Self-driving technology will become a commodity. Operating it will not.
Two days ago, Bolt CEO Markus Villig wrote that DeepSeek and Kimi (Moonshot AI) had demonstrated how Chinese open-source laboratories are commoditising frontier AI, adding: “Most people don’t realise that self-driving cars are next.”
I think he is half right.
Self-driving capability will probably advance much faster from here. Chinese autonomous-driving companies are already moving into Europe, and mobility platforms such as Bolt are forming partnerships with technology and vehicle providers. What looks like a rare technological advantage today could become something manufacturers and operators can source from several companies sooner than many expect.
But a self-driving car is not simply an AI model on wheels. It is a safety-critical system combining software, sensors, computing, vehicle integration and enormous amounts of validation. It must cope with weather, unusual road layouts, human behaviour and local regulation. Reaching the same level quickly will therefore be much harder than reproducing the performance of a language model.
Robotaxis make the problem more manageable. They can operate within a geofenced area, use centrally maintained vehicles and rely on remote support. A service in Luxembourg (Bolt), Zagreb (vene.io) or perhaps on an island does not have to solve every road in Europe before it can begin. Bolt, Pony.ai and Stellantis are following precisely this approach with their Luxembourg pilot.
Yet this is where the second half of the challenge starts. Somebody still has to charge, clean, position, monitor and recover the vehicles. The service needs sufficient utilisation, sensible pricing, insurance, local permission and a convincing reason for customers to use it.
Bolt has good reason to believe that the driving technology will become more widely available. It is positioning itself as the operating platform while partnering with Pony.ai, Stellantis and NVIDIA for different parts of the autonomous ecosystem.
If the autonomous-driving stack becomes a commodity, much of the value may move from building the “driver” to operating the service around it.
My conclusion is therefore slightly different from Markus Villig’s.
Self-driving capability may arrive surprisingly quickly. A viable autonomous mobility system will take longer.
The winner may not be the company with the most impressive demonstration. It may be the one that learns how to operate thousands of autonomous vehicles reliably, safely and profitably in the places where people actually need them.