Self-Driving Cars: Building a Team to Bring TaaS to Market

Print Friendly
Share on LinkedIn0Tweet about this on Twitter0Share on Facebook0

SEE LAST PAGE OF THIS REPORT Paul Sagawa / Artur Pylak


psagawa@ /


July 5, 2017

Self-Driving Cars: Building a Team to Bring TaaS to Market

We believe that the endgame for self-driving cars is Transportation-as-a-Service (TaaS), requiring excellence in three complementary roles – Technology, Platform, and Operations. Technology has two necessary elements: 1. 3D Mapping – fusing inputs from complementary sensors with map data to yield a detailed real-time model of vehicle’s environment and optimized routes for reaching the destination; and 2. Autonomous Controls – interpreting the real-time 3D map and directing the vehicle to safely and efficiently complete the requested driving task. GOOGL has a profound lead on both, with a defanged Uber and BIDU the next best. The companies focused on adding incremental autonomous features to private vehicles (e.g. TSLA and other automakers, INTC/MBLY, Bosch, etc.) and the relative newcomers (e.g AAPL, Comma.AI, Nutonomy, etc.) do not have the driving data, mapping technology, or AI chops to compete at the same level. The Platform role manages the customer facing aspects of the service – marketing, customer service, dispatch, etc. Uber, Didi, Lyft have a head start, but the major internet players – GOOGL, AMZN, FB, Tencent, BABA, and others – could be well placed to compete. The Operations role has been largely ignored, but will be critical. Someone will have to buy the vehicles and hold them on their balance sheet and someone will need to manage that physical plant – charging, cleaning and maintaining the vehicles. GOOGL’s deal with CAR, and Uber’s with Daimler are the tip of the iceberg, as hundreds of vehicles become tens of thousands. As TaaS becomes closer to commercial reality, we expect building partnerships with excellence in all three roles will be the most important factor for success.

  • Transportation-as-a-Service is the end game for self-driving cars. We have written about the substantial benefits of fleet-based autonomous vehicles ( providing transportation-as-a-service (TaaS), driving down the costs of transportation while improving convenience, safety, efficiency, municipal burdens, and other substantial advantages. In the short run, automakers and their suppliers will push increasingly autonomous driver assistance capabilities, but longer term, we believe robo-taxis and autonomous delivery will slowly displace vehicle ownership, increasingly leaving private cars as a luxury for urban and suburban consumers, and a growing expense for those out of range.
  • TaaS will require excellence in three distinct roles. Most of the market hype surrounds the AI and sensor technology at the core of self-driving solutions, but commercial TaaS service will require excellence on two other dimensions. First, the service must be sold to consumers and operated efficiently. This is the platform invented by Uber for coordinating rides in driver-owned cars. Second, the vehicles must be built, owned and maintained – fleet operations will be critical to keeping appealing vehicles on the road at peak efficiency. Best-of-breed partnerships will be a prerequisite for success
  • The technology role has two necessary elements. Self-driving tech can be split into two complementary elements. The first is 3D mapping, fusing input from multiple sources – sensors (LiDAR, radar, cameras, etc.), detailed map databases, and real-time road condition updates – to intuit a complete picture of the vehicle’s immediate surroundings. The second is the autonomous control system which must interpret the 3D Map and make driving decisions. Of the two, 3D mapping is the more difficult job and a critical driver for success on the second.
  • GOOGL is FAR ahead on Technology. GOOGL has been working on multi-sensor 3D Mapping solutions for nearly a decade. It has its own proprietary LiDAR believed years ahead of its rivals. It has the broadest and most detailed map database in the world. It also has more than 3M miles of fully detailed data collected in varied driving conditions – by far the deepest dataset available. These assets fuel a deep and talented AI team that has been working on self-driving since 2008. All of this makes GOOGL’s Waymo #1 in both the 3D mapping and autonomous control elements. Well behind are Uber (somewhat damaged by scandal) and BIDU (late to the game, but favored by Chinese authorities). None of the companies working to incrementally enhance driver assistance technologies toward full autonomy (e.g. TSLA, other automakers, INTC/MBLY, Bosch, etc.) have sufficient 3D mapping capability or driving data to realistically challenge GOOGL. New entrants (e.g. AAPL, Nutonomy,, and many others) are far too late to make up the ground.
  • Internet leaders positioned for the Platform role. TaaS requires reach to consumers, outstanding UI/customer service, and an efficient system to match riders to vehicles. This is the business invented by Uber and emulated by global rivals like Didi, Lyft and many others. While these firms have a leg up with brand recognition and logistics experience, the barrier posed by driver networks disappears with robo-taxis, and internet platform companies, like AAPL, AMZN, FB, and GOOGL (and Tencent, BABA and BIDU in China) have the reach and skill set to challenge for this role as well. In particular, we believe integration with virtual assistants, like Alexa, Google Assistant, and Siri, could further erode the position of independent ride sharing brands.
  • Operations – the forgotten but critical role. The Silicon Valley companies that lead in self-driving tech and are positioned to manage TaaS platforms have no experience in 1. maintaining massive fleets of vehicles nor do they have interest in 2. holding them as assets on their balance sheets. Yet managing the physical inventory efficiently, maximizing vehicle availability, and assuring ride quality for customers will be critical for success. This is the obvious rationale for GOOGL’s deal with CAR (less so for AAPL’s leasing arrangement with HTZ), and auto rental firms are good fit for fleet operations. This could also be place for auto makers and experienced commercial fleet operators (local taxi franchises, delivery services, field maintenance units, etc.). While it has not been widely discussed, TaaS consortia will also require financial partners willing to carry many thousands of rapidly depreciating autonomous vehicles on their balance sheets. We are very skeptical of the idea of private owners.
  • Ecosystem building is just beginning. With technology moving from experiment to commercial trials, leaders must begin recruiting partners to fill the key roles needed to bring TaaS to market. GOOGL’s deals with Lyft (platform) and CAR (operations) are clear early steps in that direction, as is Uber’s non-binding operations agreement with Daimler. Other players remain in much earlier stages of business development, perhaps still too far behind in technology to begin filling ecosystem holes. N.B. AAPL’s vehicle lease deal with HTZ for test cars is very different from the GOOGL/CAR partnership. We believe ecosystem’s will begin to line up behind the most promising go-to-market approaches, with financing the final, and perhaps, most important step to commercialization.

It Takes an Ecosystem

Since GOOGL started its self-driving moonshot in 2008, the market and media focus has been on the AI technology at its core. We have taken the same tack, publishing our detailed survey of autonomous vehicle tech last year (, concluding that the path to full autonomy being taken by GOOGL’s Waymo and a few other disruptors intent on delivering fleets of robo-taxis offering Transportation-as-a-Service (TaaS) had a head start and a more direct engineering path than the crowd of automakers and suppliers looking to evolve into full autonomy through incremental improvements to smart driver assistance functionality.

Amidst the noise of dozens of startups and the posturing of the auto industry, GOOGL has established a dominant lead in self driving technology. We divide the technology role into two parts. 3D mapping involves fusing the inputs from multiple sensors (LiDAR, radar, cameras, microphones, etc.) with highly detailed digital maps (updated in real time for traffic and road conditions) of the intended route. The ideal result is a perfect view of the world around the vehicle – the roadway, the other cars, street signs, pedestrians, obstructions, weather phenomena, etc. – to inform the decisions of the autonomous control system. The control system is the AI brains, that having been trained through many iterations on huge bases of actual driving data, will conduct the car more safely and efficiently than any human driver. With proprietary LiDAR hardware, 9 years of engineering work, and the world’s most accurate and complete digital maps, GOOGL has a big lead on 3D mapping, and with 3M+ miles and 20PB of driving data, billions of miles in simulations, and the deepest and most experienced staff of AI experts, it is way ahead on autonomous controls as well. The proof is in the pudding – Waymo’s 0.2 human interventions per 1,000 miles of California testing is 2 magnitudes better than any would-be rival.

However, Technology is just one of the ingredients for TaaS, and as Waymo, Uber and others begin to build partnerships to come to market, the others will begin to take part of the spotlight. TaaS will need a consumer-facing Platform to reach customers, dispatch their rides efficiently, and delight them in the experience – this is what Uber, Didi and Lyft do for their networks of riders and drivers. While these services have a head start, we believe that the major internet players – GOOGL, AMZN, AAPL, FB, BIDU, BABA, and Tencent – have the consumer reach and skill set necessary to be formidable competition. We believe that AI virtual assistants, like Alexa, Google Assistant, and Siri, could eventually subsume the platform role entirely. Interestingly, Uber’s greatest asset in the platform role may be its hard experience negotiating for approval with local governments all over the world.

TaaS will also need Operations partners to handle the physical and financial requirements for the thousands of vehicles that will be part of a successful TaaS operation. Cars need to be charged, cleaned and maintained, and someone needs to pay for them and hold them as assets on their balance sheet. Here, Waymo’s deal with CAR is particularly interesting – rental car operators have well located service depots with experience maintaining vehicles to consumer standards. Automakers and commercial fleet operators (e.g. taxi, delivery, service fleets, etc.) could also fill this role. As for financing, we would expect traditional equipment leasing companies (e.g. GE, CIT, etc.) to take interest.

Three roles: Technology, Platform, and Operations – successful partnerships will have to cover all three. Waymo (CAR, FCAU, Lyft) and Uber (Daimler, AMZN, AAPL) have begun to build ecosystems. The others do not appear ready to start.

Vehicular Autonomy

In September 2016, we published our first look at the nascent market for self-driving technology (, identifying two distinct philosophies. In one camp were the incrementalists – companies that were approaching autonomy via increasingly sophisticated driver assistance features, such as self parking and highway autopilot. The automakers and suppliers in this group tended to eschew LiDAR for its unwieldy bulk and expense – an expedient approach with significant long-term performance disadvantages for full autonomy. Incrementalism also demands engineering for seamless transfers of control from driver to computer and back, a scenario rife with potential convenience and safety issues (Exhibit 1).

Exh 1: Two Roads to Level 5 Autonomy

Taking the alternative tack were the disruptors – led by Alphabet’s Waymo, Uber and Baidu – who looked to engineer for full autonomy from the start. These players engineered their sensor suite for the richest possible set of inputs – LiDAR, camera, radar and other less critical inputs (ultrasonic, microphones, etc.) – working from the start for a smooth fusion to an integrated 3D map, assuming that the cost and bulk of LiDAR would no longer be a problem once fully autonomous vehicles were a commercial reality (Exhibit 2). Moreover, by eliminating the need for driver-computer transfers, the design of the control system is much cleaner, yielding development time to market advantages. We believe that the disruptors will easily beat the incrementalists to full level 5 autonomy (Exhibit 3).

Exh 2: Disruptor Systems are the Most Comprehensive

Exh 3: Levels of Driving Automation for On-road Vehicles

The endgame for disruptors is Transportation-as-a-Service (TaaS) – a logical extension to the on-demand ride hailing service pioneered by Uber and mimicked by rivals like Didi and Lyft. By maximizing the utilization of vehicles, increasing overall efficiency and safety, and eliminating driver payments, TaaS operators can reduce operating costs by half or more, while improving availability for customers. We expect a subscription model – unlimited rides within geographic boundaries – will be the primary product offering, with substantial appeal for consumers in urban and suburban markets, replacing private vehicle ownership for many. This is the paradigm shift feared by the many companies invested in the private automobile value chain.

Exh 4: Key Capabilities of Transportation as a Service (“TaaS”) Providers

The Three Roles in TaaS

Most of the attention in the emergence of self-driving vehicles has been on the technology. Obviously, the systems able to perceive the environment around the car and make driving decisions as well (or better than) a human are the primary obstacle to the emergence of the TaaS paradigm, but there are other roles critical to a fully realized service. We believe winners in TaaS will combine excellence in the Technology, with similarly capable user facing Platforms, and fleet management Operations (Exhibit 4).

Technology. This role is comprised of two intertwined elements (Exhibit 5). The first is 3D mapping – generating the digital model of the real-time environment around the car and along the planned route of its travel. The accuracy and detail of this model is critical to the safety and efficiency of the service, identifying obstacles and risks and allowing the vehicle to adjust accordingly. The first element of 3D mapping is a suite of sensors, with LiDAR, radar and cameras the most typical centerpiece. Each of the sensor types has a different set of strengths and weaknesses (Exhibits 6-9). LiDAR offers precise distance reckoning for hard and soft objects, is unaffected by lighting conditions, and can reveal identifying shape detail, sometimes for objects behind other objects. However, LiDAR is unable to see color contrasts, so it can’t read signs or interpret traffic signals and becomes much less effective in rainy, foggy or snowy conditions. Cameras can handle the signs and signals, fills in identifying details that would be missed by LiDAR alone, and has the longest range of the typical sensors. On the downside, cameras are inherently 2D and do not provide distance reckoning directly (it must be approximated via complex image processing systems). Moreover, cameras rely on external light and are stymied by dark or high glare conditions and by inclement weather. Radar is unaffected by weather or lighting, and offers precise distance reckoning for hard objects. The serious downside to radar is lack of shape detail and its disturbing inaccuracy in identifying soft objects, like pedestrians. Other input sensors, like microphones (to identify sirens, audible alerts, and spoken commands) and ultrasonic transceivers (for very precise, close distance reckoning) round out the array.

Exh 5: Self-Driving Technology Elements

Exh 6: Lidar Performance Against Key Parameters

Exh 7: Camera/Optical Performance Against Key Parameters

Exh 8: Radar Performance Against Key Parameters

Exh 9: Parameters of Autonomous Vehicle Sensor Performance

Exh 10: Maps and Navigation Compared

Another critical input to the mapping process are actual maps (Exhibit 10). Ideally, these maps would provide fine detail about the fixed conditions at every point on a planned route – roads, lanes, exits, overpasses, driving regulations, curbs, potholes, blind turns, building entrances, the content of all signs, etc. Furthermore, it would also be ideal to get real-time updates, on things like traffic, road conditions, lane closures, accidents, road work, weather, etc., provided through crowdsourcing and municipal authorities. All of this provides strong foundation under the information provided by the sensor suite.

Finally, the separate inputs of the various sensors and the information from the digital map must be logically combined into a single digital depiction of the environment around the vehicle – a process called sensor fusion. This is not at all simple, requiring a sophisticated AI model able to prioritize the highest quality information and provide a maximally detailed, real-time to the microsecond, 3D picture of everything within a radius of at least 1,000 feet around the vehicle, in almost any driving conditions. This is why it is important to have a complete set of sensor inputs from the start – adding LiDAR later, for example, would require taking development back many, many steps in order to fuse its input into the AI mapping model.

Exh 11: Three Basic Ingredients for Artificial Intelligence

The second element of self-driving technology is the autonomous control system. This is the AI brains of the car, making the split-second decisions to accelerate, brake, turn, signal, etc. in response to the information provided by the 3D map. Our usual AI recipe ingredients are data, computing and talent (Exhibit 11). For self-driving systems, having a broad and detailed dataset is critical – it should offer a representative view of the extremely varied conditions under which human drivers must operate, including as much potentially accident inducing unusual phenomena as possible. It should also provide the same level of mapping detail as would be available in real-life application. This data is then used to simulate driving over hundreds of millions of miles, requiring time, patience and powerful AI-tuned computing resources. Finally, the quality of the scientific talent that writes, trains and tweaks the models as they learn is a critical resource – unlike traditional software AI development is a holistic process where leadership from the top has a disproportionate influence on the quality of the solution and the time it takes to build it.

Platform. The platform role consists of three primary areas of responsibility – logistics, customer care, and government relations (Exhibit 12). TaaS customers will demand rides from a platform provider, which will handle the logistics of identifying the most appropriate vehicle and routing it to the pick-up point. It will also need to determine the best possible route to the destination, taking the passenger’s preferences into account, including multiple pick-ups in a ride sharing situation. All of this will happen within the context of managing the overall efficiency of the full network of cars, keeping the optimal number on the road and positioned to meet expected demand.

Exh 12: Self-driving Platform Responsibilities

The platform provider will also reach out to attract new customers, and to cosset the riders with an intuitive interface and responsive customer service. We expect that monthly subscription will be the primary product, likely geofenced but with generous (or even non-existent) usage limits – the platform provider will administer payments and compliance to terms.

Importantly, the platform provider will likely be responsible for gaining the necessary government approvals, a substantial potential obstacle that will require political deal making across every level of government. Given the well-publicized impact on jobs, the task may be fraught, although the many municipal benefits (i.e. safety, pollution, parking, service for the disabled and elderly, reductions in traffic infrastructure costs, etc.) will outweigh the possible employment impact for many communities. We believe as those benefits become more concrete through experience, more recalcitrant municipalities will begin to come around. The platform provider will have to manage that political process deftly.

Operations. The operations role has been largely ignored, but it will be a critical step to the rise of TaaS. There are three aspects of this. The most straightforward is manufacturing – car makers will make these cars, perhaps of their own design and perhaps based on specifications provided by a technology or platform player. The designs will maximize rider comfort, energy efficiency, durability, and serviceability, while housing the sensors, processing and communications technology required by the self-driving system.

Exh 13: Self-driving Operations Responsibilities

The vehicles will need regular servicing to high consumer standards – charging, cleaning, maintenance, etc. – and will need to be secured during their downtime. Thus, fleet management partners will operate depots, convenient to the populated areas that will generate most of the demand, where the autonomous vehicles will return for their servicing. Fleet operations must balance efficiency (which puts the cars back on the road quickly) and quality (which assures each customer a clean, well maintained ride). We believe execution in this role will prove a paramount factor in the success of TaaS operations.

Finally, someone will need to pay for these cars. The companies at the front of the development of TaaS – GOOGL, Uber, BIDU, etc. – are not likely to want to hold thousands of robo-taxis as assets on their balance sheets. Finance partners will almost certainly be involved, buying and leasing the vehicles back to either the fleet operator or the platform operator. The cost of financing on these rapidly depreciating assets will be a meaningful consideration for the spread of TaaS.

Rating the Players

Technology. Alphabet’s Waymo is FAR ahead of the pack on 3D Mapping. It has been working to fuse LiDAR, camera, radar, and ultrasonic input since 2008. It has internally developed LiDAR hardware – the centerpiece of its intellectual property theft lawsuit against Uber – believed to be a dramatic improvement on commercially available alternatives, delivering very high resolution and range, robustness to weather conditions and much lower costs. It also has, by far, the most complete and detailed digital map database in the industry, including elements such as 360 degree photographic images, real-time crowdsourced traffic data, building details (businesses, entrances, etc.), and many others, covering every potential market for TaaS services. Add to this Alphabet’s extraordinary roster of AI talent, leading that 10-year effort to optimally fuse the input from those sensors (Exhibit 14).

The headline grabbing lawsuit establishes that Uber also has a proprietary LiDAR design, which Alphabet contends is a copy of its own technology. Indeed, the schematics for Waymo’s sensor were amongst the 14,000 files reputedly lifted by former Google scientist Anthony Levandowski. In its filings, Uber claims any similarities are coincidental – if so, the company may have made up ground, but the final say will happen in court. Uber has also been hard at work building its own maps for the geographies that it serves, although anecdotal experience suggests that the company is a long way from rivalling Google Maps, and its drivers are a great source for traffic data.

While we see Uber as years behind Waymo on 3D mapping, the rest of the self-driving pack may be well behind Uber. Baidu recently offered its self-driving solution in open source, with a goal of attracting partners to strengthen its sensor fusion solution. Planned Intel acquiree Mobileye plans to crowd source maps from the vehicles equipped with its driver-assistance autonomy solution, but is committed to a limited camera-radar sensor array that we believe hamstrings their efforts to derive a 3D map adequate for true L5 autonomy. The same is true for Tesla. Elon Musk will not put an ugly LiDAR turret on top of a Model X, and until commercial solid-state LiDAR arrives at reasonable price, Tesla will suffice with video and radar. This understandable aversion to expensive turrets carries to all auto manufacturers, all of whom will need to revamp their sensor-fusion development to add LiDAR down the road.

Exh 14: Technology Leaders Summary and Score

Waymo leads in autonomous control system technology as well. 10 years of work, more than 3M testing miles and 20PB of data, and an all-star team of AI engineers have yielded a driving system that appears much closer to prime time than any other. In California, which requires the 36 companies conducting self-driving tests on its roads to report a detailed record of miles driven and instances where human intervention was required, Waymo’s cars traveled more than 5,300 miles between disengagements on average in 2016, logging more than 635K miles of travel. The next best on disengagements was BMW, which noted just a single instance during its meager 635 miles of testing its highway autopilot. GM’s Cruise Automation posted 9,846 miles of fully autonomous testing but suffered a disengagement every 414 miles – performance that is several years behind Waymo (Exhibit 15).

We note that autonomous driver assistance technology, like Tesla’s Autopilot, have delivered millions of miles of autonomy under limited conditions. Almost all these miles have been driven under highway conditions ideal for self-driving systems – no cross traffic, no pedestrians, no bikes, no sharp turns, etc. As with many engineering problems, the last steps are the most difficult – urban driving, steep hills, traffic circles, blind entries, police directing traffic by hand, inclement weather, etc. Only Waymo has even begun to address these situations – the others don’t even know what they don’t know.

Exh 15: California Reported Self-Driving Activity

Platform. Uber has been one of the great successes of the mobile app era, creating a platform to match drivers and their vehicles with users looking for rides. Uber’s franchise is protected by the difficulty in building a critical mass of customers and drivers, by the intricacies of managing dispatch logistics, and by the political obstacles that stand in every local market. Still, Uber’s competitive moat has not scared off would-be rivals – Didi beat it on its Chinese home turf, Lyft is pouncing on its larger rival’s recent scandals, and a raft of local players are scrambling for relevancy (Exhibit 16).

In the TaaS world, its ride-share honed Logistics strength will still be a formidable asset, positioning it in the lead of its current rivals for its skill in dispatching vehicles, managing ride shares and successfully routing passengers to their destinations. Still, companies like Amazon, Alphabet and Alibaba, already in the delivery business would seem to have the skill set needed to compete successfully. In the Customer facing roles, the internet giants may even have the upper hand. Alphabet, Facebook, Apple and Tencent all have deeply engaged digital relationships with more than a billion people each, with engagement platforms ready to absorb ride sharing as an added function. Amazon and Alibaba may have fewer than a billion users, but each has commerce platform well-tailored for brokering transportation. We are particularly interested in the role that virtual assistants – like Alexa, Siri, and Google Home – may play as an interface between their users and TaaS services. This would advantage those companies in the platform role, while commoditizing 3rd parties engaged through the virtual assistants.

The Government facing element of the platform role may be Uber’s biggest advantage. The battle for approval will be fought at every level of government, with local licensing and regulation the most nettlesome hurdle. With driver employment on the line and the ongoing resistance from the traditional taxi and limousine lobby, would-be TaaS platform operators will need to maneuver aggressively and adroitly market by market. Uber has already been through it, in markets all over the world. Other ride-hailing rivals are regional – Lyft in the US, Didi in China, Ola in India, and others – perhaps limiting them, in this regard. Google has obvious history negotiating with governments on the national and even international scale, but its experience at the local level is more limited to its fiber broadband initiative, and a variety of other small programs, like Google Express delivery. Few of the other companies with interest in TaaS would seem to have any relevant experience.

Exh 16: Platform Leaders Summary and Score

Operations. For TaaS, Manufacturing may be a commodity. Platform operators will provide detailed specs, and car makers will bid to provide vehicles fit with the requisite technology and designed to provide the comfort, durability, efficiency and flexibility demanded. The Platforms will multisource to keep their suppliers on their toes and to pressure them on price. The automotive industry hopes to avoid this scenario, but while we believe that it will be more than a decade before TaaS really has impact on the private vehicle market, it seems a long shot that the car makers will have the success in either the technology or platform roles that could give them leverage on selling the actual cars (Exhibit 17).

Exh 17: Operations Likely Participants and Announced Deals

Fleet Management finally hit the radar screen for investors and other interested industry observers with Waymo’s agreement with Avis to service the 500 autonomous minivans that it plans to roll out for its commercial service testing in Phoenix. All of the thousands of vehicles that a TaaS operation intends to use for service in cities and suburbs will need to be regularly charged or fueled, cleaned to meticulous standards, maintained in efficient working order, and stored securely during downtime. None of the headline names in self-driving – Google, Uber, Tesla, Baidu, etc. – have any experience at this, at all.

Avis is a great partner for Waymo. Rental companies have numerous servicing locations convenient to urban areas over a broad global footprint. They are accustomed to maintaining passenger vehicles to consumer standards – as opposed to most commercial fleet operators with their delivery vans or field service trucks. With the prime opposition to ride-sharing coming from the hyperlocal taxi industry and their drivers, taxi depots may not be the best place to store and service robo-cabs. In international markets, the circumstances may be different, or not. Aggressive TaaS service expansion is likely 5 years away, but we believe that Avis and Hertz have significant advantages against rivals for the fleet management role.

Of course, someone needs to Finance the thousands of heavily used and rapidly depreciating vehicles needed for TaaS. Alphabet and Uber will not carry them on their own balance sheets. As TaaS slowly displaces private vehicle ownership in urban and suburban areas, we expect the fleet sizes to eventually grow into the millions. This will be a significant market opportunity for equipment lessors like GE, CIT and others, and a significant cost to be borne by TaaS operators.

Getting the Team Together

Waymo has begun stitching together an ecosystem, with a partnership with Lyft and its deal with Avis filling two holes in its go-to-market TaaS capabilities. Going forward, it needs to sort out its position with Lyft across the three parts of the platform role – Lyft will book rides, but will it handle logistics? Who will get the local service approvals? Will Alphabet subsume the customer facing aspect into its Google Assistant? Waymo should also tighten its relationship with Avis, an ideal partner, as it contemplates expanding beyond its commercial trial in Phoenix. It is currently buying its vehicles from Fiat Chrysler, but we would expect it to broaden its base to include other manufacturers. Eventually, it will need to negotiate terms with a financial partner as well (Exhibit 18).

Exh 18: Summary of Positioning and Initial Partnerships

Uber has also begun a bit of ecosystem building, some of which can be extended from its more traditional ride-hailing app. In that regard, Uber was a launch partner for Amazon’s Alexa and is a key service for Apple’s Siri as well. Uber uses Volvo vehicles in its tests but recently signed an agreement with Daimler, which intends to operate fleets of its own vehicles to be made available through the Uber app. It is not clear whether those Daimler vehicles would use autonomous driving technology from Daimler, Uber or both. However, since those deals, Uber’s self-driving program has been thrown into disarray with the highly public lawsuit filed by Alphabet over trade secrets alleged stolen by now former Uber employee Anthony Levandowsky. This was followed by a handful of employees jumping ship. In all, we believe Uber is at least a couple of years behind Waymo in being ready for market.

Baidu’s surprise decision to open source its self-driving system is a clear admission of deficits vs. Waymo and even Uber. It will offer a limited system capable of highways and “open city roads”, comparable to Tesla’s autopilot, and looks to rally Chinese automakers and component technology developers to join it to close the gap. It will certainly gain advantage from Chinese regulators, who will roll out the red carpet for testing and early deployment.

GM, which bought Cruise Automation and owns a sizeable stake in Lyft, seems to be using equity to cobble together its own team. Still, we believe that Cruise is behind Uber, which is years behind Waymo in the pursuit of full L5 autonomy. Lyft may get more out of its recent deal with Google than from its part owner. Intel’s pending acquisition Mobileye has supply deals with several car makers that give it access to data, but isn’t really part of any apparent push for full autonomy and TaaS. Similarly, Bosch is also working on various development projects with auto companies that seem well off the pace. The press jumped on the reveal that Apple had struck a deal with Hertz, but it is a simple agreement to lease 6 cars that Apple will use for testing. It is not strategic. Tesla has floated the idea that its private car customers might eventually contribute their cars into an autonomous ride-sharing system during the times when it might otherwise be idle, taking care of the fleet management and financing tasks in the process. Even if Tesla could reach full L5 autonomy on a timely schedule (we are skeptical), we are EXTREMELY skeptical that any percentage of Tesla owners would participate in such a scheme.


Exh 19: SSR TMT AI / Self-Driving Heatmap

©2017, SSR LLC, 225 High Ridge Road, Stamford, CT 06905. All rights reserved. The information contained in this report has been obtained from sources believed to be reliable, and its accuracy and completeness is not guaranteed. No representation or warranty, express or implied, is made as to the fairness, accuracy, completeness or correctness of the information and opinions contained herein.  The views and other information provided are subject to change without notice.  This report is issued without regard to the specific investment objectives, financial situation or particular needs of any specific recipient and is not construed as a solicitation or an offer to buy or sell any securities or related financial instruments. Past performance is not necessarily a guide to future results.

Print Friendly