In this case (and the same in Australia, UK, Canada) it will be cost of procurement. Ie individual departments have to negotiate (and maintain) a contract. Providing a generic platform (Cloud Foundry, k8s, whatever) on top means that small products from individual entities can get built/deployed without each project having to do a tender to negotiate Heroku or similar.
Likewise, getting a .gov (or .gov.uk or .gov.au, etc) is a big hassle, and with a PaaS you can make the experience a lot better (not in use any more but you can see what it looked like: https://github.com/cloud-gov/cf-domain-broker)
Ie the goal is to reduce the time/steps/approvals required to get a gov website up and running, and cloud.gov in this current incarnation was started under the Obama admin.
Most of it is open source on github https://github.com/cloud-gov
(and exploit the shit out of you until the gov't stops us)
