"what exactly the difference between building own datacenter vs renting space in Carpathia/QTS operated DC which already has everything you describe"
Absolutely nothing as long as the DC, connectivity and service migration can be delivered in a similar time frame.
My comments are largely based on IBM's objections/reasons for losing the bidding process, but as Oracle doesn't have existing, large facilities in the same areas as AWS/Azure/Google FedRAMP facilities, it likely suffers the same issues.
JEDI is supposed to deliver ~US$500M/yr savings by bring DoD cloud spending down from ~US$1.2B to ~$US700M per annum - spending the first two years building your cloud service to match existing marketplace offerings significantly reduces those savings, making the bids significantly less competitive.
If this was purely around a single cloud strategy, I would be inclined to agree that it may be problematic, however splitting the JEDI contract between AWS/Azure would likely addresss this and isn't going to make the majority of the disgruntled parties (SAP, General Dynamics, Red Hat, VMware, IBM, Dell and HPE) any happier with what is in effect an exercise in attempting to provide some cost and security controls in an enormous infrastructure.