GOOGL: Waymo is worth $75B+ Right Now

sagawa
Print Friendly, PDF & Email

SEE LAST PAGE OF THIS REPORT Paul Sagawa / Tejas Raut Dessai

FOR IMPORTANT DISCLOSURES 203.901.1633 /.553.9827

psagawa@ / trdessai@ssrllc.com

twitter.jpg @PaulSagawaSSR

December 13, 2017

GOOGL: Waymo is worth $75B+ Right Now

GOOGL’s Waymo has a 3-year lead on autonomous driving technology and is poised to begin rolling out fleets of on-demand robo-taxis for commercial services market by market across the US. We believe that the addressable market for subscription TaaS could reach $121B in the US and $200B worldwide by 2030 displacing the all-in costs of primary or secondary car ownership for some consumers, while offering substantial benefits (convenience, safety, time, etc.). Rivals will move to close the gap and may have advantage (political, localized mapping and testing, etc.) in certain geographies. Still, Waymo’s substantial technical lead, extensive testing and simulation, digital mapping prowess, consumer reach, emerging ecosystem, and go to market strategy are powerful assets that give it the likelihood of market leadership and the possibility of dominance in the developing opportunity. We do not believe that GOOGL’s share price or the expectations of its investors reflect the $75B+ value of this potential.

  • Autonomous TaaS addresses $285B in US transportation spending. All in, Americans spend $931B annually on personal transportation. 64% of personal travel is local and practical – commuting, shopping, errands, etc. – addressable by services offering on-demand rides. The potential for replacing primary car ownership for urban households and secondary vehicles for suburban ones could as much as half of US personal transport spending in play over the long term. Of course, we would expect autonomous TaaS service to be highly deflationary, but a US market of $121B seems approachable by 2030, with obvious further opportunity internationally.
  • 3 main technology elements to autonomy. We have written extensively on self-driving vehicles (Autonomous Trucks: Self Driving Convoys are YEARS AwayBringing TaaS to the MarketSelf Driving Ambition). Successful solutions must a) deliver a detailed, real-time 3D model of the car’s environment; b) predict the most likely movement of all the objects in the map; and c) make and execute decisions to conduct the vehicle safely and expeditiously toward its destination. We believe solutions engineered specifically for full autonomy will deliver much better performance, much sooner, than incremental solutions based on improving autonomous driver assistance features.
  • Geofenced robotaxi subscription service is the most viable model. Self-driving systems benefit greatly from extremely detailed and accurate digital maps – e.g. showing the exact height of curbs, updated with temporary signs, etc. Restricting autonomous operation within the bounds of such maps greatly lessens the scope of information to be collected by sensors in real time and improves the decision making of the autonomous control system. Given that most personal transportation traffic is local, an on-demand service limited to specific geographic boundaries would still be well suited to many households. Because AI assistants, like Alexa or Google Home, will make it easy to arbitrage rates for ad hoc travel, we believe TaaS will gravitate toward subscription pricing, well matched to the needs of commuters and family errands.
  • Waymo has substantial advantages in developing an autonomous TaaS business. GOOGL has been working on self-driving for nearly a decade and has logged 4 times more autonomous test miles than its closest rival. Those test miles have also been compiled under a significantly more diverse set of driving conditions, have been supplemented by staged testing for unusual conditions and by billions of miles of simulations. Waymo’s California testing shows necessary human interventions nearly 2 magnitudes less frequent than others, despite its far more rigorous testing regimen. GOOGL also offers unparalleled AI scientific talent, industry-best digital mapping, significant development leads in both hardware and software elements of its solution, an emerging ecosystem of partners (CARS, AN, FCA, Lyft, etc.), reach to millions of consumers, and other valuable technical/business assets.
  • Waymo is ready to launch now – years ahead of rivals. Waymo has begun providing fully autonomous rides to customers in a Phoenix, AZ suburb and expects to expand the service to most of metro Phoenix within the year with as many as 600 minivans. Uber offers commercial trips with two company engineers in the front seats monitoring their system. It withdrew from testing in CA, objecting to the requirement that statistics on human intervention be openly reported. GM plans to begin similar testing in NYC next year, under conditions that require a paid police escort for each vehicle. Start-ups Drive.ai and Nutonomy will also begin limited testing with Lyft, working with tightly constrained geographies and engineers upfront, ready to intervene if necessary. We believe these would-be rivals are more than 2 years behind Waymo, which had logged 1M autonomous miles by June ’15.
  • Self-driving TaaS subscriptions will have big first mover advantage. Waymo’s lead gives it several major market advantages. 1. Commercial operations will add to its huge base of driving data, with new emphasis on circumstances common to “ride share” – e.g. passenger directed routes, poorly defined pickups, etc. 2. Waymo can refine its digital maps in each market ahead of rival entry. 3. Waymo can draw ideal partners (e.g. CARS, AN, etc.) to its ecosystem. 4. Waymo can integrate TaaS to its AI assistant. 5. Waymo can open negotiations with new municipalities with a track record. 6. Waymo can position its brand with consumers.
  • GOOGL’s share price does not adequately reflect the $75B+ value of Waymo. We believe that GOOGL could take more than half of a US TaaS market that could reach $15B by 2025, with runway to grow to $121B or more by 2030. Discounting a 14.2x multiple of projected 2025 sales back to current dollars at a 10% discount rate yields a $75B valuation for Waymo. Note that this valuation does not ascribe value to international opportunities for Waymo, which are substantial despite hurdles in translating its US success, or for adjacent markets, like trucking or local delivery. GOOGL’s forward P/E, adjusted for cash and for the value of Waymo, is 19.6x, modest for a company growing sales at 21% and projected to sustain 19% growth over 5 years. Its PEG ratio of 1.02 is lower than all its mega cap peers (also cash adjusted) but FB, including AAPL. With additional unmonetized value in enterprise cloud hosting, AI Assistants, and other opportunities opened by its dominant consumer reach, hyperscale platform, and AI skillset, we see GOOGL as significantly undervalued, with 30% or more upside.

Forgot About Waymo …

Autonomous on-demand Transportation-as-a-Service is huge opportunity. Americans spent $1.23T in 2014 on personal transportation, of which we believe at least 25% could be addressable by self-driving TaaS. However, with automakers and startups tripping over themselves to promote their various autonomous driving projects, a cottage industry of skeptics has risen to poke holes in balloons full of hot air. As a result, a perception that self-driving has crested the Gartner hype-cycle on the way to the trough of despair has grown increasingly common. The skeptics are sleeping on Waymo.

GOOGL began investing in self-driving nearly a decade ago and has logged more than 4 times more autonomous miles than the next competitor. These miles have been driven in deliberately diverse circumstances, and particularly vexing conditions were recreated at a private testing facility to gain more data and then run thousands of times in sophisticated computer simulations. No one else does this to this degree. In testing on California roads in 2016, Waymo reported 20 times more miles than its nearest rival and delivered a rate of human intervention 2 magnitudes better than the next best, despite testing under far more rigorous conditions. Waymo has the most and best scientists with the deepest experience addressing the three main problems of self-driving: a) delivering a detailed, real-time 3D model of the car’s environment; b) predicting the most likely movement of all the objects in the map; and c) making and executing decisions to conduct the vehicle safely and expeditiously toward its destination. It can already leverage the most detailed, accurate and extensive digital maps in existence.

While would-be rivals like GM, Ford and Delphi are contemplating live tests for 2018, with real customers on real roads but with engineers behind the wheel poised to take control, Waymo is already moving on to offering a commercial TaaS service in Phoenix (at first with observers in the BACK seat) with as many as 600 Chrysler minivans. It has lined up CARS to handle daily servicing for the cars and AN to manage repair and maintenance. It may create its own app to interface with customers, or it may use Lyft, where it has a partnership and an ownership stake. In either case, we believe hailing cars will become a utility, accessed with a simple request for a ride from the device at hand – a huge advantage for GOOGL with its Android platform and its Assistant AI interface.

Expect Waymo to roll out its on-demand TaaS as a subscription service (to avoid ride-by-ride rate arbitrage) geofenced to areas where it has completed detailed upgrades to its maps. As the maps expand, so will the service area, pulling in new neighborhoods and better facilitating commuting. Once Phoenix is running well, expect a launch in another city – perhaps Austin, where Waymo has already been testing. With good results – demonstrating safety improvements, access for the elderly and shut-ins, cost effective service for the disadvantaged, reduced demands on public infrastructure, and enthusiasm from citizens – new municipalities will sign on. And so on, and so on, and so on.

We believe Waymo’s technical lead, business head start and developing track record will give it big advantage over would-be rivals, and expect it to capture at least half of a US market that we believe could be $121B by 2030. There are other considerations internationally – very different driving conditions will require retraining models – and Waymo is unlikely to establish such dominance, but it will compete in those markets. Taking a 14.2 x multiple of projected 2025 sales of $11.4B, when we expect 100%+ annual growth from Waymo, suggests a $75B value for the business today. This is NOT reflected in GOOGL’s share price.

Getting Up to Speed on Autonomous Driving

We have written extensively about autonomous driving (Autonomous Trucks: Self Driving Convoys are YEARS AwayBringing TaaS to the MarketSelf Driving Ambition). Our research divides development of self-driving systems into two basic approaches: 1. Incremental autonomous driver assistance systems (ADAS) that enable functions such as hands-free cruise control or self-parking, with the intent that the increasing sophistication of such systems might eventually achieve fully driverless operation; and 2. Disruptive vehicles designed to be entirely autonomous from the start (Exhibit 1). We believe that the second approach has a much faster path to full autonomy. Incremental approaches have forced car makers to technical choices that are expedient to sell cars now (i.e. eschewing high priced and aesthetically challenging LiDAR sensors) but that compromise progress toward full autonomy. Furthermore, maintaining the possibility that a human user might take control requires extensive development and testing to be sure that the transitions can be handled smoothly and safely, a functionality unnecessary for a disruptive vehicle intended to be fully autonomous from the start.

Exh 1: Two Roads to Level 5 Autonomy

We divide the development of self-driving systems in to three parts (Exhibit 2): 1. 3D mapping – The system must construct a wide, detailed, accurate and real-time 3D digital map of the vehicle’s environment. It does so by fusing the inputs from a variety of sensors (e.g. LiDAR, cameras, RADAR, ultrasonic, etc.) with static digital maps to give the system the ability to identify and place everything that might interact with the vehicle. 2. Prediction Engine – The system then predicts what is most likely to happen in the immediate

Exh 2: Self Driving Technology Elements

future and what could happen to create a risk of an accident. Cars in the left lane will likely turn, but could abruptly pull out when the human driver realizes that they are in the wrong place. 3. Autonomous Control – The system takes the map and the assessment of the likely and possible movements of the various objects and decides on the best course of action to safely continue its course toward its destination.

All three elements rely on complex and powerful AI-based systems to operate, although the solutions are quite different from one another and each has different obstacles to traverse on the way to superhuman performance. The quality of a 3D mapping system depends on the mix of sensor data being fused to create the map, but also on the detail available from static map databases (Exhibit 3). Each sensor class has its own strengths and weaknesses – camera images show details but not precise distances and rely on available light, RADAR offers accurate distances and speeds in any light or weather, but only on hard objects, LiDAR picks up soft objects as well as hard, but can be foiled by weather, etc. Fusing the inputs involves relying on the best aspects of each and combining the information into the best possible 3D picture of the vehicle surroundings. This is very difficult and it must be done in real time with sub-10ms lag. If the AI system can reliably depend on the digital map for information like curb heights, the meaning of signs, temporary restrictions, and other important, but often NOT included, minutia, it will not have to figure it out for itself, saving precious time. This is an enormous shortcut to autonomy in geographies where such detailed maps have been produced, and the reason why the first commercial implementations of TaaS will be strictly geofenced.

The Prediction Engine demands comprehensive data. If an autonomous car reaches an 8-lane round-a-bout traffic circle, it needs to understand how human drivers behave in navigating such a traffic configuration. It needs to know how bicycle messengers weave between cars and cheat on red lights. It needs to know that if a ball rolls into the street, that the child on the sidewalk might run out to get it. It needs to know that manual transmission cars on a steep hill often roll back before moving forward. This takes lots of miles, but can be supplemented by setting up troublesome situations on a test course, and by simulating those situations with a computer subtly changing the parameters to cover as many bases as possible over

Exh 3: Disruptor Systems are the Most Comprehensive

thousands of virtual test runs. The prediction engine is also location and culture specific – the streets of San Francisco are very different from the streets of New York, and the drivers in Arizona are not the same as the drivers in London or Bangkok Thailand. This means each new market requires tweaks to the prediction model – some tweaks more serious than others.

Finally, the autonomous control system does the driving, constantly taking input from the other two systems to make optimized decisions based on the learned experience buried in its code. More miles driven under more diverse circumstances, including prep for unusual but dangerous situations like erratic drivers or inclement weather, add to the range and appropriateness of chosen maneuvers. Simulation also plays a key role, nudging the system to mastering complex behaviors like navigating traffic circles or reacting to emergency vehicles amidst erratic drivers. While safety is the highest priority, less advanced autonomous control systems are typically so cautious and deferent to other vehicles that riders are frustrated by delays. Sometimes, a driver needs to pull out into traffic with the faith that the other drivers will accommodate them. This is a hard skill for computers to master, but with practice and guidance, it can be learned.

Fleets of Geofenced Robo-taxis

The US Highway Traffic Safety Administration published a paper in 2013 which defined 6 levels of automation (Exhibit 4). Level 0 indicates that the driver is always in full charge of the dynamic tasks of driving, even when the car is enhanced by warning or intervention systems. Almost all cars on the road today are Level 0. At the other end of the spectrum is Level 5, where all the dynamic tasks of driving are under full control of the autonomous system, with no requirement for human intervention under any circumstance. These are the vehicles that can safely shuttle children or the elderly, can allow a commuter to work productively door to door, and that get drunk passengers home in one piece without a designated driver.

Exh 4: Levels of Driving Automation for On-road Vehicles

We noted earlier that we expect that disruptive vehicles, designed from the start to be fully autonomous, will get to Level 5 long before the incremental approach of adding autonomous capability bit-by-bit will put fully self-driving cars in private garages. The reasons are technical and practical – incremental approaches must engineer to manage the transition of control between the driver and the autonomous system and they must stay within cost parameters and the aesthetic tastes of private buyers. Disruptors, all of whom seem to be targeting a fleet-based ride hailing model, have another major advantage: the vehicles can be tuned for a single geographic market, with highly specific digital maps enhancing the 3D mapping capability of the autonomous solution (Exhibit 5). The service would be strictly geofenced to the area defined by the maps, not accepting requests requiring travel outside the boundaries. Given big enough territories, geofenced robo-taxi service could suit a wide range of travel needs – commuting, local errands, family transport, etc. – obviating the need for secondary vehicles, and even, in some cases, replacing the primary family car.

Exh 5: Differences in Incremental and Disruptive Approaches

Subscription Pricing

Most pundits seem to presume that autonomous robo-taxi fleets will follow the familiar Uber/Lyft ride hailing model of independent drivers and per ride pricing, quoted at the time of request. We believe that this approach will prove problematic. The value of pure ride brokers, who do not operate their own fleets, would be greatly diminished – we expect most fleet operators would integrate forward into managing their own customer logistics. Even more importantly, we believe riders will increasingly arbitrage ride quotes across services to find the cheapest and/or quickest option, further squeezing the brokerage role. AI assistants, like Amazon’s Alexa, Google’s Assistant or Apple’s Siri, will facilitate this as they gradually become primary interfaces for consumers and displace individual apps. Users will ask their assistant to get them a ride, not necessarily an Uber.

Because of this, we believe the future of TaaS will be in subscription services. Taking a page from Amazon’s wildly successful Prime membership program, TaaS operators will offer packages of rides for a monthly rate, perhaps with tiers based on time, boundaries or total ride caps. This approach would eliminate ride-by-ride price arbitrage, lock in customer volumes, and give incentive for riders to explore using the service for new purposes.

A $120B Opportunity by 2030 – Just in the US

The most recent US government statistics show that American’s spent $1.23T on personal transportation in 2014. About $931B of that is the all-in costs of car ownership, plus the use of livery services (Exhibit 6). Obviously, not all of that would be available to geofenced self-driving TaaS, but the average length of a car trip is just 9.7 miles round trip and over half of the trips are for predictable categories, like commuting, shopping, errands,

Exh 6: Spending on Vehicle Ownership and Maintenance in US

Exh 7: Vehicle Trips by Purpose in US

school runs, and church. Urban/suburban communities account for 80% of US households. The average US household has roughly 2 cars. Ride hailing, provided by human drivers in taxis or driving their own cars for ride brokers, is a $12B industry, growing at a nearly 23% pace owing to the on-demand app revolution spurred by Uber. These stats speak to the potential of TaaS.

Autonomous driving will greatly accelerate the transition. Robo-taxis will be superior to traditional ride hailing in several important ways. First, it will be dramatically cheaper. Drivers account for 78% of the costs for Uber and the like. Operators will not have to raise prices during “surge” demand to get more cars to the right places. Idle cars can be positioned ideally in anticipation of likely demand and assigned rides immediately, reducing pick-up delays. Cars will know exactly where to pick up and drop off, and will take optimized routes. Driving will be significantly safer than with human drivers, and trips will be monitored by remote observers, further assuring passenger safety. Without drivers, seating configurations can be adjusted to give passengers more room. Autonomous TaaS has clear advantages for riders vs. their private cars as well – travel time can be productive, no need to consider parking, door-to-door service, and, if a family car can be eliminated, dramatically lower all-in costs.

We believe that self-driving fleet service will accelerate the annual growth in TaaS to 25% over the next 12 years, bringing the total market in the US to $135B or so, with autonomous services accounting for about $121B of that (Exhibit 8). The services will likely be deflationary, meaning that this should displace an even greater amount of spending on traditional automobiles – some cars won’t be purchased, maintained or insured, less gas will be needed, parking fees will be unnecessary, and so on. This is also far from full saturation of the market for self-driving – we see plenty of growth beyond that arbitrary mile marker in 2030.

Exh 8: Autonomous TaaS Market in US Forecast, 2017 – 2030

International markets could be even larger, although detailed 3D mapping and road testing appear a bit behind the US. We note that driving behavior varies greatly around the world, and that a prediction engine and control system weaned on US driving habits would require extensive retraining to be ready for other countries, as anyone who has ever rented a car overseas can attest. All in, global personal transportation is a $5.5T market. At least half of it could be replaced by autonomous TaaS over the long-term, at significantly lower all-in costs.

Waymo – Way Ahead

Google’s self-driving car initiative was born from DARPA’s Grand Challenges. Beginning in 2004, the US defense agency ran 3 competitions for autonomous vehicles fielded by university-led robotics teams. While no car completed the 150km obstacle course during the first challenge, a year later, 5 teams finished with Sebastian Thrun’s Stanford team edging out 2 teams from Carnegie Mellon University. The 2007 urban challenge, which required teams to obey traffic signals and merging with other traffic, reversed the order, with CMU just ahead of Stanford.

By 2009, Sebastian Thrun was working for Google, tasked with building a team to work on a commercial self-driving system. He built an all-star team of DARPA Grand Challenge vets, signing most of his Stanford team (including the now controversial Anthony Levandowski) and reaching across the aisle to nab CMU luminaries like Chris Urmson, Brian Salesky and Dave Ferguson. This high-powered team was given the mandate and the funds to go after autonomous driving as fast as they could.

Exh 9: Waymo Self Driving Milestones

Google has been first to hit every milestone (Exhibit 9). It began collecting driving data in 2010. Nevada granted a Google car the first driverless license in 2012 – requiring a human driver at the ready and an engineer on hand to monitor the system. In 2014, it announced a working prototype car with no steering wheel or pedals and began testing with it in California in 2015. In 2016, Alphabet (having changed its corporate name from Google) broke the self-driving car initiative from the Google-X new research incubator and launched it as full-fledged business unit dubbed Waymo, giving it a clear mandate to move toward commercializing the technology. A number of the Waymo founding fathers have since moved on to found their own self-driving projects – Urmson founded Aurora, Salesky started Argo (bought by Ford), Levandowski found his way to Uber (since fired under accusations of IP theft), and Thrun stayed loosely under the Alphabet umbrella to found “flying car” start-up Kitty Hawk backed by Alphabet CEO Larry Page and higher education MOOC company Udacity.

Exh 10: Artificial Intelligence Scientists by Organization and Number of Citations, Major TMT Companies

Still, Waymo sits pretty, with a deep and experienced team of AI talent – Alphabet has 60% more scientists with at least 1000 citations in academic journals for AI work than its next closest rival, Microsoft (Exhibit 10). It has a database with highly detailed records of more than 4 million autonomously driven miles (combined with data from more than a million human driven miles as well). By contrast, Uber announced that it had hit the million-mile mark just two months ago and no other rival is thought to be close to that

Exh 11: California Reported Self Driving Activity

mark. In California – popular with developers for its liberal testing rules – 2016 data showed that Waymo had logged 636K miles of testing while all other companies had just 20K miles combined (Exhibit 11). California requires that testers report the frequency with which human drivers must disengage the self-driving system and take control. In all those miles, Waymo engineers intervened just once every 5,000 miles. Meanwhile, the next closest rival, BMW, drove just 638 miles between disengagements, with most well below that figure. Furthermore, Waymo has begun tailoring its testing to expose its system to complicated circumstances – bad weather, steep hills, oddly engineered intersections, emergency situations, etc. – making its record of intervention even more impressive. It hired the Chandler Arizona Police and Fire departments to run a day of exercises so that its vehicles could learn from real emergency vehicles and personnel. Everyone else is just solving the easy questions of highway driving and straightforward intersections.

Earlier this year, Waymo pulled back the curtain a bit, offering reporters a peak at “The Castle” – a test facility built on a retired Airforce Base in central California. There, Waymo builds replicas of traffic configurations that have vexed its software and runs repeated tests with hired drivers, bikers, pedestrians and other human actors. The cars run through their paces and the information is fed back into the training system. Later, Waymo engineers can use the data to run computer simulations to train the software through thousands of additional repetitions under subtly shifting parameters – perhaps adding a virtual firetruck crossing the intersection. Through simulation, Waymo multiplies its “miles driven” several folds – the simulator posted 2.5 BILLION miles in 2016. While rivals have also begun testing in simulation, Waymo has been at it for years longer, and can base its simulations on its unmatched base of driven miles and painstaking recreations.

Fully Self Driving Robo-taxis – in 2018!

Waymo began testing its driverless minivans in the Phoenix area in 2015, beginning in the suburb of Chandler and expanding to the broader metro area with time. Earlier this year, Waymo began offering rides in autonomous vehicles (with an engineer or two up front) to members of an “Early Rider” program. It began taking delivery of the first of 600 Chrysler Pacifica Minivans kitted out with its self-driving hardware this past fall, and has hired on Avis to service the cars daily – cleaning, fueling, simple repairs, etc. – and contracted with AutoNation to handle the more extensive repair and maintenance needs of the fleet. These deals reveal Waymo’s ambitions – if it were just 600 minivans in one location, Waymo could probably handle it on its own.

It recently announced that it would pull the driver for Early Rider trips in the 600 square miles around Chandler, leaving an engineer in the back seat for the time being. With time, the service will expand to the whole greater Phoenix metro area and begin to accept ride requests beyond its Early Rider members, eventually, pulling the engineers from the car entirely (Exhibit 12). The trial will proceed cautiously, but the road to commercial geofenced self-driving robo-taxi fleets has begun, and we expect the service to be widely available in Phoenix by 2019. Waymo’s other testing cities – notably Austin TX, Kirkland WA, and Mountain View CA – are almost certainly next on the list, and we would expect testing (and detailed mapping) to be expanded to new markets quickly. Expect the move to full commercial service to accelerate with each city. We would expect to have several new markets in operation by 2020.

Exh 12: Waymo Progress Capture on Self Driving Vehicles

Waymo is at LEAST two years ahead of everyone else (Exhibit 13). Car makers, including Ford and GM who have been vocal advocates of self-driving tech, still see full autonomy coming post 2020. Elon Musk has promised that Tesla owners would be able to sleep behind the wheel while the car does all the driving by 2019, but the company has fallen behind on the incremental updates that would suggest the progress that might make that fantasy reality. Uber, thus far the most capable competitor, as seen its engineering ranks roiled by Alphabet’s legal action over trade secrets allegedly transferred to Uber by former Waymo employee Anthony Levandowski. Uber just hit the 1 million autonomous mile market this year, more than two years after Waymo announced that it had hit the mark in June 2015. Uber is also offering driverless rides to the public, but does not seem close to being able to pull the engineer from the driver’s seat. A recent Buzzfeed article describes a ride with constant human driver intervention and slow reactions to the road environment. Uber keeps its cards close to the vest, but California testing data from 2016 was poor (Uber is now testing elsewhere to avoid California’s reporting requirements), supporting our not close to ready hypothesis. Other companies investing in full autonomy, including Ford and GM, have promised availability in 2021 or later. We think even that might be a stretch.

Exh 13: Autonomous Vehicle Leaders Summary

Waymo is ready today, with the beginnings of a capable ecosystem (Avis and AutoNation are premium partners) and substantial corporate assets. Google has overwhelming superiority in digital maps with better coverage and dramatically more detail than any competitor. These maps can form the basis for expanding 3D mapping into new territories. Alphabet also has almost unparalleled reach to consumers – seven different services reach to more than a billion users, each with the means to support a self-driving TaaS platform. We are particularly enthusiastic for AI assistants (AI Assistants The Next User Interface Paradigm) and see Google Assistant as a powerful tool that could circumvent the “app model” user interface championed by Apple. Rather than open an app to order a car, a consumer might simply ask an AI assistant for a ride. In this context, Google, with its control of Android, could be sitting pretty for TaaS.

We believe Waymo will establish its business model as it expands in Phoenix, testing systems for assigning vehicles to riders and managing fleet logistics. As it grows confident in the details of a go to market strategy, it will begin expanding to new markets, most likely in Arizona, California, Texas and Washington. We would expect new testing markets – Nevada, Michigan, Utah, Florida, Tennessee, North Dakota, Louisiana and the District of Columbia have all have passed legislation allowing self-driving cars – to herald future market launches, as Alphabet management undoubtedly understands the advantages of being first to scale in Internet based businesses.

With at least a two-year lead over troubled Uber, and likely considerably more over anyone else, we believe that this game is Waymo’s to lose (Exhibit 14 & 15). There are also international opportunities, although a system trained on US drivers will have to be retrained to reflect the laws and customs of each market and there are competitors like Baidu, which will be more focused on markets outside of the US.

Exh 14: Possible and Initial Partnerships in Self Driving

Waymo could also expand its focus to adjacent markets, like trucking (Autonomous Trucks are Years Away) or local delivery, which we believe will develop more slowly than the TaaS opportunity.

Exh 15: Disruptors with1M+ Tested Miles

What’s Waymo Worth?

Valuing a business like Waymo is difficult. It hasn’t really revealed its business model to the public, much less generate any revenues. Still, when the sales begin to roll in, we expect them to grow very, very quickly. Ride hailing is a nearly $12B market today in the US, growing a bit more than 20% annually as Uber and Lyft siphon business from taxies and livery (Exhibit 16), while bringing new riders to the party. We expect self-driving to materially accelerate that growth as it spreads market to market. Our model projects growth between 2018 and 2030 to average about 25%, peaking in 2025, with autonomous TaaS going from 0% today to roughly 80% in 2030 (Exhibit 17). This would put Waymo’s addressable market at about $120B in 2030, about 12% of personal transportation spending today, and we expect it to have captured about half of the market.

In the long run, we expect Waymo’s business to be nicely profitable – it will likely end up leasing the vehicles from a financial partner and contract with partners like Avis and AutoNation for management of the vehicle operations. Its roles will be maintaining advantage in self-driving systems, managing the logistics of matching riders and cars, and signing new customers and keeping them happy. These elements will have significant scale economies, and we expect Waymo to have substantial scale advantage over all comers. Still, it may remain unprofitable for a while as it builds the business, relying on the foresight of investors to understand the magnitude of the opportunity at hand.

Exh 16: Uber Contribution Margins and Driver Economics