The question that drove the research for my first book, The Big Data Opportunity in Our Driverless Future, and the two that followed was how AI will change ground mobility. Our firm’s recent research shows that consumers focus on three factors when acquiring new vehicles. First, the services they value, such as driving automation and driver safety. Second, the total cost of owning the vehicle (TCO), including insurance and the vehicle’s residual value. Third, the convenience, where in-cabin personalization and mobility services play a role. Of the three, TCO concerns mainstream consumers most acutely.
At the recent Paris Auto Show, I saw progress in two areas I have been writing about: AI for driving automation and AI for cabin personalization. I also saw European automakers introduce moderately priced vehicles that can compete with Chinese imports. I did not see an effort to lower what an electric vehicle costs to own after its purchase. There are reasons for that absence, and technology alone will not remove them.
What Paris showed
AI had a prominent role in Paris. Tesla and XPeng were the only automakers in the show’s public-road program, where they demonstrated supervised automated driving. Many of the 20 Chinese brands that exhibited promoted the driver-assistance capabilities of their vehicles, and according to the organizers, 2,700 of the show’s 6,000 test drives were of supervised driving systems. The automakers aimed this AI at premium buyers. Reuters reported that Chinese brands used the show to pitch “intelligent luxury” to Europe’s premium buyers. XPeng, for example, launched the G9L, whose driving system relies on a second-generation vision-language-action (VLA) model, at a starting price of €74,990. In the cabin, AI assistants are reaching mainstream models. Hyundai’s new Tucson includes the Gleo AI voice assistant, which draws on live internet data.
SBD Automotive’s recent testing in China shows what cabin AI can already do for mainstream tasks. In the Li Auto L9, a driver can ask the assistant to pay for parking, and the vehicle’s camera scans the payment code and completes the payment without the driver taking out a phone. SBD also found that the assistant sometimes confirmed an action without completing it. Chinese automakers must now show that such features work with European languages and services.
European automakers focused on small, affordable cars, including electric versions of models people remember. Stellantis showed a revived Citroën 2CV that, according to Reuters, will use new rules that let automakers drop some features to offer more affordable EVs. Renault showed its electric Renault 4 and Renault 5. These vehicles lower the purchase price. No exhibitor showed anything that lowers the vehicle’s TCO.
Where the EV owner’s money goes
AAA estimates that owning and operating a new vehicle in the United States costs $12,863 a year. Depreciation is the largest item: the average vehicle loses $4,422 in value each year, about a third of the total, and an electric medium sedan loses $6,786. AAA also found that EVs cost up to 70 percent less to charge than gasoline vehicles cost to fuel, yet cost more than $3,000 a year more to own because they depreciate faster.
The same pattern holds in China, where a three-year-old EV resells for about 45 percent of its original price, down from almost 55 percent in 2023. In Europe, Ayvens’ 2026 Car Cost Index, which covers 30 European countries, finds that depreciation accounts for 42 percent of a battery-electric car’s total cost of ownership, compared with 32 percent for a gasoline car, while energy accounts for only 12 percent of the electric car’s cost, compared with 26 percent for the petrol car.
The batteries do not explain the depreciation. Tesla reports that its Model 3 and Model Y Long Range batteries lose about 15 percent of their capacity after 200,000 miles. Geotab’s study of more than 22,700 EVs found that batteries lose 2.3 percent of their capacity a year on average. Researchers at Stanford found that real-world driving extends battery life by up to 38 percent compared with laboratory tests. The batteries last longer than the market believes. Dealers price used EVs mainly on age, mileage, and warranty, because they don’t have a way to measure battery health. BEUC, the European consumer organization, has called for a certified battery health report for every used EV.
How AI could lower the cost of ownership
AI can address the largest TCO item directly. A model that uses a vehicle’s own charging and driving history to predict how its battery will age can give the next buyer and the lender a residual value they can rely on. A higher, more credible residual value lowers both depreciation and the lease payment, because the lender prices the lease on the residual value it assumes. The same model can plan the vehicle’s charging to slow battery wear; Geotab found that vehicles relying on fast charging above 100 kW lose capacity about twice as fast as vehicles charged mostly at lower power.
AI can also lower the other large items. Driving data and the performance of the vehicle’s driver assistance can be used to price insurance for each vehicle. A vehicle that diagnoses its own faults can decide whether a fault needs a repair, a software update, or no action, and can have the parts ordered before the owner reaches the service center.
Paris showed none of this
The first reason is the way most incumbents develop software. Each of these AI applications needs a closed data loop. The automaker
- collects data from customer vehicles,
- trains models in the cloud, and
- deploys the models back to the vehicles.
Most incumbents still buy electronic control units as closed boxes from their suppliers and release software on model-year cycles. A Software-Defined Vehicle using a zonal architecture helps. A zonal architecture provides better access to the vehicle’s data. It also lets the automaker update the whole vehicle over the air. But an automaker can build a zonal vehicle and still lack the data loop. As I wrote in The Flagship Experience, the automaker must own its software stack and release software continuously.
The second reason has nothing to do with technology. To lower the buyer’s insurance or guarantee a residual value, the automaker must carry the financial risk of its own predictions. BYD has shown that an automaker can make this choice. Since June 2026, BYD has paid for the losses from accidents that occur while owners in China use its urban navigation assistance or smart parking.
Tesla has made a narrower version of the same choice. It sells its own insurance in 15 states. Since April 2026, it has given every mile driven on FSD a perfect Safety Score. This lowers the premium on new policies in six of those states. For most owners, however, that discount offsets part of the FSD subscription. It doesn’t lower the cost of owning the vehicle. Most automakers have not made this choice. Their captive finance arms have already taken losses on BEV leases whose residual values were set too high. Those losses make them reluctant to stand behind new predictions.
The third reason is consent. The data that feeds these models describes where and how people drive. Automakers can use it only with the driver’s agreement. GM learned this when the Federal Trade Commission barred it from sharing drivers’ location and driving data with consumer reporting agencies for five years. In exchange for the data, the automaker must offer the driver a visible benefit. For example, a lower premium or a guaranteed resale value.
Conclusion
Paris showed that automakers have made real progress with AI for driving and for the cabin. The exhibiting European automakers intend to compete with Chinese imports on purchase price. Paris did not address the cost that concerns mainstream buyers most, which is what an EV costs to own. Depreciation is the largest part of that cost.
Lowering that cost does not require automakers to wait for the AI-defined vehicle. It requires properly built zonal SDVs with a closed data loop, a willingness to carry the financial risk of the models’ predictions, and the driver’s consent to use the data. The AI-defined vehicle will let the vehicle manage itself. However, it will learn only from the data loop that automakers build now. The first automaker to stand behind a prediction of its batteries’ health will compete with the Chinese imports on the factor mainstream buyers care about most.