Method/system is disclosed for recovering computing capacity and critical applications after a catastrophic failure. The method/system involves distributing the computing capacity over multiple computing clusters, each computing cluster having concurrent access to shared data and software applications of other computing clusters. Sufficient backup computing capacity is reserved on each computing cluster to recover some or all active computing capacity on the other computing clusters. Message traffic throughout the computing clusters is monitored for indications of a catastrophic failure. Upon confirmation of a catastrophic failure at one computing cluster, the workloads of that computing cluster are transferred to the backup computing capacity of the other computing clusters. Software applications that have been designated for recovery are then brought up on the backup computing capacity of the other computing clusters. Such an arrangement allows computing capacity and critical software applications to be quickly recovered after a catastrophic failure.