I think there are several factors at play:

First, the $100MM per year Lyft was spending seems like a drop in the bucket to the billions of dollars GM and Waymo are spending every year. Combining with a large automobile manufacturer may have been their best bet to remain competitive.

Although there are two companies testing level 4 technology, we don't know anything about how much time a safety driver is required to take over in a remote operation center. Despite being on the road, for all we know safety drivers are required at least once in the vast majority of trips (thus not saving money or being economical on a large scale).

SF is definitely hard city driving, that doesn't mean it'll adapt well to other locations and climates.

I do think Lyft was likely behind the other companies, but given the slow rate of progress over the last three years, and my own painful development on alternative AI systems, large scale rollout is probably 10 years away.