Privately-owned Autonomous Vehicles are Coming
November 2025
Autonomous vehicles (AV) are no longer a thing of the future. Robotaxis operate commercially in five cities in the U.S. and a dozen in China, with respectively one and four companies which all have big plans to scale. The two leaders in terms of deployment, Alphabet’s Waymo and Baidu’s Apollo Go, now each deliver over 250,000 fully autonomous paid rides per week, with fleets of about 2,000 and 1,000 vehicles. These companies are currently increasing their rate of scaling. They are now coming to Europe, not just with robotaxis but also autonomous shuttles.
Major automakers have played key roles in shaping the autonomous driving landscape, investing heavily in technology and partnerships. Since the mid-2010s, they have also been providing the necessary platforms for autonomous driving (AD) system developers to deploy their tech. Without them, robotaxis and autonomous shuttles would not exist.
Incremental Waves of Partnership between OEMs and AD Tech Developers
The early partners, i.e., FCA, JLR, Ford, BAIC, GAC, GM, Toyota, Nissan, Hyundai have provided standard vehicles that required significant retrofitting to provide redundancy for safety functions as well as extra power and cooling. This has resulted in high costs per vehicle.
A second wave of vehicles emerged that are designed specifically to serve as robotaxi or autonomous shuttles, thus preventing costly modification. These vehicles are designed for commercial use, i.e., long lifecycles and readiness to carry careless riders. The OEMs behind them are fewer: Zoox for their own needs, Geely’s Zeekr for Waymo, JMC for Apollo Go’s RT6, and Yutong for WeRide’s Robobus.
A third wave of partnerships between AD system developers and OEMs focuses rather on derivatives of standard vehicles that are pre-disposed by design to host a Level 4 system, i.e., “AD-ready platform” for the purpose of shared use. As such, these vehicles provide a relatively low-cost foundation thanks to their high-volume origins. They also require much less invasive, thus cheaper customization before they can join fleets of robotaxis or autonomous shuttles. Yet, they may not be necessarily designed for intense use and long lifecycles as those described earlier. Hyundai, Toyota, Renault, and Stellantis among others are engaged on this path respectively with Waymo for the first two, WeRide, and Pony.ai.
How about flipping the script? What if a vehicle was designed with autonomous driving (Level 4) capability, focusing primarily on private use — yet with the potential to be shared? For one, this would require that the AD system does away with the HD maps typically used today. Indeed, a buyer would not accept that the operating design domain of their vehicle’s AD capability be limited to a certain geographic zone (except for regulatory limits).
In fact, this has been Tesla’s plan for some time, but their AD tech (FSD) has yet to mature. Toyota and Waymo have said last April they will “explore how to leverage Waymo's AD tech and Toyota's vehicle expertise to enhance next-gen personally owned vehicles”. Likewise, Mercedes-Benz and VW Group have indicated they were considering adding such vehicles in their portfolio before the end of this decade. Most recently, China’s Xpeng announced the introduction in 2026 of three vehicles capable to operate at Level 4.
Autonomous Vehicles Targeting Private Use: Tensor’s Robocar
Let’s focus here on Tensor, a company getting ready to leapfrog established players. Tensor was founded as AutoX in 2016 in Silicon Valley. AutoX first developed its own AD tech system and leveraged it to deploy a fleet of robotaxis in several Chinese cities, reaching about 1,000 vehicles in 2022. Last August, AutoX morphed into Tensor.
At that point, the company unveiled a vehicle it has been working on for five years. It is designed for private use with the capability to operate autonomously, building on AutoX’ extensive Level 4 experience. Today, the company has a staff of about 400 people based in Silicon Valley, Spain, Singapore, and the UAE. I had a chance to experience a representative prototype, although statically for now. Thank you to my host, who did not sponsor this article (images used here are from Tensor).
The vehicle is large, 5.5 meters in length, i.e., slightly more than a Maybach S-Class although the cabin feels smaller than one would expect given the overall dimensions. It is intended for people with certain means given the amount of tech it contains. Target buyers may be accustomed to having a chauffeur, whose presence prevents them from having certain conversations. They may want to just make the most of their commute time, while they can trust the autonomous driver. Access is easy with fully powered, wide angled, clamshell doors (see above).
The car can be driven with assistance ranging from none to Level 1, L2, and L3. For L4, the pedals and yoke-type steering wheel retract, and the center display slides across in front of the would-be driver, creating space for five passengers. The ability to select various levels of assistance or autonomy makes the vehicle regulation agnostic. Their use could be geo-fenced to ensure compliance.
There is a large amount of tech and care to details on the vehicle, especially related to the private use of autonomy since typically the equipment (sensors, etc.) is monitored frequently and maintained by skilled technicians. The 37 (mainly 17 MP) cameras, 5 lidars (designed in house) are equipped with cleaning, defrosting, and shutters to protect them when not in use. There are infrared cameras under the vehicle to detect objects (or a child) and avoid any accidents when the car is called remotely. There are water sensors to automatically move the car in case of a flood. There are also 11 radars, 22 microphones, and more. The vehicle is also fitted with small displays at each corner to communicate with other road users.
The electrical / electronic architecture is built around three high performance controllers fitted with eight NVIDIA Thor which provide 8,000 TOPS of compute. This enables not only L4 autonomy but also features such as an embedded Large Language Model to operate the conversational assistant so that the data stays onboard.
This care for a high level of privacy impacts other areas. No data leaves the vehicle, not even to help the company train its AD models. All processing takes place onboard, thus the high compute power. The five cameras inside the cabin can be covered with manual shutters for extra precaution. Last, the drive containing sensitive data can be easily removed.
The vehicle was designed and engineered with major Tier 1 suppliers, e.g., Bosch, ZF, Autoliv, and will be produced in Vietnam by VinFast. Deliveries are expected to start in the second of 2026, first in the Middle East, then in Europe (where the vehicle dimensions may be an issue for dense cities) and the U.S.
I invite curious readers to watch this extensive video (102 min) which provides significantly more information.
The Future Shared Use of Personal AVs
Several years ago, Tesla promised that the activation of its FSD feature (currently L2) would eventually allow owners to share their vehicles on a ride-hailing platform, thus generate revenue. Tesla launched its robotaxi pilot this summer in Austin and San Francisco, using company-owned vehicles. Today, the very small fleet operates with restricted access and safety operators but is intended to scale — the latest promise calls for 1000 vehicles in the San Francisco Bay Area and 500 in Austin by year end, which seems again unlikely. Eventually, privately-owned Teslas would be invited to join the fleet to generate revenue for their owners.
Tensor — or Xpeng — may beat Tesla at this game. Indeed, the vehicle is designed to be integrated with ride-hailing platforms. The company already signed a partnership to this end with U.S.-based ride-hailing platform Lyft in October. The expectation is that wealthy owners may consider their vehicle as a financial asset and put it to work to increase its ROI as they do with their secondary homes or boats.
Will we see personally owned AVs developing massively beyond the premium market? I don’t see that happening for years to come. Why would the “average” person pay for the extra equipment to be driven while they have so many other potential uses for their funds? This could nevertheless happen if the incremental cost for L4 becomes marginal or the business model for sharing your personal AV makes financial sense. At that point, one could own a fleet of AVs for this sole purpose as people do on Turo, the peer-to-peer car sharing platform.
Marc Amblard
Managing Director, Orsay Consulting