Robotaxis in 2026: Where Self-Driving Cars Actually Operate

Robotaxis stopped being a demo in 2026 and became a real, paid transportation option in more than a dozen American cities. Waymo alone is now giving over half a million driverless rides every week, Tesla has put unmanned cars on the streets of Austin, and Amazon's Zoox is shuttling people around the Las Vegas Strip. But the gap between the hype and the map is enormous: true self-driving is still confined to a handful of metros, one major player (GM's Cruise) is dead, and the most-watched debate of all — Tesla's "Full Self-Driving" — is still legally a system that requires a human. Here is where things actually stand in mid-2026, and why it matters for what you drive.

The state of play: three companies, very different bets

After roughly a decade and tens of billions of dollars, the U.S. robotaxi race has narrowed to three serious commercial operators, plus a graveyard of expensive failures. The leaders are not betting on the same thing — and understanding the difference is the key to reading every headline.

Operator	Where it runs (mid-2026)	Who's "driving"	Roughly how big
Waymo (Alphabet)	Phoenix, San Francisco, Los Angeles, Austin, Atlanta, Miami, Dallas, Houston, San Antonio, Orlando	SAE Level 4 — no human in the car	~3,000 vehicles; 500,000+ paid rides/week
Tesla Robotaxi	Austin (metro-wide), plus Dallas and Houston launches	Mostly safety monitors; limited fully unsupervised	~135 in service (late 2025), but only ~20 carrying riders without a safety monitor
Zoox (Amazon)	Las Vegas, San Francisco; testing in Austin and Miami	SAE Level 4 — purpose-built, no steering wheel	~350,000 riders served; paid rides rolling out in 2026

The single most important thing to understand is the technical and legal distinction between these services — because it determines who is responsible when something goes wrong.

Level 4 vs. Level 2: who is actually driving

Engineers classify automation on the SAE J3016 scale that NHTSA officially uses, running from Level 0 (no automation) to Level 5 (drives anywhere a human can). Two rungs matter for this story:

• Level 2 (partial automation): The car can steer and control speed at the same time, but the human must watch the road continuously and is legally the driver. Tesla's "Full Self-Driving (Supervised)," GM's Super Cruise, and Ford's BlueCruise all live here.
• Level 4 (high automation): Within a defined area and set of conditions, the system performs the entire driving task — including the fallback if something fails — with no human needed. This is what makes a true "robotaxi" with an empty driver's seat legal.

Waymo and Zoox operate at Level 4. Tesla's consumer software is Level 2, no matter what the name suggests, and the company itself sells it as "Supervised." That is why a Tesla owner is on the hook for a crash while using FSD, but a Waymo passenger is not. As covered in our look at where car prices are heading in 2026, this distinction is also reshaping liability and insurance — the carmaker, not the buyer, owns the risk at Level 4.

Waymo: the runaway leader

Waymo is no longer a science project; it is a scaled business. The Alphabet subsidiary now provides more than 500,000 paid rides every week across 10 U.S. cities — a tenfold jump from about 50,000 weekly rides in May 2024. It does this with a fleet of roughly 3,000 vehicles built on its fifth-generation "Waymo Driver" hardware.

The footprint has exploded. Beyond its original markets of Phoenix, San Francisco, and Los Angeles, Waymo added Austin and Atlanta (offered exclusively through the Uber app in those two cities), Miami, and an aggressive Texas-and-Florida push into Dallas, Houston, San Antonio, and Orlando. The company has confirmed further Texas and Florida cities for 2026, along with planned launches in markets like Washington, D.C., Nashville, and San Diego.

Two 2026 milestones moved Waymo from "novelty" to "actual replacement for a car":

• Freeways. In November 2025, Waymo began carrying passengers on freeways in San Francisco, Phoenix, and Los Angeles, which it says cuts some trip times by up to half. Surface-street-only robotaxis were a tough sell for longer journeys; freeway access changes the math.
• Airports. Waymo already serves Phoenix Sky Harbor and San Jose's airport, and on January 29, 2026 it began rides to and from San Francisco International (SFO), initially from the Rental Car Center. Airport runs are the highest-value rides in any ride-hail business.

The safety case

Waymo's pitch rests on a growing pile of data. Through December 2025 the company had logged 170.7 million rider-only miles — driving with no human behind the wheel. Compared against human-driver crash rates in the same neighborhoods, Waymo reports 82% fewer injury-causing crashes, 92% fewer serious-injury-or-worse crashes, and 92% fewer pedestrian injury crashes. Independent reviewers have backed the trend: a peer-reviewed analysis at 56.7 million miles found statistically significant reductions across injury and airbag-deployment crashes, and reinsurer Swiss Re measured far fewer bodily-injury and property-damage claims.

That doesn't mean Waymos are flawless — they still get stuck, block traffic, and occasionally need remote help. But the directional evidence that a mature Level 4 system is safer than the average human is now hard to dismiss.

Tesla Robotaxi: the high-stakes bet

Tesla took the opposite approach. Instead of building a custom vehicle bristling with lidar like Waymo and Zoox, it bet that camera-only software trained at fleet scale could deliver autonomy in any Tesla. The robotaxi service is the test of that thesis.

Tesla launched its Robotaxi service in Austin on June 22, 2025, using modified Model Y vehicles — but with a human "safety monitor" sitting in the front passenger seat, ready to intervene. That is a crucial caveat: a supervised ride is not a driverless ride. In January 2026 Tesla began integrating genuinely unsupervised vehicles into the Austin fleet in a limited way, and by spring 2026 it had expanded the Austin service area metro-wide and begun launches in Dallas and Houston, with plans floated for Phoenix, Miami, Orlando, Tampa, and Las Vegas.

But the scale is tiny next to Waymo. Tesla reported roughly 135 robotaxis in service in late 2025 — though that figure counts cars used for Bay Area employee testing, and Texas state filings show only about 42 autonomous vehicles registered there. Independent trackers found as few as 20 cars actually carrying passengers without a safety monitor in mid-2026, a fraction of Waymo's thousands, with limited daily availability. Tesla's advantage, if the software proves out, is cost: it doesn't need a bespoke $100,000-plus vehicle. Its risk is that "good enough to impress on a demo" and "good enough to remove the human entirely, everywhere" are separated by a very long, very expensive tail of edge cases.

The regulatory cloud

Tesla's autonomy claims are under intense federal scrutiny. In March 2026, NHTSA escalated its Full Self-Driving probe to a formal Engineering Analysis (EA26002) covering roughly 3.2 million FSD-equipped vehicles, after crashes tied to the system failing in conditions like sun glare and fog. Separately, the agency has fielded complaints of FSD running red lights or entering wrong-way lanes. Tesla's robotaxi fleet itself reported about 14 crashes over roughly 800,000 miles through early 2026 — a rate that critics flag and supporters argue is early-stage noise, especially since several incidents were low-speed rear-endings of a stopped Tesla. Either way, every one of those incidents is public because of one rule.

How regulators watch all of this

There is no single federal robotaxi license. Instead, the U.S. relies on a patchwork: NHTSA sets vehicle safety standards and investigates defects, while states (California, Arizona, Texas, Nevada, Florida) decide who can actually deploy on their roads. California's DMV and Public Utilities Commission, in particular, hold real power — they can and have suspended permits.

The connective tissue is NHTSA's Standing General Order on crash reporting, which requires every operator of an automated driving system — and many Level 2 systems — to report qualifying crashes, in some cases within a day. That order is why we can compare Waymo's and Tesla's safety records at all. The ongoing fight is over transparency: Tesla has been criticized for redacting crash details as confidential business information, though it later unredacted a set of robotaxi crash narratives under pressure.

The cautionary tale: GM Cruise

It's worth remembering how fast this field can humble a giant. In December 2024, after a high-profile San Francisco crash and a permit suspension, General Motors pulled the plug on Cruise robotaxi development — having spent more than $10 billion. GM folded the team into its broader engineering org and redirected the technology toward personal-vehicle driver assistance like Super Cruise, citing the time, capital, and competition required to run a taxi fleet.

The lesson for buyers: robotaxi ambitions and the driver-assist features in cars you can buy are now diverging. The autonomy GM is shipping to consumers is hands-off-but-eyes-on Level 2 — not a self-driving car.

Zoox: Amazon's purpose-built wildcard

The most futuristic-looking entrant is Zoox, owned by Amazon. Its vehicle is a toaster-shaped pod with no steering wheel, no pedals, and bench seats facing each other — built from scratch for autonomy, seating up to four. Zoox launched free driverless rides around the Las Vegas Strip in September 2025 and opened a public service in San Francisco later that year. By spring 2026 it had served around 350,000 riders, with a long waitlist, and was preparing to start charging fares — pending an expanded NHTSA exemption — while testing in Austin and Miami.

Zoox is behind Waymo on scale and behind Tesla on geographic ambition, but Amazon's balance sheet means it isn't going anywhere soon — and a purpose-built vehicle with no controls is, in some ways, the purest expression of the Level 4 idea.

What it means for ordinary drivers and car ownership

So should you care if you live in Cleveland or Kansas City rather than Phoenix? Increasingly, yes — even if a robotaxi never picks you up.

• In the launch cities, a second car is becoming optional. When reliable driverless rides reach freeways and airports at competitive prices, the case for owning a rarely-used commuter car weakens — the same pressure pushing some households toward EVs and away from multi-car ownership.
• The car you buy is not self-driving. Every system you can purchase in 2026 — Tesla FSD (Supervised), Super Cruise, BlueCruise — is Level 2. You are the driver, legally and practically. Treat marketing names with suspicion.
• Insurance and liability are shifting. At Level 4, the operator owns the crash. That's already changing how insurers price these miles, and it's a preview of debates coming to personal vehicles.
• The hardware keeps improving underneath all of it. Better sensors, compute, and — critically — energy density will determine how far and how cheaply these fleets can run; see our explainer on solid-state EV batteries arriving in 2026 for why the battery layer matters to autonomy economics.

FAQ

Can I ride in a fully driverless car in 2026?

Yes, in specific cities. Waymo offers fully driverless paid rides (no human in the car) in Phoenix, San Francisco, Los Angeles, Austin, Atlanta, and several Texas and Florida metros. Zoox runs driverless service in Las Vegas and parts of San Francisco. Outside those areas, you cannot — robotaxis are geofenced to mapped, approved zones.

Is Tesla's "Full Self-Driving" actually self-driving?

No. The consumer product is officially "Full Self-Driving (Supervised)" and is an SAE Level 2 system: you must watch the road and keep your hands ready, and you are legally the driver. Tesla's separate Robotaxi pilot in Austin is where it is testing genuinely unsupervised operation, and even that still uses human safety monitors in many cars.

Are robotaxis safer than human drivers?

The strongest evidence comes from Waymo, which reports 82% fewer injury crashes and 92% fewer serious-injury crashes than human drivers over 170+ million miles, with peer-reviewed and insurer studies backing the trend. Newer or smaller programs like Tesla's robotaxi don't yet have comparable mileage to draw firm conclusions, and federal investigations into FSD are ongoing.

Why did GM's Cruise shut down?

After a 2023 San Francisco crash, a permit suspension, and more than $10 billion spent, GM concluded in December 2024 that the robotaxi business required too much time and capital in an increasingly crowded market. It redirected the technology to personal-vehicle driver assistance instead of running a taxi fleet.

Will robotaxis replace owning a car?

Not broadly, and not soon. In dense launch cities with freeway and airport coverage, they can replace a second or occasional-use car. But for most Americans outside those zones, robotaxis remain unavailable, and personal car ownership — increasingly electric — is still the default.

The Bottom Line

In 2026, "self-driving cars" is really three different stories. Waymo is the clear leader and a genuine business: if you live in one of its 10 metros, a driverless ride is now a normal, often safer, way to get around. Zoox is a credible, Amazon-funded challenger with a striking purpose-built vehicle and real momentum in Las Vegas and San Francisco. Tesla is the wild card — enormous ambition, a tiny fleet, an empty-promise track record on timelines, and a legal status (Level 2) that its branding obscures.

Our advice: if you can ride a Waymo or Zoox, try it — the technology has earned a look. But when you buy a car, ignore the "self-driving" labels entirely. Nothing you can purchase in 2026 drives itself, and treating Level 2 like Level 4 is how people get hurt. The robotaxi future is real, but for now it lives on a map you can count on one hand.