Thanks Benjamin! Really appreciate your kind words! Already working on 2 new deep dive articles - the VMS software market and the optical networking industry.
Really enjoyed this. The “audit the silicon” framing makes Uber feel much closer to a real-time logistics utility than a taxi app.
Also agree with the key nuance: H3 or transformers aren’t the moat by themselves, it’s the compounded loop of marketplace liquidity, spatio-temporal data, and operational ML that’s hard to reproduce city by city.
Thanks, Manu! The graphics are a time-sink, but essential. I’m trying to strip away the noise to get to the signal, so I’m glad to hear they’re helping visualize the architecture without getting bogged down in the weeds
Great article. I agree with your unit economics logic, hybrid model (AV + Real drivers), but not the long-term moat.
If Waymo scales up 5-10x and unit economics improve, Google can build an app on par with Uber or integrate it into Google Maps, as you have mentioned. There is no reason for them to pay Uber for software, and I believe Google makes better software. I guess Google will also be collecting AV data religiously. Once they have meaningful data for a few years, this is not a difficult problem for Google to solve. When the upfront investment to build a dedicated application/service makes financial sense, Google will do it.
Ride matching for AVs is a simpler problem than ride matching for actual drivers because there is no auction, and there is no fraud check. They will simply offer their app in parallel to Uber for Waymo-only bookings. Even if Google makes a slightly inferior application, it can undercut the price because it owns the app, Google Maps, and the car. It also has location data and travel history for tons of users.
Uber will still not go anywhere anytime soon, but it is a bit difficult to guess what will happen if Waymo scales up meaningfully. Users don't really care if the algorithm is greedy or sophisticated or slightly less sophisticated, as long as the service is decent and they get cheaper fares.
I am not talking about anything in the antitrust lawsuit zone, such as providing slightly inferior Google Maps to Uber or charging more for Google Maps API services, etc.
The original research paper for the transformer architecture by Google Employees(2017). So Google is more than capable of replicating Uber's technical solution once they have enough data, assuming the data from Google Maps and users is not enough.
Amazing article! Can’t wait for the upcoming pieces.
Thanks Benjamin! Really appreciate your kind words! Already working on 2 new deep dive articles - the VMS software market and the optical networking industry.
I liked the writeup but one clarification, Cruze shuttered it's robotaxi fleet
Love Uber too! Great pick
Really enjoyed this. The “audit the silicon” framing makes Uber feel much closer to a real-time logistics utility than a taxi app.
Also agree with the key nuance: H3 or transformers aren’t the moat by themselves, it’s the compounded loop of marketplace liquidity, spatio-temporal data, and operational ML that’s hard to reproduce city by city.
Great work. High-signal piece.
Excellent piece my friend!
Thanks! Glad you enjoyed the read!
Amazing article and amazing graphics to go with it.
Thanks, Manu! The graphics are a time-sink, but essential. I’m trying to strip away the noise to get to the signal, so I’m glad to hear they’re helping visualize the architecture without getting bogged down in the weeds
Exceptional analysis and clarity of thought. Hat tip for the root/* structure.
I would really like your analysis around Ubers network effects/externalities and how you see those going forward
Great article. I agree with your unit economics logic, hybrid model (AV + Real drivers), but not the long-term moat.
If Waymo scales up 5-10x and unit economics improve, Google can build an app on par with Uber or integrate it into Google Maps, as you have mentioned. There is no reason for them to pay Uber for software, and I believe Google makes better software. I guess Google will also be collecting AV data religiously. Once they have meaningful data for a few years, this is not a difficult problem for Google to solve. When the upfront investment to build a dedicated application/service makes financial sense, Google will do it.
Ride matching for AVs is a simpler problem than ride matching for actual drivers because there is no auction, and there is no fraud check. They will simply offer their app in parallel to Uber for Waymo-only bookings. Even if Google makes a slightly inferior application, it can undercut the price because it owns the app, Google Maps, and the car. It also has location data and travel history for tons of users.
Uber will still not go anywhere anytime soon, but it is a bit difficult to guess what will happen if Waymo scales up meaningfully. Users don't really care if the algorithm is greedy or sophisticated or slightly less sophisticated, as long as the service is decent and they get cheaper fares.
I am not talking about anything in the antitrust lawsuit zone, such as providing slightly inferior Google Maps to Uber or charging more for Google Maps API services, etc.
https://arxiv.org/abs/1706.03762
The original research paper for the transformer architecture by Google Employees(2017). So Google is more than capable of replicating Uber's technical solution once they have enough data, assuming the data from Google Maps and users is not enough.