Roughly twenty billion dollars are sitting in a market that, by all accounts, should already have a clear winner. This figure is quietly discussed in industry decks and analyst notes. It doesn’t. Self-driving cars don’t have a Geico. Robotaxis does not have a progressive. The category is huge, growing, and strangely vacant at the top.
You can practically feel the hesitation when you walk through the lobby of any midsize insurance provider in Des Moines or Hartford. The task of pricing something that doesn’t behave like a human is being given to underwriters who have been trained on decades of human-driver data. A Level 4 robotaxi doesn’t text, drink, or become sidetracked. However, there are times when it misreads a construction cone or hesitates at a four-way stop in ways that no actuarial table has ever simulated. The twenty billion lives in that gap between what insurers know how to price and what is actually happening on the street.
It’s difficult to ignore the peculiar silence surrounding this. Years ago, Tesla introduced its own insurance product, claiming premiums that were about 20 to 30 percent lower than those of its rivals. The rollout has been gradual but not particularly explosive. Lemonade stirred. Root made an effort. Naturally, the large carriers are keeping an eye on things. Everyone seems to be waiting for someone else to construct the thing before stealthily replicating it.
The part that keeps lawyers up at night is the liability question. Who pays when a Mercedes Drive Pilot system, which is presently the only Level 3 setup authorized in California and Nevada, makes a choice that results in a collision? The driver wasn’t operating the vehicle. The vehicle was. Last summer, Goldman Sachs analysts released a note that suggested autonomous adoption could eventually cut insurance costs in half. However, the same note discreetly acknowledged that no one had figured out how to assign fault when the algorithm, not the human, made the decision.
Additionally, there is a more profound issue that is practically philosophical. Conventional auto insurance is a human-centered product. Age, marital status, driving history, and zip code. The entire system operates under the assumption that a flawed human is in charge. The entire underwriting logic falls apart if you remove that and replace it with a software stack that has been updated overnight through OTA patches. The code, not the client, would have to be priced by insurers. Not many have the engineering bench to accomplish that.
Last winter, a row of white Chevy Bolts waited for the morning shift outside the Cruise depot in the Mission District of San Francisco. No one at the company seemed to know exactly who insured what, a maintenance worker said, half-jokingly. At three in the morning, the cars, the routes, the riders, and the software updates were pushed. The policies that cover each layer are pieced together from older frameworks that were never intended for this, and each layer carries its own risk.
According to a research note published by Bank of America, autonomous vehicles may actually increase insurer profits by shifting liability to automakers. In the long run, that might be accurate. It produces a vacuum in the short term. Automakers have no desire to turn into insurance firms. The technology is not fully understood by insurance companies. Furthermore, there isn’t yet a large-scale embedded finance product that combines financing, software warranties, coverage, and roadside assistance into a single subscription linked to the car.
It’s easy to draw comparisons between this and the early days of mobile payments, when everyone acknowledged the enormous potential but argued for five years over whose logo should appear on the screen. The dominant product will eventually be built here by someone. It could be a carrier that no one anticipates. Tesla may be subtly continuing what it has already begun. It could be a business that doesn’t currently exist. The twenty billion have patience.