an outage someday.
TCP/IP was never designed for 100% network uptime, it was designed to make sure the packet arrived in one piece.
Trying for 100% server uptime is not really good in the long term, eventually a reboot needs to be done unless you're looking to find one of those uptime bugs that may strike after so many days which can be disastrous. A reboot every 6 months is generally a good way to flush out any bad memory and also let fsck do it's thing every once in awhile.