That was slightly loose wording in the article: yes, the single data center is currently a single point of failure (though we can spin up a replacement stack elsewhere very quickly). Andrew was referring to the way we had rearranged our software stack to reduce the SPOFs (our previous architecture had a single load balancer for example).
