You're right that CDKTF with dependent stacks is probably better than nothing, but (a) CDKTF's compatibility with OpenTofu depends on a lack of breaking changes in CDKTF, since the OpenTofu team didn't fork CDKTF, so this is a little hairy for building production infrastructure; (b) CDKTF stacks, even when they can run in parallel, still run on the same machine that invoked CDKTF. When you have (ludicrous scale) X number of "stacks", this isn't a good fit. It's something that should be doable in one of the managed Terraform services, but the pricing if you try to do (ludicrous scale) parallelism gets to be insane.