by Jason Costa
It’s unclear when exactly self-driving cars are going to hit the road en masse, but it’s going to take some time to get there. What is clear though is that the autonomous vehicle race is going to be one of the most exciting business battles of our time, and it will continue to unfold over the next decade and beyond. It’s my belief that the world is slowly moving from a model of car ownership, to a “Mobility as a service” model. A world in which one can walk outside of their home or office to a waiting autonomous vehicle ready to take them to their next destination. The stakes in winning swaths of that value chain will be huge — and there are already a lot of players vying to win here.
Google’s intentions in the space are intriguing. While Uber’s ambition of making the unit economics per ride work is clear, in my opinion Google’s plan here is about distribution. The ~1 hour a day that the average American spends in the car is one of the last great frontiers in “time availability” for software services to attack. If a human is freed up for an hour to focus on other tasks within the cockpit while the vehicle drives itself, this opens up the possibility of a lot more searches for Google to monetize. Even more important, the car’s cockpit (much like the home) will be a critical battleground for Google Assistant, YouTube, and the entire G Suite of products to gain ground in the everyday lives of users. The car will become a distribution mechanism, much like the phone, for Google to maintain it’s stronghold when users are in need of a digital concierge.
And if I were to guess, Google’s launchpad for such a ride hailing service would live inside of Google Maps. As people bring up the app to plan their next trip, they’ll be able to hail a Waymo vehicle natively within Maps. That would bootstrap the Waymo service with more than a billion users right out of the gate.
Of all the OEMs looking at autonomy, GM seems to be the most aggressive with a vertically integrated approach. Over the past few years, they’ve bought companies across several layers in the stack: Cruise (software controls), Strobe (lidar / perception), Sidecar (fleet orchestration), Zippy (SLAM, computer vision). Other OEMs have made large bets here as well, such as Ford’s ~$1B “investment” in Argo; but GM stands out as the most aggressive. The benefit of being vertically integrated with a technology that’s further out at the frontier is clear: by not having to deal with external dependencies inherent in modular partnerships, a team can accelerate its pace to market by virtue of being “in the tent” together. So in the short to mid term, while this nascent technology has such a high degree of evolving interdependencies, I believe it’s likely to see many of the OEMs try to vertically integrate many of the core technology assets.
The exciting thing about all of this is that the traditional OEMs are not standing still. They’re incredibly savvy players, many of whom have been around for a century or more. As incumbents of industry, OEMs are playing a different game than their record label or handset manufacturer colleagues before them. They’re not willing to be commoditized. The OEMs want to own the brand experience end to end, and none of them are looking to cede the cockpit to Google. Manufacturing cars at scale will be their primary leverage point in doing this, because that’s arguably the hardest part of the value chain — just look at what’s happening with Tesla right now. Expect to see waves of consolidation from the OEMs over the next few years.
Back in 2015 when Uber siphoned off a large cohort of researchers and roboticists from Carnegie Mellon, the outfit illustrated just how focused they were on building their own autonomous vehicle technology. And if there was any confusion on whether Uber is serious about maintaining its own fleet of AVs powered by ATG, that debate is over with their purchase order of 24k Volvos valued at ~$1.4B.
Driving the cost curve down on rides will be imperative to Uber’s quest for profitability. In a sense, making AV work is more of an existential question for the ride hailing service. They *have* to triple down here to control their own destiny. Autonomous vehicles are incredibly expensive, but as transportation transitions from car ownership to mobility as a service the ability to amortize the asset by maximizing usage will become a reality. Uber is going to have these cars out on the road doing as many miles per year as they can — more than 100k+ miles annually. And as they drive the cost per mile closer to $1 / mile, suddenly the unit economics become very interesting.
Tesla’s plans in the autonomous vehicle space are fascinating to consider. The most interesting piece for this player is the data that they’ve collected to date. They’ve got a fair number of cars out on the road, and even if only a small fraction of the overall miles driven in those cars have the Autopilot driving assistance software engaged — that’s still a great deal of data. And with shadow mode enabled, Tesla is able to capture far more data for verification and validation purposes (i.e. what the autonomous car should do, versus what it would do).
What’s unclear is Tesla’s aspirations for autonomy. My sense is that Tesla is really focused on building an aesthetic electric vehicle that fits in with being an energy company (solar panels, battery packs, etc.), and that driver assistance is a nice add on to upsell as part of the vehicle. The company today is still not using lidar, which I suspect has less to do with BOM costs, and more to do with the fact that it ruins the car’s aesthetic in its current form factor. In my opinion though, that’s a signal that L4 capabilities are simply less important to Tesla.
Delphi made noise late last year when the company purchased NuTonomy for ~$450M. This was a strong message that Tiers 1s were not going to be left behind in the race to self-driving dominance. It’s highly likely that these players (Continental, Magna, etc.) consider elements such as software controls, mapping, validation, etc. to be components of the car that the Tier 1s themselves should be producing and selling to OEMs.
There are many layers in the autonomous stack that will likely move to more modularized services sooner. Whoever solves these areas will stand to reap the benefits, and these players could end up becoming a new vintage of Tier 1s in their own right. Some areas of potential opportunity for faster modularization include:
- Vehicle security
- Labeling & annotation tools
- Network integrity
- Verification & validation
- Sensors / lidar
- HD Mapping
- Simulation services
- In- car experiences
Autonomous Systems & other verticals
As discussed above, autonomous systems are coming. And they’re coming to other categories faster than to the road. Ultimately, getting L4 systems on the road is going to be a longer time horizon because the failure case is so extremely high. If something goes wrong with a self-driving vehicle at high speeds, mortality is a very real possibility.
As the technology gets cheaper (lidar costs being cut in half, etc.) and more performant (image classification rapidly improving, etc.), many of the engineering innovations such as sensory, motion planning, and control from autonomous cars are spilling over into other verticals. I suspect that we’ll see a massive acceleration of automation play out in other industries sooner.
These markets will materialize faster because the failure case here is far safer: if a delivery robot accidentally bumps into something while going ~3 mph, a package of glasses might break, but the likelihood of serious injury to a person is extremely low. In fact, I wouldn’t be surprised if many of the software control players going after L4 today move down market into the automated transport of goods for last mile delivery, as that market is likely to materialize before the automated transport of people.
In either case, the self-driving race is going to be incredibly fun to watch play out over the next decade and beyond. The stakes are high, but countless other industry categories are going to benefit immensely from autonomous systems. Hopefully we’ll see a slew of great new companies built in the process. If you’re building one of those companies now, we at GGV would love to speak with you.
Jason Costa is currently a Venture Partner at GGV Capital. This post is part of an ongoing series aimed at exploring topics such as consumer product development, platform analysis, and strategy.