[TradingPlaces]

It definitely works. I was in one just yesterday for over a half hour and it’s a pretty smooth thing, and in some ways better than a human driver (e.g., always yields to pedestrians entering a crosswalk, never exceeds speed limit,etc.)

But both Waymo and Cruise lose money on every ride. All the sensors and on-board compute are expensive, and also the R&D, and the 3D maps. Both services rely on detailed 3D maps of the areas they service, and are geofenced to that.

So right now, it does not scale yet until they can bring down the costs, and/or reduce reliance on those maps, which take a long time to make.

Videos of my rides: https://youtu.be/q-l23cMs1Fs?si=R1hjPQokqAMCff_T https://youtu.be/FZTWxDe1i6g?si=mGbcixGTKCQchnVv

[0xDEAFBEAD]

>All the sensors and on-board compute are expensive, and also the R&D, and the 3D maps. Both services rely on detailed 3D maps of the areas they service, and are geofenced to that.

I think we need to split this into fixed vs variable costs. It's hard for me to believe that the sensors+compute is more expensive than a human. Those sound like the variable costs. I'd be surprised if selling an additional ride causes e.g. Cruise to lose a bit of money, but I'd love to be corrected.

[TradingPlaces]

I have been told it loses money by Google employees. Not just sensors+compute. R&D and the map-making are substantial costs. Waymo is reported in “Other Bets” which had $1.1 billion in revenue in 2022 and $6.1 billion in operating losses.

[JeremyNT]

> R&D and the map-making are substantial costs.

But that's largely a sunk cost that (in SF anyway) has already been spent, right? It's not like that cost goes up on a per-ride basis.

For places where that's already done, you only need to account for the cost of sensors/compute/maintenance of the vehicle themselves, which seem (intuitively) like they would be lower than the expense to users. Which would mean that a ride is net positive.

Maybe it's the case that configuring and maintaining the cars is incredibly expensive. That would be the only way I could envision actually losing money on a "per ride" basis.

[flutas]

> But that's largely a sunk cost that (in SF anyway) has already been spent, right?

For mapping, no.

Waymo maps the city constantly with every car, and they have a mapping team that reviews those changes before they go to the other cars. So ongoing cost of the human side is a thing (for now?).

They also have a team of "remote operators" that watch the fleet in real time in case they have a panic attack over something, so more ongoing cost that's not just sensors/cars.

[TradingPlaces]

There are substantial non-cash costs that go into cost-of-revenue. These come from amortizing assets like patents and the maps, and depreciation on all the equipment, including whatever backend servers there are. Share-based compensation is typically also a big one. They are "sunk costs" as far as pure cash accounting goes, but that’s not how GAAP accounting works.

[TradingPlaces]

And also, for whatever reason they decided to go with the Jag I-Pace, which starts at $72k

[jhp123]

there is also a team of human remote operators.

[gghffguhvc]

If they could 100x the number of cars and trips in SF then the cost of maps per ride would be significantly lower.

[TradingPlaces]

There are only 808k people in SF https://www.census.gov/quickfacts/fact/table/sanfranciscocit...