Startups dealing with big data and web presences are almost all heading in this direction. Rackspace and Amazon cloud services are the most commonly used infrastructure, supporting spooling up farms of compute resources in a few minutes to support demos and large-scale processing cycles, then dropping the compute instances to management levels until the next round of needed growth. At the moment, though, the economics don't quite work out over the long run. The monthly costs of maintaining large storage and compute instances in a commercial cloud become more costly than self-managed solutions over time. Thus, private clouds are an interesting alternative, providing similar resource flexibility while reducing outlays. Network collocation cloud solutions "in a box" is yet a third way, reducing the construction and infrastructure outlays while maintaining the virtues of pure cloud solutions.

Moving from one solution to the next is dictated by validation at each level of the business effort, funneling down the failures with low cost rented commercial clouds initially, then transitioning the successes to localized solutions as the economics point in that direction. The first part of the play is almost always lean, though, and clouds can help companies fail fast, fail cheap, and always learn and document what transpired.