Hello from Editor Z. On October 21, 2026, during Tesla's Q3 2026 earnings call, AI chief Ashok Elluswamy gave a date for unsupervised FSD: the second quarter of 2027. That part is a familiar ritual by now. The part worth your attention is the sentence that followed, because it changed what the date refers to. Step one, Elluswamy said, is not a software update that lets you climb into the back of your Model Y and sleep while it drives you where you asked. Step one is that owners add their vehicles to Tesla's managed Robotaxi network — a service Tesla operates, inside zones Tesla draws, at speeds Tesla permits. Buyers who paid for a car that would eventually drive itself anywhere are now being offered the option to lend that car to a fleet. And HW3 owners are not even being offered that: their computers, Tesla has already conceded, cannot run unsupervised FSD at all, with no clear upgrade plan.
What Was Actually Said on October 21
The Date Survived. The Product Didn't.
Treat the two halves of the announcement separately, because they move in opposite directions. The timeline half is a slip: in July 2025, Elon Musk said unsupervised FSD would be available "for unsupervised personal use by the end of this year in certain geographies." That end-of-2025 window is now roughly eighteen months in the past, and the replacement target is Q2 2027. The product half is the bigger revision. Electrek's summary of the call puts the shift bluntly: the goalpost moved from "your car drives you anywhere while you sleep" to "your car can work as a Robotaxi inside a small geofence that Tesla controls." No highway road trips. No unsupervised drive home from the airport across town.
Why "Enrollment" Is a Different Thing Than a Software Unlock
A software unlock is a product feature: it ships, it's yours, and what you do with it is your business. Enrollment in a managed fleet is a commercial relationship, and the call coverage does not spell out its terms. Nothing in the reporting on October 21 addresses who carries liability when an enrolled customer car is driving itself, how revenue from those rides is split, what happens to insurance, or whether an owner can pull a car out of the fleet on a given morning and still use the capability privately. Those are not footnotes — they are most of what distinguishes a feature you bought from a side business you've been invited into. As of now they are unanswered.
What "Inside the Geofence" Means in Practice
Six Metros, and the Zones Are Small
Unsupervised operation is bounded by where Tesla's Robotaxi service already runs: Austin, Dallas and Houston in Texas, plus Orlando, Tampa and Miami in Florida. Six metros sounds broad until you look at the shapes. Tesla's Miami coverage is mostly West Miami with strips reaching toward Doral and Sweetwater. Orlando and Tampa launched in July with similarly tight geofences. Inside those boundaries the cars run at low speeds, stay off highways, and avoid the harder stretches of urban road. That is a reasonable way to launch a driverless service. It is not the thing anyone pictured when they bought a car advertised as eventually self-driving.
The Bay Area Tells You the Constraint Isn't Only Software
California is the instructive exception. Tesla operates in the San Francisco Bay Area, but those rides still carry a safety driver — not because the software is a different build, but because Tesla declines to share data with the California DMV. Bay Area miles consequently aren't fully autonomous at all, and the market is excluded from Tesla's six-metro autonomous tally. If you live there and bought FSD, the Q2 2027 date does not currently describe anything you can use, and the blocker is a regulatory relationship rather than a release note.
The Mileage Disclosure, Read Properly
Austin Is Two-Thirds of the Entire Program
Tesla disclosed nearly 2 million cumulative unsupervised miles by the end of Q3, then published 3.1 million miles driven autonomously between January 22 and October 18, 2026, with a live counter past 3.3 million by the time Not a Tesla App wrote it up. The per-city split is where the story is, and the concentration is extreme (percentages are my arithmetic on Tesla's own figures, which total 3,112,541 miles).
| Market | Autonomous miles, Jan. 22 – Oct. 18, 2026 | Share of total | What it indicates |
|---|---|---|---|
| Austin, TX | 2,118,015 | 68.0% | The only market with real operating history; 300+ purpose-built Cybercabs deployed |
| Dallas, TX | 608,130 | 19.5% | Second site; unsupervised Cybercab rides slated to begin next month |
| Houston, TX | 225,964 | 7.3% | Meaningful volume, far behind the two leaders |
| Orlando, FL | 61,053 | 2.0% | Launched July 2026; Waymo was already operating there |
| Tampa, FL | 53,117 | 1.7% | Launched July 2026; Waymo already operating there too |
| Miami, FL | 46,262 | 1.5% | Geofence is mostly West Miami plus strips toward Doral and Sweetwater |
| San Francisco Bay Area | Not counted | — | Safety driver required; Tesla won't share data with the California DMV |
Two Texas cities are 87.5% of every autonomous mile Tesla has logged this year. The three Florida metros combined contribute 160,432 miles — about 5.2%, or roughly what Austin produces in under a month at this year's average pace. "Six metros of coverage" and "six metros of operating experience" are not the same claim, and only the first one is true.
The Growth Rate Is Steep, Which Cuts Both Ways
Set the two disclosures side by side: nearly 2 million miles by the end of Q3 on September 30, and 3.1 million by October 18. That is roughly 1.1 million miles in about 18 days, or on the order of 60,000 miles a day. The figures come from different disclosures and may not be measured identically, so treat the rate as approximate — but even approximately, it means more than a third of Tesla's lifetime autonomous mileage was driven in the final two and a half weeks of the window. The optimistic reading is that scaling is working. The cautious reading is that the safety record below rests heavily on a very short, very recent sample.
Zero Injuries — and the Fine Print
Tesla's safety report claims no people injured by a Robotaxi, no airbag deployments, and no pedestrians or cyclists struck. The conditions breakdown: 862,000 miles after dark, 113,000 miles in rain, 575,000 miles during rush hour. By my arithmetic that is about 28% of miles at night, 18.5% in rush hour, and just 3.6% in rain. Tesla also estimates human drivers would have caused roughly 9.1 injury collisions over the same mileage, which implies an assumed human injury-collision rate of about one per 342,000 miles. The record is genuinely good. It is also a record earned at low speed, off highways, away from the hardest urban geometry, and with two-thirds of it collected in a single city — exactly the envelope that makes the Q2 2027 promise narrow in the first place.
HW3 Is the Part Nobody Should Skip Past
Here is the admission that should reframe the whole announcement for existing owners. HW3 cars whose owners bought FSD have been told their computers cannot run unsupervised FSD at all — not in a geofence, not in a fleet, not at low speed. There is no clear upgrade plan. For a feature that commenters on the Q3 coverage note was being sold for private cars something like seven years ago, that is a durable gap between what was purchased and what will exist. And note how it interacts with the fleet framing: if the only route to unsupervised capability is enrollment in Tesla's Robotaxi network, then a car that can't run the software can't enroll, which means it has no path to the feature even in its diminished form.
What Each Tesla Owner Actually Gets
Based strictly on what was disclosed on and around October 21, 2026, here is the honest matrix. Find your row.
| Your situation | What Q2 2027 offers you | Limits that apply | Confidence |
|---|---|---|---|
| HW3 car, FSD purchased | Nothing — the computer can't run unsupervised FSD | No clear upgrade plan announced | Stated outright by Tesla |
| Newer car, inside one of the six live metros | The option to add your car to Tesla's managed Robotaxi fleet | Geofenced zone only, low speeds, no highways, harder urban areas avoided | The actual Q2 2027 target |
| Newer car, San Francisco Bay Area | Unclear — rides there still require a safety driver | Blocked by Tesla's refusal to share data with the California DMV | Low; no date given |
| Newer car, Las Vegas / Phoenix / Jacksonville / San Antonio | Listed as "coming soon" for Robotaxi service | Tesla gave no launch dates for any of the four | Low; dateless |
| Newer car, anywhere else in the US | Supervised FSD, as today | No announced timeline for unsupervised personal use outside service zones | None offered |
| Anyone wanting unsupervised highway or cross-town driving | Not part of this announcement | Highways are explicitly outside the operating envelope | None offered |
Why the Business Model Moved Instead of the Product
The Financial Backdrop
The reframing landed in the same week Tesla reported a 43% drop in Q3 profits. That juxtaposition doesn't prove intent, but it does explain the appeal of the structure: a managed fleet of customer-owned vehicles is capital Tesla doesn't have to spend. Tesla has 300-plus Cybercabs in Austin, which it had to build and own. Enrolled customer cars arrive already paid for — by the customer. Shifting the delivery of unsupervised FSD into the fleet is simultaneously a technical retreat to an easier operating envelope and a cheaper way to add supply.
The Scale Question Nobody Has Settled
The counterweight to Tesla's mileage chart is the vehicle count. Third-party tracking in spring 2026 put fully driverless Teslas in the dozens — a figure that predates the 300-plus Cybercab deployment and so understates today, but it marks how recently the real-world fleet was tiny. Waymo, for its part, has over 3 million miles and was already serving both Orlando and Tampa before Tesla arrived. The honest summary is that Tesla's mileage is now in the same order of magnitude as Waymo's while its geographic envelope is tighter and its unsupervised capability still routes through a fleet an owner joins rather than a feature an owner holds.
The checkable test arrives before Q2 2027. Dallas is supposed to begin unsupervised Cybercab rides next month, which will show whether Austin's record travels to a second city. Watch for three things specifically: whether any of Las Vegas, Phoenix, Jacksonville or San Antonio gets an actual date; whether Tesla publishes fleet-enrollment terms covering liability, revenue share and insurance; and whether HW3 owners are offered anything at all. Until those land, the fair description of what Tesla committed to on October 21 is a date for a service, not a date for the product people bought.
-EditorZ
Post a Comment