Tesla Leans on FSD to Sell Cars as Robotaxis Scale

More than 75% of Tesla’s new North American deliveries in the third quarter included Full Self-Driving (Supervised). Chief Executive Elon Musk told investors on October 21 that the software is now the biggest driver of car demand.

He also said the Robotaxi fleet would soon become a real revenue line, and that Optimus humanoid robots could ship to companies outside Tesla by the end of 2027. GAAP net income still fell 43% to $787 million on $28.05 billion of sales that barely moved from a year earlier.

FSD Is Why Buyers Show Up at Tesla Stores

Tesla’s own letter put the claim in writing. FSD (Supervised) still needs a person ready to take over, and the company repeats that it does not make the car autonomous. Buyers in North America, Tesla’s most mature market, are paying for it anyway at the highest attach rate Tesla has disclosed.

The share of new North American cars that included FSD rose by nearly 20 percentage points from the second quarter, to more than 75%. Chief Financial Officer Vaibhav Taneja said those sign-ups were paid subscriptions at delivery, not the free one-month trial Tesla already bundles. Musk’s line on the call was blunt. The biggest driver of demand, he said, is full self-driving.

People do not attach a paid driver-assist pack to three out of four new cars unless that pack is why they walked into the store. The owned Tesla is supposed to be the old business. In the quarter Tesla just reported, it was the business that closed.

FSD SUBSCRIPTIONS BY QUARTER

Tesla’s third-quarter update with the SEC counts those 1.74 million the way Musk’s 2025 pay package defines them, mixing monthly plans with older upfront purchases and leaving out free trials. That is a 67% jump from 1.04 million a year earlier, and a record for net new additions, up 0.26 million from 1.48 million in the second quarter. Taneja said more than 1.7 million customers now pay, and that for the first time more than 50% of them are on monthly subscriptions.

Croatia, Slovenia, the Czech Republic and Slovakia approved FSD after the second quarter, bringing the system to nine EU countries. Tesla said more than 100,000 customers there have logged more than 160 million kilometers. Musk said he would be surprised if EU-wide approval slipped past the first quarter of 2027, while stressing that Tesla does not control the clock. Tesla posted the quarter’s highlights as the numbers went out.

Q3 Shareholder Update → https://t.co/sXBSeLiJIj

Highlights
– FSD Supervised attach rate grew to over 75% of new vehicle deliveries in NA

– Unveiled the next-gen Roadster

– Record deliveries in Canada, Australia, Japan & Taiwan

– Tesla Semi now in volume production. Top… pic.twitter.com/yQQRq9LwSd

— Tesla (@Tesla) October 21, 2026

A 2% Delivery Dip and a Four-Year Backlog

Tesla delivered 486,532 vehicles in the quarter, 2% fewer than a year earlier, and still 22,141 more than the 464,391 it built. Inventory fell to 11 days of supply from 15 in the second quarter. Taneja said Tesla had now posted two straight quarters above 480,000 deliveries and had exited September with its largest passenger-car backlog in four years. Production, he said, was the constraint, and battery supply was the limiter on how fast factories could run.

Almost all of that volume was Model 3 and Model Y, at 478,237 deliveries. Other models, the bucket that holds Cybertruck, Semi, Model S and Model X, fell 48% to 8,295. Tesla stopped building S and X in Fremont earlier this year to free floor space for Optimus. Record deliveries in Canada, Australia, Japan and Taiwan, plus record output at Gigafactory Berlin, did not hide that mix shift. Tesla also produced and delivered its 10 millionth vehicle, a cumulative count that, under the same pay-package rule, includes unsupervised robotaxis put into commercial service. The figure now stands at 10.2 million.

TESLA’S THIRD QUARTER, YEAR OVER YEAR

Automotive sales were down 3%. Energy generation and storage fell 16% to $2.88 billion even as Tesla deployed 13.7 GWh of storage, up from 12.5 GWh, and said orders and backlog hit records. Services and other rose 33% to $4.61 billion, with a record $734 million of gross profit. Operating expenses climbed 39% to $4.78 billion, which is why operating income dropped 56% to $718 million and the operating margin compressed to 2.6% from 5.8%. Gross profit still rose to $5.49 billion, and GAAP gross margin reached 19.6%. Non-GAAP net income was $1.70 billion, or $0.48 a share, against $0.50 a year earlier. GAAP earnings were $0.22 a share.

Musk opened the call by saying demand is growing rapidly and exceeding production by a significant margin. That is the FSD story in hardware form. Tesla is selling about 2 million cars a year at this pace, aiming at about 5 million around 2030, and telling investors the car is no longer the point.

Robotaxi Miles Jumped Eightfold in One Quarter

Cybercab joined the Austin Robotaxi fleet on September 3. Tesla said cumulative unsupervised miles, rides with no safety monitor in the vehicle, reached nearly 2 million by September 30, eightfold growth from the end of June, on sustained double-digit weekly gains. Musk said the Austin Cybercab fleet has grown by a factor of 10 since those rides began. Hours there now run 6 a.m. to midnight, and Grok is in the cabin for voice control of mail, calendars, files and tasks.

Those miles are still a rounding error next to the car business. Tesla did not break out Robotaxi revenue. Management’s own timeline puts a noticeable P&L contribution in 2027, with a bigger print in the third or fourth quarter of that year. Ashok Elluswamy, vice president of AI software, said Dallas gets Cybercab in November 2026, the first city outside Austin to receive the two-seat vehicle, and that the same stack can then spread with Model Y robotaxis. Musk said rides will stretch toward 2 a.m. and toward round-the-clock service by the end of 2026. He also said the vast majority of Cybercabs will stay Tesla-owned as transport-as-a-service, while Model 3, Model Y and the Roadster remain the cars people will want to own. Texas already lists Cybercab capacity above 125,000 units a year. Every Cybercab built so far has gone into Tesla’s own fleet.

THE AUTONOMY CALENDAR TESLA JUST SET

1. September 3, 2026: Cybercab starts Robotaxi rides in Austin.
2. October 21, 2026: Tesla reports nearly 2 million unsupervised miles and more than 75% FSD attach in North America.
3. November 2026: Cybercab launches in Dallas, the first city outside Austin.
4. End of 2026: Tesla targets 24/7 Robotaxi hours.
5. Around Q2 2027: Some customer-owned cars join unsupervised operation through the V15 software release.
6. End of 2027: Tesla aims to deliver Optimus to companies outside Tesla.

The safety page Tesla published with earnings is the other half of the pitch. Through October 18, Tesla’s Robotaxi Safety Report lists zero injuries, zero airbag deployments and zero pedestrians or cyclists hit in the unsupervised window it counts from January 22. Austin accounted for 2,118,015 of those miles, Dallas 608,130 and Houston 225,964. Dallas already had Model Y robotaxis; Cybercab is the extra hardware arriving in November. On Tesla’s consumer site, Robotaxi rides in six U.S. cities are listed as live: Austin, Dallas, Houston, Miami, Orlando and Tampa. A live counter on those pages had passed 3.3 million autonomous miles by October 22, a running total that is wider than the nearly 2 million unsupervised miles Tesla booked at quarter end.

Taneja said Tesla still sees a path to 30 cents per mile as the fleet grows. Musk went further, saying that wherever Tesla is allowed to run freely it will end up doing a majority of passenger miles. That is the 2027 argument. The 2026 evidence is a geofenced fleet whose miles are compounding and whose sales line is not yet large enough to show up beside $20.56 billion of cars.

When Tesla Expects to Ship Optimus to Other Companies

Tesla has started Optimus production on high-volume lines in Fremont, the floor that used to build Model S and Model X, and it wants outside companies to take robots around the end of 2027. The bots coming off that line are the first production-quality units Tesla has built, the company said, and the process is now suitable for a commercial product. Musk said most of the equipment for a line designed to make about 1 million robots a year is already in Fremont. Reaching an annualized 20,000 units a week will take at least a year, and maybe longer. A second Optimus site in Texas is still in construction.

We’ve started early production. The growth of Optimus is an S curve, but the initial curve is a shallow slope. There’s still some things we need to tweak about the design. There’s still quite a lot of work to do with the AI in the robot. I would expect that the AI, the electromechanical design, reliability, usage life, expect these issues to be solved by the end of next year. I remain convinced Optimus will be the biggest product on Earth.

Elon Musk, Tesla CEO, Q3 2026 earnings call

He has said the robot cannot wear out after a month, and that Tesla still needs to deliver a machine that is safe, reliable and long-lived. A new Optimus V3 demo, meant to show tasks that were not pre-programmed, is due in one or two months. The AI5 chip is slated for the robot around April 2027. On the call he compared the training problem to raising a toddler: Tesla does not have a vast robot fleet to harvest data from, so it has to build thousands of units, run an Optimus academy, and let the machines practice. That is a factory story and a data story. It is not yet a customer-shipment story.

THE WORK OPTIMUS STILL HAS TO CLEAR

• The AI: Musk said there is still quite a lot of work in the robot’s software before outside deliveries.
• The hardware: Design tweaks remain, and the hand is the piece he keeps calling the hard problem.
• Useful life: The bot has to run for a long time, not wear out after a month.
• The ramp: A 20,000-a-week rate is at least a year away, even with most of the million-unit line already in Fremont.

Early units are for Tesla’s own plants, dirty and dangerous jobs, and for the training flywheel. Outside companies sit at the far end of that sequence. If that date holds, Tesla becomes a robot vendor to other firms at the same moment it is trying to turn unsupervised miles into a taxi P&L. Neither line paid the third-quarter bills.

Merger Questions Were Off the Table

Travis Axelrod, Tesla’s head of investor relations, opened the question session by saying the company would not take questions on a potential merger with SpaceX. Musk had already talked about that on SpaceX’s call on October 20, calling the exercise tap dancing through a minefield. Special committees of independent directors at both companies are discussing a combination at shareholders’ request. Musk, CEO of both, has recused himself as a conflicted party. The committees have not agreed on terms, and he said a deal is not a certainty.

Any Tesla vote, he said, would need support from a majority of all shares, not just shares that are voted, which he called a high hurdle. He still said there is a lot of merit in combining the companies around software, chips, the Texas Terafab project, robotics and connectivity. That is a 2027 governance fight sitting on top of a 2026 car company that is using FSD to keep the order book full. Investors who wanted merger math on Wednesday did not get it.

Tesla Trimmed 2026 Capex to About $20 Billion

The cash line is the one Wall Street did not model. Free cash flow came in at $1.53 billion, after $6.26 billion of operating cash flow and $4.73 billion of capital spending. Analysts had been braced for a multi-billion-dollar burn and for capex closer to $8 billion. Cash and investments rose $1.9 billion to $45.4 billion. Taneja cut full-year capex guidance to about $20 billion from about $25 billion, saying Tesla would fund it mainly from operating cash flow. Musk said the spend gets more meaningful next year and then into the tens of billions in 2028, as Optimus lines, Cybercab, training compute and the Austin chip fab scale. Tesla is aiming for 400 MW of training compute by the end of 2026. Megafactory Texas is running, with planned capacity of 50 GWh, and Tesla shipped the first Megapacks built entirely with its own Nevada LFP cells. Taneja said 2027 Megapack output is nearly sold out, and energy gross margin reached 33.1%.

Volume production of Tesla Semi has started in Nevada, at a listed 25,000-unit annual capacity. The top variant, Tesla said, can go 500 miles at 82,000 pounds and 1.7 kWh per mile, about a third of the energy a comparable diesel uses for the same load. The next-generation Roadster was unveiled on October 15. Supercharging added more than 2,100 stalls, up 14% from a year earlier, to 8,904 stations and 84,465 connectors. Those are the support systems around a car business that is still waiting on cells.

The dated proof points Tesla just put on the calendar are a Cybercab launch in Dallas in November 2026, an Optimus demo in a month or two, and a hope of EU-wide FSD approval by the first quarter of 2027. Until those land, the product Tesla can actually ring up is a human-owned car with Full Self-Driving attached.

Disclaimer: This article is news reporting and analysis of Tesla’s third-quarter results and earnings call, and it is for information only. It is not investment advice, a recommendation to buy or sell Tesla or SpaceX stock, or a forecast of any merger, robot shipment, or robotaxi profit. Readers should consult a licensed financial adviser or other qualified professional who can review their own holdings and risk tolerance before making any investment decision. Figures, fleet counts, mile totals, and product timelines reflect Tesla’s filings, safety pages, and executives’ remarks as of the dates named above and can change with later reports, software updates, or regulatory action.