When a project has 200 engineers and 50 in "overhead" (designers, marketing etc.) behind it though, and you break it off - how is that a startup? It's still a 250-person company. I don't see how just cutting it off restores agility that a tiny team with a nascent project has. A large project has more stuff to break with more people that need more links to communicate, increasing the inertia of the whole thing and slowing it down.
What might be possible is some type of Skunk Works with independent leadership, which generates new ideas but leaves the fleshing out to the rest of the company. It wouldn't be associated with longterm management of one project though, more as an incubator and consultancy that spins off and checks in on projects they initiated.