Tesla's path does not start with Robotaxi: in 2015, it first integrated assisted driving into mass-produced cars with Autopilot, and in 2016 claimed that new production vehicles had the hardware necessary for future full autonomous driving, selling FSD as a paid feature. The issue lies in the product naming and capability promises running long ahead of regulatory validation; U.S. regulators later initiated reviews around FSD performance in low visibility, focusing on the system's failure to recognize capability degradation and insufficiently warning drivers.

The capital path has shifted from 'selling a car once' to 'vehicles collecting data—training clusters iterating—software confirming revenue—Robotaxi service monetization.' Tesla is expanding its Texas Cortex training cluster, building Cortex 2, and expects capital expenditures to exceed $20 billion by 2026; its 2025 documents also list FSD, Cybercab, and autonomous driving mobility services as part of the same growth and infrastructure deployment plan.

Unlike Waymo, which first operates driverless taxis in limited areas and then expands through external financing, Tesla attempts to turn sold vehicles into training terminals and potential service providers. Waymo completed $5.6 billion in financing in 2024 to expand its autonomous ride-hailing service; Tesla relies on the software penetration rate of its existing fleet and is currently transitioning from driving assistance subscriptions to a mobility network.

This change is essentially a shift in pricing power: the price of cars is determined by hardware costs and brand, while ongoing software charges depend on data loops, computing power investment, and regulatory permissions. If continuous no-takeover mileage can be replicated across different roads, weather, and drivers, pricing power will concentrate on platforms that control fleet data and model updates; if it cannot be replicated across scenarios, capital expenditures will precede cash flow realization.

ABAB News · Cognitive Laws

Hardware is sold once, software is sold for a lifetime
Data is not a moat, a closed loop is
The valuation of technology ultimately settles on replicability