https://investor.uber.com/news-events/news/press-release-det... shows Uber generated $65B gross booking in 2019, and we know substantial fraction of that goes to the driver.
https://www.sfchronicle.com/business/networth/article/For-a-... says driver takes about ~60% of the booking.
So this is a question of whether the capital cost / the engineering cost / operating cost for autonomous vehicle is higher than $65B * 0.6 = $39B per year for the same number of rides.
The engineering cost will be amortized over long enough time that it's likely not a significant factor in the long run. So, the limiting factor will be the capital cost (for vehicles and its support infra) and the operating cost. As you say, at this point we don't know for sure - one thing we know is that the capital cost and the operating cost likely will see significant reduction thanks to the economy of scale.
if FSD actually ends up working on most roads, I don't see why not. there's no inherent reason why engineer headcount would scale linearly with the number of cars.
as for the cost savings, they don't necessarily have to beat uber's pricing today. they could alternatively try to drive uber's costs up by lobbying hard to get drivers classified as employees and catch some PR points at the same time.