Zoox has confirmed it will begin charging passengers for robotaxi rides in Las Vegas from 10 August, after the National Highway Traffic Safety Administration (NHTSA) granted the company a commercial exemption permitting it to deploy up to 5,000 units on public roads over the next two years. The announcement marks a milestone for a company that has spent the last year offering free rides across Las Vegas, San Francisco, Austin and Miami, covering more than three million miles and carrying almost a million passengers in the process.

Fares will follow a base charge plus time-and-distance model, priced to sit more or less in line with the comfort tier that human-operated ride-hailing apps offer. This is noteworthy: robotaxis have often been hailed as a means of bringing down the price of private transport, but Zoox will charge around 20-40% more than the standard fare. Any destination-specific charges, such as airport fees, will be disclosed upfront, and the company has said riders will not be charged extra if the vehicle ends up taking a longer route than planned.

Unlike most rivals, which retrofit familiar passenger vehicle models car with sensors, Zoox built its robotaxi from a wholly unique concept. The vehicle is bidirectional, with a carriage-style cabin containing two rows of inward-facing seats, four-wheel steering, and software designed to check cabin cleanliness and flag items riders leave behind. 

“Because we’re vertically integrated, we can do things that other companies can’t,” Co-Founder and Chief Technology Officer Jesse Levinson told Axios. A finalised production version of the Zoox robotaxi was revealed early in July, offering multiple iterative improvements including bidirectional reflectors refined and relocated for better visibility, as well as the door interface gaining an expanded two-way audio system.

Zoox is arguably the least established of the three main attendees, but also the most ambitious in their approach

NHTSA concluded the vehicle offers safety equivalent to a conventionally controlled car, but attached extra reporting requirements covering crashes and inappropriate stops. The approval follows a string of software recalls, most recently in July after a Zoox robotaxi failed to detect heavy smoke and drove into an active fire scene in Las Vegas, part of a wider pattern of driverless vehicles interfering with emergency responders that has drawn a pointed warning from regulators to the entire industry this year.

The commercial launch also marks a genuine step change for Zoox itself rather than a formality. Half a million riders and years of free testing have effectively been a long dress rehearsal; ramping its Hayward, California plant toward a target of 100 vehicles a week, while simultaneously running a live paid service, shifts the challenge from proving the technology works to manufacturing, maintaining and scaling it under commercial pressure. 

To be sure, 100 units a week would be quite a step up: Zoox has only built around 100 vehicles to date. The eventual goal is to reach 10,000 a year; Amazon, which paid US$1.2bn to acquire the company back in 2020, has said it expects to invest “double-digit billions” more to get there.

The launch also puts Zoox’s purpose-built approach to a test that testing alone couldn’t provide. Designing out manual controls entirely means there’s no fallback to a conventional vehicle platform if production or regulatory problems emerge, unlike rivals like Waymo which can always default to a human-driven version of the same car. Like Waymo—and Tesla, which is also planning to deploy a robotaxi without manual controls—it has the fallback of teleoperation, but even this comes with constraints that human-driven operations simply lack. 

Both Zoox and Tesla have spent years arguing that a robotaxi built with no human fallback is safer, not riskier, than one which keeps manual controls as a safety switch. Charging fares in Las Vegas is where that argument stops being a design philosophy, and there will be nothing to fall back on if either the vehicles or the business model come up short.