Does it scale? Does the data center operator have a realistic sense of how much load this rack can handle when fully populated, with tens or hundreds of thousands of virtual machines? Does the owner have confidence enough in the capacity in order to effectively price its services, and back them with money-back SLAs?
For most data center operators, the answer is no: They cannot accurately predict the performance of a stressed cloud, especially one that spans hundreds or thousands of servers, as well as the requisite storage and networking infrastructure. Add in the additional layers of hypervisors and SDN, and capacity is an educated guess at best, calculated by studying product spec sheets.