A data center in Downtown Los Angeles may boast about the multiple backup power generators they possess, but is it enough to deal with power problems that span the whole city for multiple days? Although problem scenarios may seem farfetched when an information architecture is functioning normally, problems do occur as exemplified by the extended power outage around New York City. Consolidated Edison, a 10 billion dollar company, was unable to provide highly available service. A majority of co-locations do not spend as much as ConEd to create an environment that supports high availability, and a majority of companies are not willing spend too much on minimizing downtime.
To minimize the cost while maximizing service availability, a solution for providing high availability should include at least two data centers that are in separate geographic regions. This is more advisable than building up a single super data center. As an example at a smaller scale, buying two servers and configuring them for high availability is cheaper and safer than buying a single server that is supposedly fault tolerant. A single server, by definition, cannot provide redundancy that is needed to foster high availability. When dealing with high availability, scaling out is simply better than scaling up.