Tesla begins Optimus mass production at Fremont; Musk sets 20,000-a-week, late-2027 external supply, and a video-demo window
From the Robots digest for 2026-10-22
What happened: Tesla has begun initial mass production of its Optimus humanoid robot at its Fremont, California factory, with coverage specifying that the robot line is replacing the Model S and Model X production lines at the plant. On the ramp, Musk says reaching 20,000 robots a week will take at least about another year, and has set a late-2027 target for external supply — robots going to customers outside Tesla. Separately, Tesla says it will be "maybe a month or two" before it can roll out a video demonstration of the Optimus robot, a delay investors watching the Q3 earnings cycle are reading closely (MarketWatch frames Musk as someone who "could move the entire robotics sector" with what comes out this earnings run). On training, Musk compared teaching Optimus to raising a toddler: the robot has to learn by watching people do tasks and then practicing them itself, and Tesla says it needs 10,000–20,000 Optimus robots practising in an "Optimus Academy" because the program still lacks the training data for general tasks (The Next Web, ~8:51 AM ET October 22; echoed in MSN syndication at ~8:47 AM ET and in RobotsBeat's afternoon write-up at ~4:26 PM ET). The earnings-call detail layer lands mid-day (Electrek, ~11:56 AM ET October 22): Musk put external Optimus deliveries "around the end of 2027" and named the hands, the robot's durability, and its AI as the problems still needing work before then; Electrek's own framing is that those are the same three problems every humanoid maker is fighting, and that XPeng and Hyundai's Boston Dynamics "look at least as far along" as Tesla. The same-day skeptic close-out (The Verge, ~12:56 PM ET) argues Tesla is manufacturing humanoid robots while still working out how to make them useful, reliable, and durable enough for real work — "in production" is not the same as ready. The supply-chain layer gets its first per-unit chip framing (Yahoo Finance, ~12:56 PM ET): investor coverage positions Texas Instruments as a chipmaker that "can get paid on every one" of the Optimus robots off the Fremont line — an investor thesis rather than a named contract; the piece itself notes the chipmaker never mentions Tesla by name. Two late-evening additions round out the picture: Decrypt's figures put the quarter's profit at $787 million, down from $1.37 billion a year earlier (−43%), with spending on robots and self-driving cars cited as climbing (~4:50 PM ET October 22); and CFO Vaibhav Taneja says construction has started on a dedicated Optimus factory in Austin that will make far more robots than Fremont (Teslanorth, ~4:49 PM ET October 22) — the first CFO-confirmed, named larger site beside the Fremont line.
What's genuinely new: Development of the Tesla Optimus thread — the October 21 record already had production-quality units coming off Fremont's high-volume lines and an announced 2027 sales start; the overnight window added the first named production-line casualty (Model S/X), the first quantified ramp horizon (20,000/week in roughly a year), the video-demo timeline, and an external-supply date of late 2027. October 22's morning added the first named training program and a fleet-size requirement for it. The late window then named, from Tesla's own earnings call, the first explicit engineering-blocker list (hands, durability, AI) behind the late-2027 date, the first fuller skeptical read on the production claim, and the first named-chipmaker per-unit supplier thesis. The evening closes two open gaps: the financial drag first filed as a percentage (profit −43%) now has its dollar figure ($787M vs $1.37B), and the factory question has its first named executive statement — a dedicated, larger Austin facility under construction, which gives shape to the October 1 Giga-Texas skeleton record and resolves the October 21 Fremont-vs-Texas contradictory headlines toward "both sites, Austin the bigger one." The training-method claim from October 7–09 (learning chores from watching human video) now has an operational shape, and the 10,000–20,000 figure is itself a unit-count ambition landing while the external record on this thread shows several hundred robots a week with the hands breaking. The standing distinction holds: announced production, not shipped product — no external customer is named, no video has been released, no Academy footage has been shown, the Texas Instruments relationship is analysis rather than a confirmed contract, and Austin is confirmed construction, not an operating plant; all fleet and rate figures remain company claims.
Watch: Whether the "month or two" video demo lands by year-end and shows the Gen 3 design publicly for the first time — and whether it shows the hands, durability, and AI progress the earnings call flagged as blocking; whether an Optimus Academy and its 10,000–20,000-robot fleet materialize on camera, with practice that is autonomous rather than teleoperated; whether the 20,000-a-week pace survives contact with 2027 the way earlier Optimus unit promises have not; who becomes the first external customer behind the end-of-2027 supply date; and whether the Austin factory gets its own ramp date or unit-capacity figure to put beside Fremont's.
Links: BigGo Finance · Drive Tesla Canada · MarketWatch (video demo timing) · MarketWatch (sector read) · The Next Web · MSN (toddler framing) · RobotsBeat (Optimus Academy) · Electrek (earnings-call blockers) · The Verge (readiness skepticism) · Yahoo Finance (Texas Instruments thesis) · Decrypt (Q3 profit figures) · Teslanorth (CFO on Austin factory)