Record car sales, silence on robots: Tesla postpones Optimus launch again

Tesla delivered 480,126 vehicles in the second quarter, more than in any quarter before. Hardly anyone cared on Wednesday evening. The question analysts and shareholders have been waiting for months to hear answered is: How many Optimus robots has Tesla built? The answer was not forthcoming. Again.

In the presentation of quarterly results, it states that the first production lines for the humanoid robot are currently being installed at the Fremont plant, with manufacturing to start "soon". That is a remarkable formulation. Just in April, CEO Elon Musk had scheduled the production start for late July or August, now there is talk of "later this year". Tesla did not provide a verified unit count, neither in the presentation nor when asked directly. The first robots are also not supposed to reach customers at all, but rather gather training data internally.

Musk himself sounded unusually cautious. "This will be the hardest product we've ever had to scale at Tesla, because everything about this robot is new," he said in the conference call. That sentence captures the problem precisely. Optimus consists of around 10,000 unique components for which there is no established supply chain. Alone the over 40 servo motors require approximately 3.5 kilograms of high-performance magnets made of neodymium, dysprosium and terbium per robot. Over 90 percent of the processing of these materials is controlled by China, and the current exception in Chinese export controls expires in November. Just when production is supposed to ramp up.

The stock market responded coolly. The share fell almost four percent after trading closed, even though revenue of 28.2 billion dollars exceeded expectations. Profit declined 57 percent year-over-year, and investments of 25 billion dollars this year are flowing largely into projects that have not yet generated a single franc in revenue: Optimus, the Cybercab robotaxi, proprietary AI chips. Morgan Stanley now calculates the Optimus contribution to the stock price with a probability of only 50 percent. That is the sober translation of investor confidence into a number.

How large the gap to the competition has become is shown by their business reports. Agility Robotics reports over 65,000 operating hours for its Digit with nine named customers, including GXO Logistics and Toyota. Figure deployed sheet metal parts in the production of over 30,000 BMW X3 at BMW's Spartanburg plant for ten months, documented by BMW itself. And Unitree from Hangzhou delivered around 5,500 humanoid robots in 2025, more than all Western providers combined. If you want to know today what humanoids can really do, you don't look to Fremont.

Yet the restructuring in California is quite real. Tesla stopped production of the Model S and Model X in early May, after 14 years and well over 610,000 vehicles. A line with planned capacity of up to one million robots per year is being built on the now vacant space, with an even larger plant planned to follow in Texas starting in 2027. The ambition is not the problem. The problem is the gap between ambition and the first confirmed unit. Five months remain to meet the official annual target of 50,000 to 100,000 robots.

For the industry, Wednesday evening is nonetheless revealing. The humanoid sector is learning to distinguish between announcement and delivery, and capital markets are increasingly rewarding those manufacturers who present operating hours and customer names instead of capacity targets. For Swiss investors who hold Tesla through funds or directly in their portfolio, this close look at the details is worthwhile: The most expensive part of Tesla's valuation currently rests on a product of which there is not a single publicly confirmed unit.