ScalabilityThe great thing about the cloud and scalability is the scalability-on-demand aspect of the offerings of the largest providers. If you need more more processing power and/or memory at certain times of day, week or month or for ad-hoc periods, then you can temporarily schedule an increase in power and are only charged for the time you are using the higher specifications. The primary example of this is Amazon's Elastic Cloud Computing (EC2).
This is “scaling-up”. However, with all n-tier applications there is still a need to “scale out” to provide robust online applications. This is done in the cloud in the same way you would do this in-house or hosted, by authoring applications to be stateless, clustered and/or load-balanced with robust RDBMS back-ends.
Availability & ResilienceAmazon offers its EC2 service in a number of geographically distinct zones which do not rely on the other for availability. Therefore when one zone is unavailable for whatever reason, the other zones remain active. Scaling your applications out over multiple zones, or using Dynamic DNS to cut between zones when one zone is unavailable, means that cloud computing offers a lot of potential to guarantee the 99.97% up-time that many service level agreements require.
The platform-as-a-service (PaaS) vendors also have a great deal of geographic redundancy in their systems, meaning that your PaaS applications may well be available when a particular part of the providers infrastructure is unavailable. Economies of scale mean that having a third-party provide this kind of infrastructure is going to be a lot more cost-effective than providing it in-house.
Disaster RecoveryThe cloud is a great option for disaster recovery. If you have DR capability, or are thinking of investing in it, then you have to consider the cloud. The key area to consider is transferring large amounts of data over an internet VPN, which you may need to do in order to guarantee a seamless cut-over in the event of a disaster. The bandwidth provided by your ISP may not be enough to compensate for the dedicated WAN which most companies are using for DR at the moment.
Statutory requirements regarding privacy and storage of certain information may mean that you are unable to move all of your disaster recovery needs to the cloud. But there is certainly scope for every organization to implement cloud DR for at least some systems.