Earlier this month, GM announced it will be combining Cruise and GM technical teams - purchasing the remaining outstanding shares and marking the end of Cruise’s robotaxi efforts. Cruise had already been facing issues since a pedestrian crash incident in San Francisco from October 2023. I won’t comment on the incident itself, but it marked a decline for the company that GM had to deal with.

In spite of this, Google’s Waymo is growing in usage, though the exact numbers are contested. A partner at a16z has Waymo and Lyft at 22% and Uber at 55% in November 2024 compared to Lyft at 34% and Uber at 66% in August 2023. Lyft’s CEO has them at 30.6% in November 2024 compared to 30.7% in November 2023. Regardless of the exact numbers, the Waymo project has been successful, despite the Cruise shutdown - in California (SF and LA), Waymo reported 20,000 passengers across 13,000 rides in August 2023 and 499,000 passengers across 312,000 rides in August 2024.

Table of Contents

• Self-Driving … In The Future
• Self-Driving Implementations
  • Tesla
  • Waymo
• Economics
  • Scaling Costs
  • Partnerships

Self-Driving … In The Future

Almost 10 years ago, then CEO Travis Kalanick was convinced Uber’s future was in self-driving cars as the ultimate transportation disruptor. Google, Apple, Tesla, Ford, and countless other companies agreed with this vision to replace a million human drivers. Then came the nail in the coffin: Uber sold ATG, its autonomous driving division, 4 years ago. Part of this is due to the current CEO Dara Khosrowshahi focusing on making Uber profitable, or the fact ATG bled money constantly. In my eyes, Uber - the same Uber that startups would reference by pitching them as “Uber for …” - giving up on building self-driving cars indicated it was a pipe dream.

Elon Musk has tweeted many times for over 10 years with varying predictions for full self-driving capability timelines. Despite this, Tesla Autopilot and Comma openpilot are impressive, but keep responsibility with the user (more to come on these systems). If a driver has to keep looking at the road, it doesn’t align with my vision of a fully autonomous, self-driving car.

This is why the most fascinating piece, for me, is how regular Waymo’s success is. Self-driving cars used to be a clear sign of the future that no one believed would actually happen. Yet Waymo is just another thing that happens every day. No one gives it a first glance on the cloudy streets of San Francisco.

I believe there’s a few components for why Waymo’s success is normalized:

1. Tesla set expectations high with Autopilot
2. Waymo is still restricted to certain areas (due to mapping requirements)
3. Google is expected to win

Self-Driving Implementations

Tesla

Tesla Autopilot uses 8 cameras and radar (re-introduced with HW4). around the car - notably, this does not include Lidar sensors. In combination with publicly available maps, this hardware is sufficient for route planning “in the future for full self-driving in almost all circumstances”. These assist the driver via autopilot (AP) and full self-driving (FSD) capabilities, which control the car to some destination automatically. The driver is required to pay attention to the road, else the system disengages. There are some restrictions on what regions AP and FSD are available for, but it’s quite large.

Waymo

Waymo is similar, but has two key differences:

1. A custom mapping system - why Waymos are restricted to certain regions
2. More sensors around the vehicle

Waymo leverages a custom 3D map of geofenced areas to drive. Vehicles are able to adjust to new changes, such as lane closures or new traffic lanes. Waymo is able to use their fleet to ingest these changes into their mapping system. When a vehicle drives, it matches what sensors see with these maps to get better predictions for when other vehicles are stopping, whether there is a mannequin or a pedestrian, or if the curb cut is actually a driveway.

Waymo vehicles also use more sensors around the car: 13 cameras, 4 Lidar, and 6 radar units, reduced from the previous version. These vehicles even have auto-cleaning equipment for sensors on board.

Economics

Scaling Costs

Uber (and Lyft after it) create a dynamic pricing market: when demand spikes, prices go up, encouraging more drivers to take rides from that area (increasing supply, and decreasing prices). This automatically scales demand and supply to some near equilibrium. Since Uber does not pay any costs when vehicles are not being actively used (given a consumption-based, not time-based payment model), this works out great for them.

Waymo, however, owns its vehicles. These are nice Jaguars too - probably chosen to give riders a feeling of luxury for a futuristic feel. Waymo has a fixed cost with operating and maintaining these vehicles, which leads to a challenge: how many vehicles should they deploy? There is no answer, as long as they own. The Waymo Driver, as software that can drive (theoretically any vehicle) instead of the actual hardware, potentially allowing Waymo to leverage a dynamic fleet. The important assumption is that someone else is owning these vehicles, but who would want to do that?

Tesla took a different approach for ride-hailing to use consumer vehicles with their robotaxi CyberCab concept. Their existing vehicles sold to consumers have driver assist software, but these vehicles have no steering wheel or pedals. Tesla will operate these CyberCabs and owners of regular Tesla can deploy their vehicles into this network to get extra money, with no geographic limitations (due to no need for 3D mapping). This creates an elastic supply adjusting to demand, where users would give up their cars during popular hours for more money. The markets are convinced this will happen soon.

Partnerships

Waymo has already partnered with Uber in Phoenix for rides and Uber Eats. In early 2025, this partnership will expand to Austin and Atlanta for rides, which are booked directly via the Uber app. This was Uber’s plan when selling ATG - wait for someone else to figure out self-driving (and light money on fire in the process), then leverage the Uber network to charge commissions. Uber still has options, but the key is to make sure self-driving taxis operate through their network to maintain their market stronghold and network effects.

Personally, I see Waymo continuing to improve economics and grow to new areas, leveraging partnerships to get riders. Waymo still has its own app, Waymo One, in addition to using the Uber app (for Austin/Atlanta). Users on Uber get to pick a human driver, a Waymo, or even Lime scooters. The Waymo app has no options: it’s just Waymo - better for their revenue.

The real test is whether “waymo”-ing around catches on. No one says “I’m gonna Lyft somewhere”; it’s always “I’m gonna Uber, I’ll check Lyft”. The first mover advantage is real. Until this happens, self-driving is still in the future, and will continue to impress me.

Image sources: 1 2 3 4 5 6 7