Uber, Tesla and Google have spent billions on autonomous vehicles (AVs) over the last few years. So have the traditional automakers, their suppliers and other startups in Silicon Valley and beyond. 

But why has more than $80 billion been pumped into this race? What is the future of the self-driving car, and how will it impact people’s lives and cities? 

Who will ultimately win this game?

AVs can bring several benefits to society. First, the safety argument. More than 1.3 million people die every year in road traffic accidents, according to the World Health Organization. More than 90% of those deaths are due to human error and a significant percentage due to driving under the influence of alcohol. By eliminating the human component of driving, AV could reduce the risk of accidents and improve safety on roads, assuming that engineers manage to solve even the most challenging cases of AV deployment.

Second is the argument that AV could greatly expand mobility, either by giving mobility to people who are physically unable to drive or by accelerating the adoption of on-demand car services such as Uber and Lyft. According to some estimates, removing the drivers could reduce the ride costs by 70-80%. 

Ultimately, these trends will change who buys cars — and what they look like.

A world in which on-demand cars operate at scale leads to many second-order effects. 

Parking, for example, will never be the same.

Vehicles today are parked more than 90% of the time, and the average driving occupancy is little more than one passenger per car. If most individuals no longer own cars and if cars do not sit idle waiting, cities and commerce would be transformed.
What happens to traffic overall — does it get better or worse? Would we shift from public transportation to cars if roads start emptying out? What happens to retailers and home prices if you no longer need to build an underground parking lot? Will insurance premiums go down as accident rates fall? Would people live further away from large cities if commuting becomes seamless? What will happen to car and truck driver jobs? 

But where does the VA  industry stand at the moment, and how far are we from ‘full autonomy’? 

There are five different levels of autonomy and we are at Level 2 for commercial purposes. Level 2 essentially accelerates and slows down the car and stays within the lane, but the driver still needs to have his/her hands on the steering wheel. Level 3 does the driving for you, but it requires the driver to be ready to take control. Level 4 will drive for you in certain situations but not others. Level 5 will not need a human driver and will not have a steering wheel.

There is still a lot of disagreement amongst experts in the industry as to when Level 5 will become a reality. Enthusiasts believe that in the next 5 years full autonomy will be ready to be deployed at scale in certain areas in the US. However, the majority believe it will take around 10 years or more until Level 5 hits large cities. Skeptics are confident that the future of AV is the same as that of flying cars.  

Google started researching AVs 10 years ago and still leads the race. 

In December, Google opened Waymo One (its self-driving taxi service) to a select group of residents of Phoenix, Arizona.  The cabs operate mostly with a safety driver behind the wheel who is ready to intervene whenever needed, reinforcing the perception that fully autonomous Level 5 is still a long way from being a commercial reality. 

Google now has more than 10 million miles driven, while Uber, the runner up in the race, has only a fraction of that. 

But if the future is so uncertain, why is there so much capital being invested in AVs in what seems to be a race against time? 

Industry experts believe two opposing scenarios will be the most likely outcome of the AV race.

Either automakers will be able to acquire an ‘autonomous vehicle package’ from OEMs, similar to what happens today when Volkswagen buys a fuel injector from Bosch, or Google and Uber will be the sole providers of such technology. In such a scenario this duopoly would dictate the rules of the game and capture a disproportionate share of value in the auto industry. Detroit would be done (again).

The value capture potential is substantially greater if AVs come to fully replace driving.  The winners would eat taxi and truck drivers’ lunch and dessert by selling subscriptions to ‘mobility as a service’ plans for individuals who no longer own cars and use their driving time in a more efficient way instead.  The Union Bank of Switzerland estimates that by 2030 the self-driving car industry will be worth $2.3 trillion. Intel says the market may reach $7 trillion by 2050.  

There is little debate that market control will not come from the hardware and sensors, where there may be strong manufacturing scale effects but no network effects. 

The key to the power shift will be data. 

The gold mine in the value chain will be in owning the software and the mapping and driving data, which go hand-in-hand. 

The more data points on mapping a company has, the more accurate is the information given to the autonomous car, which translates into fewer unknown situations and a safer ride.  Therefore, the more miles driven and simulated, the more updated and precise the system becomes. 

In addition, after understanding its surroundings, the car needs to work out what to do based on what other cars and drivers are doing. Here again, the more situations the system has seen, the more the machine learns and the better the software reacts.

So it seems that accumulating as much data as possible, as quickly as possible, is the secret to building strong network effects and creating an insurmountable barrier of entry for competitors.   

But although this argument makes logical sense, and to a large extent explains the race against time, there may be a fallacy to it. This fallacy would eventually make the power dynamic pendulum swing back to the scenario in which several manufactures can supply an ‘autonomous vehicle package’. 

First there is a question of who owns the data. Automakers have argued that they are the ultimate seller of the vehicle and therefore they should own the data. If that view prevails, instead of only Google and Uber controlling the data, ownership would be fragmented, thereby reducing network effects.  

Second, data value is not linear. There might be a point at which adding more data to the algorithm does not make any incremental improvement. This is a common topic in machine learning and, as engineers continue to develop the field, it might be that the amount of data needed to have a safe AV system may decline substantially. Then the question becomes how many people can get that amount of data. If the answer is ‘several’, then the argument to pour billions into R&D in an attempt to be the sole winner falls apart.  

It is still very early days in the AV industry, and the answers to most of these questions remain fluid. The pace of development will depend on how fast computer science problems get solved, and it may take decades until AVs cycle into the entire global fleet of more than 1 billion cars. 

Autonomy will also have many different shapes and forms, depending on complexity. The Bay Area, Manhattan, São Paulo, Beijing and Bangkok will have very different needs and complications and each will have a different model. 

Despite all these uncertainties, we know one thing for sure: the outcome of the AV race will have very profound consequences. It will not only change the power dynamics in one of the largest industries in the world. It will change the way we interact with our cities — and our lives.  
Paulo Macedo is an investment professional at 4×4 Capital and a Stanford Graduate School of Business alum.