Tesla's robotaxi promises are clashing with reality

A Mileage Gap the Numbers Can't Hide

On the surface, Tesla's headline figure sounds impressive: 380,000 unsupervised miles across six cities in two states, with Musk claiming a 10 percent week-over-week growth rate. But context collapses the boast. Waymo, using lidar and radar alongside cameras, runs roughly 4 million miles a week — meaning Tesla's entire year-one footprint is less than 10 percent of one of Waymo's typical weeks. Crowdsourced trackers put the active unsupervised fleet at just 16 vehicles in Austin, four in Dallas, and one in Houston. A program that was supposed to demonstrate exponential scaling instead looks like a fragile pilot.

The fleet composition also matters. Tesla runs a mix of supervised and unsupervised vehicles, and the ratio is in constant flux, which makes any "growth" claim difficult to verify. Musk's framing — that the service is expanding rather than concentrating — is a strategic choice about proving geographic generalization, but it has the side effect of dispersing a small fleet across too many under-resourced markets. Investors reading the disclosures, like BNP Paribas analyst James Picariello, are no longer buying the spin; his note to clients used the phrase "an alarming decline."

The Camera-Only Bet Meets Regulatory Reality

Head of AI Ashok Elluswamy used the earnings call to relitigate an old argument: that cameras alone are sufficient for safe autonomy, dismissing the "kitchen sink" of lidar, radar, and HD maps that competitors rely on. The pitch is that Tesla's lower hardware bill of materials and absence of high-definition map dependencies will let the service scale anywhere. In theory, that is a powerful moat. In practice, regulators are pushing back.

New Jersey is considering legislation that would effectively ban robotaxis lacking redundant sensors like radar and lidar. Morgan Stanley analyst Andrew Percoco flagged this on the call, and Tesla VP of vehicle engineering Lars Morvay conceded it was "a little bit disheartening." A patchwork of state-level sensor mandates would fragment Tesla's "drive anywhere" thesis into a quilt of restricted operating domains — the opposite of the generalized stack Elluswamy wants to demonstrate. If the camera-only approach runs into regulatory walls in even a handful of states, the total addressable market contracts sharply.

Safety, Secrecy, and the Trust Deficit

Musk's quip about not wanting to "run over a pet" was a pointed jab at Waymo's well-documented animal-strike incidents. But the rhetorical deflection papers over Tesla's own transparency problem. The Robotaxi Tracker records roughly 18 incidents, 56 percent attributed to other drivers — a figure that would be reassuring if Tesla published crash data with the same granularity NHTSA requires of human-driven fleets. Instead, the company redacts the substantive details, leaving outside observers to triangulate from crowdsourced reports.

The Federal Self-Driving investigation adds another layer of risk. Tesla's robotaxis run FSD v15 while customer vehicles sit on v14.3.6, meaning the public-road autonomy stack diverges from the version regulators are actively examining. A recall or mandated software change targeting the customer fleet could spill into the robotaxi build, or expose gaps between the two. For a business model predicated on safe, unsupervised operation at scale, opacity around incident data is not a minor communications issue — it is a structural liability that compounds every quarter the numbers stay flat.