Funny thing: It turns out that Warren is not the only one doing wild future robotaxi valuations.
James Douma also has wild ideas about Teslas FDS/robotaxi valuation.
Couched in more cautious terms, and approached indirectly, but ... due to James science cred/background in programming, experts systems and AI, it is very impactful.
He does not provide a classic model, but a napkin model.
He refuses Dave's call for sharing his model, but leaves it to the audience to guesstimate based on his napkin math.
Spoiler-alert: The upper bound of James' estimate for robo-taxi earning is so high that he has is seriously considered sort of phase-change effects regarding that a such enormously successful future Tesla may have to change into another type of organization because it would at that point command the resources of several large and wealthy countries.
Above video is sort of the 'conclusion' or take away of a longer series of technical videos, where James is lecturing Dave and the audience.
If you don't have 1-2 day, the video below is perhaps the best: Condensed, and good explanations.
I haven't done any time markings on this, but the gist of the argument is the following.
We don't really know for sure that FSD is solvable. There still may be some corner cases so far unseen which may turn out to be intractable. Or really time consuming.
FSD It most likely solvable with camera only: What Tesla has shown indicates that FSD can be solved with cameras only with a high degree of certainty. Tesla now have pseudo-lidar down pat, called 'birdseye', based on either time-fusion of a single camera or multi-camera fusion - or both (perhaps with inter-relational verification)
So, rephrasing: It is not entirely a given that FSD can be solved, but if it can, camera-only works.
Dave tries to raise questions about competitors and poke holes in James' 'Tesla-pole-position'-arguments, but James very succinctly details how and why Tesla can scale easily and cheaply, and Waymo cannot.
Off course they can choose to scale - slowly and expensively. But they will be very vulnerable because Tesla has tremendous pricing power. James has Tesla's long term cost per mile at ~10 cent - all included.
Dave also repeats his sort of desire to still buy a nice Tesla FSD as a private person, but James gently talks him down by pricing FSD as being initially in the 100K-500K range - per car.
(Dave had the same discussion with two youtubers, Zac and Jesse ~two months ago where he sort of insisted and persisted that Tesla could and should sell him FSD at around 10K, even though they could earn 10-25X of that by keeping the car. I like Dave a lot, but that was not his shining hour.)
Short version:
Even though we don't know for sure if FSD can be solved, if it can, camera-only is enough (If you have very good sensor fusion AND a f**kton of data)
FSD may be solvable with lidar and camera also. He politely does not overly criticize lidar-driven solutions, saying there are good historic reasons but his statements about lidar solving for static object and position are damming, but not helping with moving objects and people - precisely 'cause he is so polite.
TLDR:
0. FSD may not be solvable - we cannot be totally sure - yet.
1. If Waymo and other companies with lidar-driven solutions can solve FSD then so can Tesla. Lidar == Pseudo-lidar.
But the converse is not true: If Tesla can solve FSD using only cameras then it does not follow that Waymo and other lidar-companies can also. (Off course they can in the primitive sense that they can just ignore all lidar input but that has
not been their claim or strategy)
2. When Tesla (as is likely) solves FSD they will in the upper bound reach large-country-level wealth. Especially if the solve it first, and use the proceeds to consolidate their position by extreme and rapid scaling.