Robotaxis in 2026: the year the driverless car stopped being a demo

For fifteen years the self-driving car was permanently two years away. In 2026 that stopped being true. Not as a headline or a demo on a closed track, but as a thing you can open an app and pay for, watch pull up with nobody inside, and ride across a real city in real traffic. The technology did not suddenly arrive. It crossed a line, quietly, from experiment to service.

That is worth being precise about, because "self-driving cars are here" has been a lie for so long that it is easy to miss the moment it became true. So here is the honest state of it, with the numbers that back it up and the parts the marketing leaves out.

Where can you actually ride a robotaxi in 2026?

In more places than most people realise, and the map is redrawing itself every few months.

The leader by a wide margin is Waymo, the self-driving arm of Google's parent company. It runs a fully driverless, public, paid service, no human in the front seat, and by mid-2026 it provides around 500,000 rides a week. It started in Phoenix, San Francisco, Los Angeles and Austin, and in 2026 it opened to the public in a wave of new cities including Dallas, Houston, San Antonio, Orlando and Miami, with Washington, D.C. close behind. It is also laying the groundwork abroad, with Tokyo and London named as early international targets. To fund all this it raised a $16 billion investment round in February 2026, which tells you how seriously the money now takes it.

You are not waiting for this. In several large American cities, hailing a car with no driver is already a mundane Tuesday.

Is a robotaxi actually safe?

This is the question everyone asks, and it is the one where the data is most surprisingly one-sided.

By Waymo's own published safety reporting, across more than 220 million miles driven with no human behind the wheel, its cars were involved in 94% fewer crashes causing serious or fatal injuries than human drivers would have been over the same miles in the same cities. The same data shows roughly 82% fewer crashes involving any injury and large reductions in collisions with pedestrians and cyclists specifically.

Two honest caveats. First, this is the company's own data, though it publishes the methodology and much of it has been examined by outside researchers and insurers. Second, "safer on average" is not "flawless": these cars still have low-speed bumps, still occasionally freeze in confusing situations and block traffic, and a rare serious failure gets far more attention than the thousands of crashes they quietly prevent. But the direction of the evidence is clear, and it is the opposite of what most people assume. The machine is already the safer driver in the places it operates.

What happened to Tesla's robotaxi?

Tesla is the name most people expect to lead this, and it is the clearest example of the gap between promise and delivery.

Tesla launched a Robotaxi service in Austin in June 2025, and it is real. But it arrived years after Waymo, it started with human safety monitors sitting in the cars, and through early 2026 it remained small: the genuinely driverless part was only a couple of dozen cars in Austin, while its San Francisco Bay Area service still ran with a safety driver in the seat, because Tesla does not hold the driverless permits California requires, the same ones Waymo already has there. Through 2026 the pace picked up: Tesla expanded the service into more cities, adding Houston and Dallas and, by August, Orlando and Tampa, sometimes within weeks of its mapping cars appearing, though US safety regulators have begun examining reported incidents of its robotaxis speeding and crossing lane lines. Set that against what was promised: for years the plan was that existing Teslas would become robotaxis through a software update, covering most of the country. The reality on the ground is a modest, closely watched pilot. It may still get there, and Tesla's approach of using cameras alone rather than expensive sensors could scale cheaply if it works. But in 2026, Tesla is a follower, not the leader, and the distance between the pitch and the product is the story.

The race the West is barely watching: China

The most under-covered part of this story is that the country scaling robotaxis fastest may not be the United States.

Baidu's Apollo Go operates across more than a dozen cities in China and, increasingly, the Middle East, and by some global rankings has pulled ahead of Waymo on raw scale. In Wuhan alone it has run over 1,000 vehicles, and Baidu says the unit has reached per-vehicle profitability there, a milestone no Western operator has publicly claimed. It is now expanding outward, with a paid driverless permit in Abu Dhabi and trials heading into Europe. It is not alone: WeRide and Pony.ai each run large robotaxi fleets across Chinese cities and the Middle East, and Amazon's Zoox has opened a public robotaxi service in Las Vegas using a purpose-built pod with no steering wheel at all.

If you think of this as a two-horse American race, you are looking at the wrong map.

The players at a glance

Operator	Backed by	Where it runs	Notable
Waymo	Alphabet (Google)	US: Phoenix, SF, LA, Austin + expanding	~500k rides/week; strongest safety record
Baidu Apollo Go	Baidu	10+ Chinese cities, Abu Dhabi	1,000+ cars in Wuhan; profitable there
Tesla	Tesla	Austin + Houston, Dallas, Orlando, Tampa	Expanding fast; camera-only bet; under regulator scrutiny
WeRide / Pony.ai	Independent (China)	China, UAE, Gulf	Large, fast-growing fleets
Zoox	Amazon	Las Vegas (paid rides from Aug 2026)	Purpose-built pod, no wheel; recalled 105 cars after a June incident

What still stands in the way

The technology working is no longer the hard part. The hard parts now are unglamorous:

• Cost and profit. Building, sensing and supporting these cars is expensive, and outside a few dense markets the economics are still unproven. Profitability in Wuhan is a signal, not a settled case.
• Weather and edge cases. Most deployments favour cities with mild, predictable conditions. Heavy snow, chaotic construction and unmapped areas remain genuinely hard.
• Cities built for humans. Double-parked trucks, hand signals from a traffic cop, a festival that closes a street: the world is full of situations that a careful machine handles slowly, if at all.
• Trust and jobs. A single dramatic failure moves public opinion more than a million safe miles, and a future with fewer driving jobs is a real cost that the technology does not solve.

So is the driverless future here?

Yes, and no, and the gap between those two answers is the whole point. In a dozen cities, for the specific trips these cars are cleared to make, the driverless future is a normal service you can already use, and by the numbers it is safer than the alternative. Everywhere else, it is still a promise, and the last stretch, from "works in mapped, mild-weather cities" to "works anywhere," has always been the longest one.

The useful stance is neither the breathless "everything changes next year" nor the tired "it will never work." It already works, in more places than the sceptics admit and far fewer than the boosters imply, and it is spreading one carefully mapped city at a time. That is less dramatic than either headline, and it is what actually happening looks like. For more from the edge, see the Future section, and our reporting on the AI that makes these cars drive.