Robotaxis

Gregory Chassapis

Jul 29

6 min read

Imagine if your car was a cash flow-generating asset.

For most people it is the opposite. A car is among the largest purchases a household makes, and it has one of the worst cost profiles of any commonly acquired asset (insurance, registration, and depreciation accrue regardless of whether it moves). It also sits idle roughly 95% of the time, which raises the question: is there a way to monetize that time?

In theory, yes - particularly in more urban areas, where transportation remains a costly and inconvenient affair.

Autonomy was once the domain of science fiction, but thanks to advancements in artificial intelligence and automotive technology, the first step in monetizing that idle asset is here in the form of robotaxi networks. How it will scale is another question entirely.

Scalability: Technology

Scalability in any industry is heavily dependent on the economics making sense, but with robotaxis, there are two difference-makers rather than one. The second comes later, but the first is the technology stack, and understanding it requires accepting that a robotaxi is not just a car. It’s a perception system, a prediction engine, and a planner wrapped in a vehicle, and the perception layer is where the industry's central tech disagreement lives.

At present, there are two major approaches to autonomy: Camera-Only (Tesla) and Multi-Sensor (Waymo, Zoox, Baidu, etc.).

Tesla’s camera-only approach involves the use of cameras and neural networks (software that learns patterns from examples rather than following rules a programmer wrote). The idea here is that since Tesla has millions of camera-equipped cars on the road that selectively upload driving clips (particularly disengagements and rare edge cases), the company can use driver data that often displays the correct way to deal with any given situation to continuously train the neural network behind its autonomous driving software. The more useful data it ingests, the better it gets, since the program relies on trying to match what the cameras “see” to the known “correct answers.” Think of it as pattern recognition at scale in that the system infers what to do based on millions of examples of humans doing it.

On the other hand, the multi-sensor approach simultaneously runs cameras, LiDAR, and radar to build a picture of the world around the vehicle, accepting a costlier vehicle and sensor stack in exchange for sensors that precisely measure depth and velocity rather than inferring from pixels. As with Tesla’s approach, neural networks continue to process data, but the multi-sensor system uses pre-collected fleet, safety-driver and simulated datasets rather than consumer driver data.

Both approaches require per-market validation and regulatory approval, and both currently send survey vehicles into cities ahead of launch. For multi-sensor, HD maps are a runtime dependency the vehicle references while driving and must keep current. Tesla's survey work is calibration of a model. If the thesis behind it is correct, the cost and frequency of doing that should decay over time as the model improves.

Scalability: The Cars

This brings us back to the car in the driveway. Tesla's Cybercab, a two-seater with no steering wheel or pedals, entered production in February 2026 with a promise to completely revolutionize urban transportation. CEO Elon Musk also described it as operating within a system in which Tesla owners add and subtract their own vehicles from the fleet (similar to how Airbnb works), which essentially converts a depreciating and idle asset into a revenue-producing one during the hours it would otherwise sit. That model is currently the most disruptive idea in the sector because it accelerates fleet expansion thanks to the creation of what is essentially a marketplace. Unfortunately, that very marketplace can only be populated by vehicles with the correct internal hardware. Older vehicles (those built prior to 2023 and depending on their hardware) will likely not be eligible, but the idea remains.

Scalability: The Economics

Every argument above eventually comes down to all-in cost per mile and estimates vary.

Today, ARK Invest puts Tesla's Model Y robotaxi just above $0.60 per mile. Morgan Stanley puts the same vehicle at $0.81 (roughly 35% higher) against $1.43 for Waymo. At scale, ARK sees Waymo settling near $0.40 per mile by 2030 and the Cybercab at $0.20, a figure Musk has called "probably true." Morgan Stanley is more conservative, modeling $0.37 for the Cybercab by 2035.

For context, ARK Invest estimates a human-driven ride-hail mile in the US cost about $2.80 in 2025.

The per-unit numbers help explain the gap. Per Jaguar, a base I-PACE launched at an MSRP of $72,500 (the car is no longer in production, but it represents a large portion of Waymo’s existing fleet). A planned Hyundai Ioniq 5 variant runs above $50,000. Neither figure includes the technology stack itself. By contrast, the Cybercab’s all-in cost is estimated at anywhere between $20,000 to $30,000.

Practical Points to Consider

While the future is an exciting place, there are obstacles to mass adoption. These are still robots, after all, and while their performance is often impressive, current iterations are not without fault.

During a December 2025 San Francisco power outage that knocked out signals and cell networks, Waymos halted at intersections citywide, and the mayor phoned the company's chief executive to have them removed. San Francisco firefighters have filed multiple internal reports since April 2025 of robotaxis obstructing emergency operations, including blocked firehouse exits.

A review of municipal complaint databases by CNN found vehicles running red lights, entering flooded and closed roads, and ignoring school-bus stop arms. Any aspiring robotaxi network operator will need to ensure these types of incidents do not happen. The good news is that these things are solvable and IIHS data has shown that autonomous vehicles are statistically safer than human-operated vehicles.

That said, scaling a network ultimately requires a solution to the vacuum left by the driver, because the driver was never there just to operate the vehicle - he was also the reason the back seat was clean when a customer got in.

And that’s the second difference-maker.

Waymo handles that with human crews at its depots, including contractors trained for biohazard cleanup. Tesla's answer is automation, which is to say that cabin cameras flag messes and trigger automatic fees based on the degree of cleanup required. A robotic arm will then proceed to vacuum, collect trash, and wipe surfaces. Tesla has indicated that this infrastructure will be available at its robotaxi depots and supercharging stations, but in a situation where any robotaxi is not in an infrastructure-rich area, that could create a large backlog of unavailable units.

Therefore, the key component is the depots themselves. Stuck cars, depleted batteries and dirty back seats are problems that are solved with pristine operations, not better self-driving software, which means the winners will be decided by the infrastructure surrounding the technology rather than solely by the technology itself. Unit availability depends on it. Manufacturing a Cybercab for $20,000 and enjoying its advantageous economics won’t matter if the end user can’t hail it.

Looking Ahead

Autonomy has crossed from demonstration to deployment, to the point where people are beginning to declare an inflection point. Hundreds of thousands of people already use some version of the representative software daily, and half a million weekly robotaxi rides are already happening in the United States. High-utilization electric fleets that replace an asset sitting idle most of its life represent one of the better transport ideas of this century, precisely because of the problems they solve. The technology remains imperfect, but it has improved markedly, and if these companies can solve depots, permits, and public trust, autonomy becomes the default way cities move.

Sources

Disclaimer: The content contained herein is provided for general informational and educational purposes and does not constitute investment advice or a recommendation, offer, or solicitation to buy or sell any securities. The content reflects the writer's views and analysis as of the time of writing and is provided for context only. It does not address every factor relevant to any particular investor's circumstances, and investors should evaluate their own facts and circumstances before making any investment decision. The writer and/or affiliated funds may hold positions in the securities discussed and may buy or sell such positions at any time without notice. Past performance is not indicative of future results.