> zero downtime for small time scales... Most of the time
I read it as "if you take small enough discrete time intervals they won't overlap with any downtime". Or in other words "no downtime between downtimes". Yes, it's very in line with your video.
Are you talking about Google Compute Engine? In that case yes, because by default VMs are live migrated between physical hosts. This can be done for schedule maintenance or upon signs that the machine is likely to fail. Furthermore there are no physical disks for a GCE VM which is one of the more common failure points. The result of this is that GCE VMs often survive for months or years without downtime. Note that the SLA allows more than 3 hours of downtime per month. https://cloud.google.com/compute/sla
For physical servers the uptime is typically quite small. Of course Google isn't optimizing for server uptime so it isn't fair to say "well even Google can't do it".
