Tesla's robotaxi miles fell 36% last quarter as stock plunges 13% on earnings miss

Tesla's robotaxi footprint grew to six cities. Its paid miles went the other way.

Tesla spent the second quarter of 2026 adding territory to its robotaxi network. The service that came out the other side was smaller by the only metric that matters to a ride-hailing business: paid miles driven.

Tesla's robotaxi fleet covered roughly 1.1 million paid miles in the first quarter. In the second, that fell to about 700,000, a decline of around 36% 1. The numbers come from a chart Tesla released alongside its Q2 results. At a glance the chart looks like growth, because the figures are cumulative. Broken into quarters, they show the opposite 1.

The mileage decline surfaced next to Q2 earnings that fell short of Wall Street's profit expectations and sent the stock down more than 13% 1. The earnings miss grabbed the headlines. The robotaxi chart tells the harder story.

More cities, fewer rides. The company expanded to six cities across Texas and Florida with a mix of supervised and unsupervised vehicles 1. Paid mileage fell anyway. That is the gap between a pilot and a product, and it is widening at the exact moment Tesla needs it to close.

Why the network shrank during an expansion push comes down to what Elon Musk admitted on the Q2 earnings call. He told analysts that Tesla needs to accumulate real-world driving data specific to the Cybercab, the company's purpose-built two-seater sedan, before deploying large numbers of them 1.

The admission cuts against the foundation of Tesla's autonomy pitch. For years, the company has pointed to its fleet of nearly 10 million customer cars silently collecting driving data in the background as the training advantage that would make a robotaxi fleet inevitable 1. Now Musk is saying that advantage does not transfer. The Cybercab has a different chassis than the Model Y vehicles currently running the service, and calibrating for it requires test cars retrofitted with steering wheels and pedals. There are not enough of them yet 1.

The millions of Teslas already on the road were supposed to be the moat. They cannot fill the gap where the Cybercab sits.

Musk also swapped his explanation for why robotaxi deployment is slow. For years, Tesla framed regulators as the primary obstacle to full-scale autonomy, without specifying which regulations were prohibitive 1. On this call, the story changed. Musk and Tesla executives framed the cautious pace as a safety decision. He said he does not want Tesla robotaxis causing accidents because a single injury could generate worldwide press coverage and prompt regulators to shut the operation down 1.

Ashok Elluswamy, Tesla's VP of AI, told the call that the robotaxi fleet has logged more than 380,000 miles without a safety operator and had zero "notable incidents" 1. He did not define what the company considers notable. That definition matters, because Tesla has reported 22 crashes to the National Highway Traffic Safety Administration in the year since its robotaxi trials began. Three were caused by remote teleoperators steering vehicles into collisions. Others involved the cars hitting curbs, utility poles, and a tow truck's bed at low speeds 1.

Zero notable incidents. Twenty-two crash reports. Whatever sits inside the gap between those two numbers is where the word "notable" earns its keep.

There is a geographic asterisk on the mileage figures, too. Tesla likely counts paid miles driven in the San Francisco Bay Area as part of its robotaxi coverage, even though those vehicles lack the state permits required to operate autonomously and carry a safety driver behind the wheel 1. If those supervised miles are inflating the total, the actual autonomous-service volume is lower than even the declining chart suggests.

Musk and Elluswamy both promised that growth is around the corner. They noted that unsupervised robotaxi miles have grown roughly 10% each week since Tesla started offering them at the end of last year 1. Elluswamy also used the moment to argue that Tesla's camera-only approach proves lidar, radar, and HD maps are unnecessary for safe autonomous driving 1.

For builders and investors tracking autonomous ride-hailing, the shape of this story is now legible. Tesla is adding cities while losing paid miles. It is conceding that its data advantage does not extend to the vehicle meant to carry the business model. It is swapping a regulatory villain for a safety rationale while reporting crashes it categorizes as not notable. And the Cybercab, the vehicle built to make robotaxis profitable at scale, cannot scale until a small test fleet gathers enough calibration data.

The distance between a working demo and a real transportation service gets measured in paid miles. Right now, that number is moving in the wrong direction.