Tesla’s Optimus Line Starts at Fremont, but Output, Price and Hands Remain Unproven

Tesla has crossed a threshold that is simple to announce and hard to prove. In its third-quarter shareholder letter on October 21, the company said Optimus production had started at its Fremont, California factory on high-volume lines, calling the units “the first production-quality bots that we have manufactured.” Yet the same document’s capacity table lists California/Optimus as “Early Ramp” and Texas/Optimus as “Construction,” with the capacity column blank.

• Production, not output: no unit count, weekly rate, price or delivery schedule was disclosed.
• Executive caution: Elon Musk calls it “early production,” with a shallow, bumpy ramp ahead.
• Unfinished hardware: hands and open-ended AI remain works in progress, and no verified demo exists.

The gap between a running line and a sellable product

Fremont’s Optimus line occupies space that built the Model S and Model X until this spring; Tesla cleared the old tooling in 46 days for a line sized at one million units a year. The letter is precise about what is validated: “we have proven that the current manufacturing process for Optimus is suitable for a commercial product.” That is process readiness, not product readiness.

Tesla also has not documented which generation is rolling off the line: “Gen 3” appears nowhere in its 2026 quarterly reports, and the unveiling promised for the first quarter slipped repeatedly.

The financial backdrop sharpens the stakes. Tesla reported GAAP net income of $787 million, down 42.7% year over year, on flat revenue of $28.046 billion. Operating expense rose $1.346 billion, with R&D contributing $1.131 billion; CFO Vaibhav Taneja tied part of the increase to “preproduction ramp costs for new products like the Optimus, Megapack 3 and Semi truck,” warning the trend would continue until commercialization. Tesla does not break out robot-program costs. Musk’s September 2025 post that “~80% of Tesla’s value will be Optimus” frames the bet.

The 20,000-a-week target is real, the deadline is not

Musk says reaching an annualized rate of 20,000 Optimus units a week will take “at least a year, maybe a little longer.” Several outlets converted that into “by the end of 2027”; that date belongs to their arithmetic, not his words. He also claimed Fremont already houses “the vast majority of a 1 million unit per line” production line, while Taneja said the Austin plant under construction will “do a lot more robots as compared to the Fremont one.”

“It’s an S-curve, but the initial part of the S-curve is — has a shallow slope and is somewhat bumpy,” Musk said. “There’s like 10,000 things that have to go right for production of the robot, and it will move as fast as the slowest least likely of the 10,000.”

Tesla plans an “Optimus Academy” where 10,000 to 20,000 robots rehearse tasks and feed results into the next model. Suppliers Tuopu, Joyson and Sanhua were certified as mass-production partners in September, and a larger plant is rising at Giga Texas, where Musk has cited a 10-million-a-year long-term target — a sharp contrast with Tesla’s own “Construction” label.

Hands, AI and the demo that has not arrived

The candid admissions concerned hardware. A Tesla executive said the “hand is one of the most challenging hardware areas,” with “some bespoke gears that are carrying a lot of burden for consumption and for reliability that we are developing still.” Musk was blunter: “we’re building robots that do not have perfected hands,” adding that much AI work remains before Optimus is “truly useful for open-ended tasks.” The AI5 chip, promising an order of magnitude more compute than AI4, should reach the robot around April 2027.

No video, independent verification or public demonstration yet shows a Fremont-built unit doing useful work. Musk promised a demo in “maybe a month or 2,” one “that you can believe what you see,” while dismissing rival humanoid showcases as “preprogrammed deterministic tasks” that “wouldn’t work basically” in the real world. External deliveries are targeted for around the end of next year, starting with dangerous, repetitive industrial jobs. No price has been announced; Taneja said costs are “trending closer” to Cybercab levels without giving a figure.

The milestone is real: Tesla pledged in January to begin production before the end of 2026 and beat that deadline by two months. But a company that has installed a line, validated a process and still admits its robots lack finished hands is describing something narrower than “production” implies. Musk’s own phrase — “Vegas betting on the come” — and his claim that “there is no one that is even close to Tesla” will be judged over the next year in verified output, not shareholder letters.