We are also in very early stages about replacing vRA with possible Terraform. But probably not in the same matter.
As a solution we require the users (developers+Support) to have no understanding of IaaS, they only want their machines/databases etc with a touch of a button. Also we have no public cloud (AWS/Azure etc). Futher more we basically use vRA as a frontend for deploying and assigning IP's. (Everthing is done in vRO worfklows based on user selection in vRA)
The trouble we see ahead is vRA 8.0. It's the second time VMware breaks almost everything (vRA 6.x to 7.x was the first). We have spend alot of time developing vRO workflows witch needs to be redesigned. And vRA 7.x will go out of support in a little over a year. (This you really need to consider, if you you want to go with Terraform and vRA 7.x!)
So we are are looking into the following solutions at the moment:
1. Replace vRA with vCD and (possibly) Terraform (keep using vRO)
Much more expenisive solution, however this mean we can keep alot of the vRO workflows. We can use vCD as frontend and it's more flexible than vRA. Futhermore it should be able to give us addition features when it comes to vRO (RabbitMQ/vCD can trigger alot more events than vRA/EventBroker). It will also prove true multitenancy (not the "fake" one vRA has.)
2. Create everything again in vRA 8.0
Alot of work, and we have no idea what will happend with vRA 9.0, experience tells us it will not be compatible...)
3. Going with Foreman/Ansible
Loses alot of features, alot to redo, but cheaper.
4. Finding another product (Resolve, BMC, Hiro etc).
So far we have not decided, but we are doing PoC on Soltion #1 and #3.