Tesla still builds cars, but Elon Musk has stopped pretending that is the point. He has turned the company's attention away from selling vehicles and toward a driverless Robotaxi service, a humanoid robot he insists will dwarf everything the company has ever made, and now the floated idea of merging with SpaceX. The automaker that by 2020 held the richest market cap of any carmaker is spending like a firm that would rather be an AI company, and the earnings tell the story.
If you want to understand where Tesla is going, stop looking at delivery numbers and start looking at where the money is being burned. The automotive business is shrinking. Automotive operations brought in $20.56 billion in the third quarter, down 3 percent year over year. The company is trying to climb out of consecutive years of declining deliveries, pressured by Chinese rivals including BYD, Nio, and Xiaomi that sell cheap, high-tech EVs in markets beyond American borders. The cars are now, in management's telling, a delivery mechanism for software.
The quarterly math is unflattering. Second-quarter operating margin fell to 1.4 percent, down from 4.1 percent the prior year, and gross margin slipped to 16.8 percent from 17.2 percent, missing the 19.4 percent analysts expected as average selling prices and regulatory credit revenue fell. Tesla sold lower-cost Model 3 and Y versions after retiring the pricier Model S and X, which helps volume and hurts margins. Meanwhile operating expenses climbed 47 percent in the second quarter to $4.35 billion, and free cash flow turned negative by $1.1 billion after generating $146 million in the same quarter a year ago.
CFO Vaibhav Taneja told investors that operating costs will keep rising through 2026 and the years after, citing commodity price increases and interest rate changes. He also told analysts that capital expenditures would top $25 billion in 2026, and second-quarter capex already soared 142% to $5.79 billion. That spending is going into AI compute, battery material, and semiconductor manufacturing, with Musk confirming Tesla already ordered equipment for a 'development fab' for AI chips. The company frames all of this as maintaining liquidity to fund its product roadmap and long-term capacity expansion. You can also read it as a car company paying to become something else.
The Robotaxi service runs on vehicles controlled by Tesla's Full Self-Driving software, and it is real, if modest. Limited service opened in Austin, Texas, in late June 2025 with human safety monitors in modified Model Y vehicles. From there it spread: full commercial service in Dallas launched in April 2026, Miami in July 2026, and Orlando and Tampa Bay the same month. By September 2026, the service was running without an in-car safety supervisor across Austin, Dallas, Houston, Miami, Orlando, and Tampa, while the San Francisco Bay Area still kept one. The company says over 1 million unsupervised miles have been driven.
The fleet is mostly Model Ys, but Tesla has started building the Cybercab, a two-seat vehicle built without a steering wheel or pedals. Purpose-built Cybercabs joined the Austin fleet on September 3, 2026, though Texas records listed only about 45 Cybercabs out of a Texas robotaxi fleet of roughly 420 vehicles. Taneja told shareholders the Cybercab will stay largely a fleet Tesla owns and runs itself, which quietly contradicts Musk's older vision of owners adding their personal cars to a shared network to generate income while they do other things.
The promises here have a long paper trail. At 2019's 'Autonomy Day,' Musk predicted a million robotaxis would be operating by 2020 and said the cars would reach feature-complete full self-driving before the close of 2019. That did not happen. The FSD product is still called Full Self-Driving (Supervised) and, everywhere it runs, needs a human prepared to take the wheel or the brake at any moment. Tesla also lags far behind Waymo and Baidu's Apollo Go, and early Austin riders documented cars straying into oncoming lanes, braking for no reason, and breaking traffic laws, which drew an NHTSA investigation.
What has worked is turning FSD into a subscription. Tesla now counts 1.48 million FSD subscribers, with active subscriptions up 56 percent in the second quarter, and Taneja claimed 75 percent of new customers pay for FSD even though the first month is free. He went so far as to say customers buy Teslas for FSD, not for the Model Y or Model 3 badge. Musk, never one for modesty, said that unless governments stop it, Tesla will handle the majority of all the passenger miles driven worldwide.
For roughly the last two years, Musk's humanoid robot claims have dominated Tesla's earnings calls. He announced Optimus in 2021 and has called it the greatest product the planet has ever produced, repeating on the latest call that it will be the most valuable product ever made, of any kind. He has promised a robot that could handle babysitting, factory work, or even world-class surgery, and said robots will someday outnumber humans.
The catch is manufacturing. Musk admitted Optimus is the hardest product to scale that Tesla has ever made, because everything on the robot is new, and that there is no existing supply chain. Tesla is remaking older Fremont lines to build it and says it is installing first-generation lines and will start production soon. The early robots will serve for training data and development, not for sale to customers.
How Tesla teaches the thing has changed. Until late June, the program used teleoperation and motion-capture suits, with workers wearing suits and VR headsets for extended periods and spending months on a single simple task. Then Tesla told employees it would lean more heavily on a camera-only approach, recording workers on five in-house cameras worn on a helmet and heavy backpack. Tesla says this lets it scale data collection faster. Program director Milan Kovac stepped down shortly before the shift, and Ashok Elluswamy took over.
Robotics researcher Robert Griffin cautioned that teleoperation lets a robot learn through hands-on contact with its surroundings, and that with video alone, "If you're just using video data, you don't have that directly." Sheffield's Jonathan Aitken said Tesla will likely need actions that transfer across many tasks: "Working at this scale, they must have a generalized set of actions or else it would take forever."
The newest wrinkle is a possible merger with SpaceX, Musk's other trillion-dollar company. Musk told the SpaceX earnings call he would welcome a merger and, asked whether it was a good idea, said "Yes." He said both boards formed special committees of independent directors and that he would recuse himself from both committees. Tesla, notably, said the merger was off the table for its own earnings call.
Journalist Kara Swisher read the move bluntly, telling NPR's Steve Inskeep that the idea would have been avoided had Tesla built its long-discussed $25,000 economy EV instead of the Cybertruck. "He's got to do something with Tesla as it descends downward," she said. "If Tesla was doing OK by itself, they wouldn't be doing this. And you can see that from the recent results."
Judge Tesla by cash, not slogans. Capex is set to top $25 billion and keep climbing, so free cash flow and margins tell you whether the AI bets are funded by a healthy car business or draining it. Watch whether unsupervised Robotaxi expands to the announced cities like Phoenix, Las Vegas, and Jacksonville and whether the NHTSA investigation changes anything. Watch for Cybercab volume production and whether Optimus first-generation lines ship. And watch the merger committees. Taneja said the aim is to dominate every form of transportation worldwide. The honest question is whether the company can afford the trip.
[Images: Tesla, SpaceX]
Become a TTAC insider. Get the latest news, features, TTAC takes, and everything else that gets to the truth about cars first by subscribing to our newsletter.