Tesla’s humanoid robot is finally a product coming off a real assembly line. The company said in its third-quarter shareholder letter that production-quality Optimus units are now being built at its Fremont, California factory, in manufacturing space that used to make cars. That milestone arrived alongside the bill for getting there: Tesla’s quarterly profit fell 43 percent to $787 million.
Musk described the milestone as “early production” and warned investors the ramp would be slow and somewhat bumpy, a characterization that understates how much of a leap Tesla is attempting. He nonetheless told the earnings call that Tesla could reach 20,000 Optimus units per week by the end of 2027, with deliveries to outside customers starting around then.
The Q3 numbers behind the bet
Tesla’s third quarter was flat on the top line and ugly underneath. Revenue was $28.04 billion, essentially unchanged from $28.09 billion a year earlier. Net income dropped 43 percent from $1.37 billion. Operating profit fell 56 percent to $718 million, research spending rose 69 percent to $2.76 billion, and capital expenditures more than doubled to $4.73 billion.
The EV business, which funds everything else, stagnated. Vehicle revenue held at $20 billion year over year. Tesla delivered 486,532 vehicles, about 6,000 more than the previous quarter but fewer than the same quarter last year. Energy and storage revenue fell 16 percent to $2.8 billion on tariff-related customer refunds. The one bright spot was services, including Full Self-Driving subscriptions, which grew 33 percent to $4.6 billion and posted a record $734 million gross profit, the only segment keeping the quarter in the black.
Optimus, AI, and autonomous driving ate the rest. A decline in regulatory credit sales added further pressure. Tesla is spending like a robotics company while still being valued like a car company, and the quarter shows the cost of that transition arriving before the revenue.
What “production-quality” actually means here
The shareholder letter’s claim is precise: “The bots coming off this line are the first production-quality bots that we have manufactured.” That means the manufacturing process is proven for a commercial product, not that the product is finished. Musk listed the remaining problems plainly: the robot’s hand is the hardest hardware challenge, durability past about a month of use is unsolved, the design still needs tweaks for safety and reliability, and there is “still quite a lot of work to do” on the AI before Optimus can handle open-ended tasks.
There is also the track record. Tesla missed its earlier Optimus target of about 10,000 robots in 2025. And the economics are brutal at current volumes: Reuters reported the materials for an Optimus Gen 2 could cost $50,000 to $60,000 at low production volumes, against Musk’s long-term price target of around $20,000. McKinsey estimates actuators alone account for 40 to 60 percent of humanoid hardware costs.
A market already owned by China
Optimus is not entering a vacuum. Counterpoint Research data shows Chinese manufacturers including Agibot and Unitree took 86 percent of the 22,000-plus humanoid units shipped in the first half of 2026, though more than 60 percent went to entertainment, performances, research, and data production rather than factory work. Chinese firms have the component-cost advantage; their robots still struggle with everyday tasks. Tesla’s argument is that scale and its AI stack will win the factory floor first.
The Cybercab, Tesla’s two-seat robotaxi with no steering wheel or pedals, is the other half of the bet. An executive said on the call the company is trying to get Cybercabs to customers by the end of 2026, while a federal safety audit remains open and Tesla’s sworn answers to regulators are due October 30.
The arithmetic investors have to do
Musk’s 20,000-units-per-week figure is worth reading carefully. At that rate Tesla would make more than a million robots a year, against a global humanoid market measured in the low tens of thousands today. Getting there requires at least a year of design iteration before the ramp even starts, by Musk’s own admission, and deliveries to outside customers land around the end of 2027.
Meanwhile, Tesla is pouring $4.73 billion a quarter into factories and $2.76 billion into R and D while its core car business declines. Services revenue is growing fast but cannot fund a robotics moonshot alone. The Optimus line at Fremont is a genuine manufacturing milestone, and it is also a marker: Tesla has now spent 43 percent of last year’s quarterly profit proving it can build the robot. The next number that matters is not units per week. It is the cost per unit, and whether anyone will pay it. Nvidia’s new open-weight agent models are coming for the same enterprise AI workloads from the software side, while Microsoft’s Maia 300 shows how much silicon is being thrown at the inference problem. The race to make humanoids useful is on; the race to make them affordable has barely started. Microsoft’s Nvidia-powered Surface Laptop Ultra, pushing 89 tokens a second on-device, shows the parallel bet that heavy AI workloads will live closer to the hardware.
