Tesla's Robotaxi Ramp Has Stalled, and the Cybercab Is Getting a Steering Wheel

No new vehicles have joined the network in over a month, unsupervised driving got rolled back from 94% to 78%, and the real 2026 story is a two-seater Tesla sells you outright rather than a fleet it runs itself.

The Robotaxi dream people bought in 2025 was a fast-scaling driverless network minting recurring revenue across dozens of cities. That's not what's happening. The network has added zero new cars in over a month, the share of unsupervised rides got walked back after briefly touching the mid-nineties, and the honest read is that the software isn't ready for prime time yet. But the more interesting conclusion isn't bearish at all. It flips the whole thesis: the biggest near-term money in autonomy is Tesla selling you a cheap, comfortable, self-driving car you own, not renting you a seat in one it owns.

Key Takeaways

• Zero new vehicles have joined the Robotaxi network in just over a month, with the one-year anniversary of the network landing on June 22nd and no birthday fleet drop to show for it.
• Unsupervised operation on the seven-day rolling average spiked to roughly 93-94% in Austin, then got pulled back to around 78%. When a fleet pauses and rolls back safety autonomy at the same time, the software is telling on itself.
• Safety, though, is not the problem. There have been zero at-fault Robotaxi accidents since February, and the three recorded collisions were the Robotaxi getting rear-ended.
• Staged and waiting: spotters have counted around 106 vehicles in an outbound lot, plus pods of roughly 50 Cybercabs each staged in Houston and Dallas, ready to deploy the moment the software clears.
• Tesla will almost certainly sell the Cybercab with a steering wheel and pedals, betting most people just buy and own a driverless car rather than pledge it to a network.
• At 20-25% gross margins on Cybercab plus recurring FSD revenue on top, selling three to five million two-seaters a year is the cash-flow story, and it's the one that could re-rate the stock.
• Q2 deliveries are tracking toward the mid-400,000s, a strong quarter, and the near-term catalyst is SpaceX, not any Robotaxi expansion.
• A "rent a supervised FSD Tesla" model is the sleeper idea here, because a couple hours behind the wheel is what converts a skeptic into a buyer.

---

A Network That Stopped Growing

Start with the tell. A robotaxi network that's working does one thing above all: it grows. More cars, more coverage, more unsupervised miles every week. Tesla's has gone quiet for over a month. No new vehicles added. The cars exist and they're staged, roughly 106 sitting in an outbound lot and pods of about 50 Cybercabs each parked in Houston and Dallas, engines effectively idling. Hardware isn't the bottleneck. The bottleneck is the code.

The clearest evidence is the unsupervised share in Austin. On a seven-day rolling average it climbed to something like 93-94%, meaning most rides ran with nobody minding the wheel. Then it fell back to around 78%. You don't roll that number backward if you're confident. A paused fleet plus a retreat on unsupervised operation, happening together, means one thing by default: not ready yet. I don't think that's a scandal. I think it's Tesla being appropriately careful with a system that carries real liability. But it does end the fantasy that this scales fast.

Safe Enough, Just Not Smooth Enough

Here's the nuance most people miss. The problem isn't safety. Since February there have been zero at-fault Robotaxi accidents. The three collisions on record were the Robotaxi getting rear-ended by another vehicle, which is the other driver's fault, not the machine's. On the metric that actually matters for regulators and lawsuits, the system is already performing.

So what's holding it back? Refinement. There's a gap between a car that's safe enough to drive empty and a car that's polished enough to satisfy a paying passenger. An empty car can stab the brakes or make a minor navigation error and nobody cares. Put a human in the back seat and that same twitch becomes the difference between a delightful ride and a one-star rating. The engineering that's left isn't "don't crash." It's "don't annoy." That's a subtler, slower problem, and it explains the pullback better than any safety scare.

The Robo-Chauffeur Nobody Wanted to Admit Was Coming

There's an old idea floating around Tesla circles that used to sound silly and now looks prescient. Call it the robo-chauffeur. The car is smart enough to drive itself to you empty, but the ride experience still benefits from a human in the loop. For years that felt like an awkward halfway house. Today it looks a lot like where the technology actually sits.

You can already see the shape of it in the short-term rental setup. Fill out a rental agreement, request a Tesla, and one shows up on its own. You climb into the driver's seat and run FSD supervised for your trip, rather than riding as a passenger in a fully unsupervised robotaxi. It threads the exact needle the technology is stuck on: autonomy good enough to reposition the car, human oversight to smooth the parts that would otherwise frustrate a rider. The intermediate step people dismissed turned out to be the product.

Why the Cybercab Gets a Steering Wheel

This is the part I've come around on hard. The original Cybercab pitch was a purpose-built robotaxi with no wheel and no pedals, a car you'd never drive yourself. I now think Tesla sells it with both. And the logic is almost too clean to ignore.

If FSD gets commoditized and the hardware gets cheap, a driverless car stops being a scarce asset you'd want to pool into a network and becomes just an affordable car people buy and keep. Why pledge your vehicle to a fleet for a thin recurring cut when owning it outright is cheap and you get all the utility yourself? The counterintuitive result: cheap autonomy could mean more people own cars, not fewer. The "nobody will own a car" future gets weaker as the tech gets better, because ownership gets so cheap the whole rental case erodes.

And Tesla has run this play before. It got hundreds of thousands of owners to pay for the hardware and beta-test FSD on public roads, effectively funding its own training data. Selling Cybercabs does the same thing at the fleet level. Let the customer eat the upfront hardware cost, book the revenue and margin today, and keep the option to fold those cars into a network later if demand shows up. Take the sure money now instead of waiting a year-plus to earn it back one robotaxi fare at a time.

The Kansas City Fleet Guy Is Going to Be Disappointed

There's a fantasy going around that autonomy is a Wild West land grab, where anyone with spare cash buys ten robotaxis and launches a local fleet. I don't think that business exists the way people imagine. You'd be competing directly with Tesla's own network in your own city.

The likelier structure is that Tesla seeds a baseline number of robotaxis in metro areas it thinks warrant one, then supplements with owner cars only when and where it needs the extra supply. Demand-based, Tesla-controlled, not an open marketplace you can just plug into for passive income. You keep the option to add your car to the network. You don't get to build an empire on it. The value for an owner is a driverless car that's cheap to own, plus a little recurring upside when Tesla actually needs your vehicle. That's a good deal. It's not a gold rush.

The Rental Trick That Sells the Whole Thing

The rent-a-supervised-FSD idea is smarter than it first sounds, because it solves the real barrier to adoption, which is psychological, not technical. People fear ceding control. Even if the car drives better than they do 99.9% of the time, it's the 0.1%, one harmless brake stab, that spooks a normal person who isn't a Tesla enthusiast. And people behave completely differently in FSD alone versus with their spouse or parents in the car.

The fix is time in the seat. The tolerance curve is steep. The first five minutes, a new user is tense. Twenty minutes in, they're curious. Then the car does the genuinely hard stuff, navigating work zones, pulling over for police, handling complex intersections and unprotected left turns, and something clicks. At that point they stop watching and pull out their phone. I've watched people who'd never have considered a Tesla buy one over FSD and then call me stunned they waited so long. If a rental gets a skeptic a couple hours of that experience, it manufactures buyers at scale. That's why the rental idea might matter more than the robotaxi network itself right now.

What Actually Moves Tesla in 2026

Let me be blunt about the near term. Robotaxi is not a 2026 catalyst. The scaling-fast dream should be set aside. A few cities were floated for the first half of 2026, and I don't expect them to land on schedule. When Cybercab does join the fleet, it'll come slowly and deliberately, one car, then five, then ten, pause, software update, twenty, pause again. That's the responsible cadence, and it's the opposite of a hockey stick. The only place it scales fast soon is inside the Gigafactories as internal testing, letting workers ride Cybercabs at Giga Texas, which is great for data and irrelevant to revenue.

The real catalyst is boring and bullish: Q2 deliveries tracking toward the mid-400,000s, a genuinely strong quarter, alongside SpaceX dominating the headlines after its IPO. Vegas is the one Robotaxi expansion I'd actually bet on near-term, and only on a small scale, because the city's layout is easy and its leadership treats every new ride as another attraction. Then there's the bigger picture. Autonomy will hit regulatory walls, job-loss politics, and officials who slow it down, and none of that lets you max out factory capacity on a network model. Meanwhile there's going to be enormous demand for a cheap, comfortable, fully self-driving two-seater. Selling it at 20-25% margins with recurring FSD revenue stacked on top, at three to five million units a year, is the kind of cash flow that makes the market re-rate the whole company. Tesla would be foolish not to sell the car. So it will.