Is Tesla Optimus Working in Factories? The Real Status in September 2026

No, there are not more than a hundred Optimus robots working shifts inside Tesla's Fremont factory. As of early September 2026 the building holds a partially installed production line where Model S and Model X used to be built, and Tesla's own quarterly filing lists that line under the status “Construction” with an installed annual capacity of “-”.

Search for the robot and you will find fleet counts, payload figures and runtime specs, all stated with total confidence. Almost none of them come from Tesla.

Three things get mixed together every time Optimus is written about: demo footage, executive projections, and shipped reality. Those three are still a long way apart. The distance between them is the story.

What Tesla itself says, quarter by quarter

Tesla publishes a quarterly shareholder update PDF on its investor relations site. It is a primary source, it is legally consequential, and it contains a table listing every product line with its installed annual capacity and its status. Checking that table is the fastest way to test an Optimus rumour, because Tesla has to stand behind what it writes there.

Here is what the last three of those documents say about Optimus, verbatim where it matters:

Three consecutive quarters, same status, no capacity number. The Gen 3 unveiling promised for Q1 2026 in the January document had not happened by the time the July document was published, and the July document does not mention it at all.

One more detail gets lost in headline coverage. In the Q3 2025, Q4 2025 and Q1 2026 updates, Tesla's outlook section read: “First-generation production lines for Optimus are being installed in anticipation of volume production.” In the Q2 2026 update the same sentence reads: “...in anticipation of production in 2026.” The word “volume” quietly disappeared. That is what it looks like when a company narrows its own commitment in the safest legal language available to it: one word, removed without comment.

The Fremont claim, checked against the filings

Tesla files a production and deliveries release with the SEC at the start of every quarter. The Q2 2026 one, dated 2 July 2026, reports 451,758 vehicles produced, 480,126 delivered and 13.5 GWh of energy storage deployed. It breaks vehicles into Model 3/Y and Other Models. There is no Optimus line. There has never been an Optimus line in any Tesla production report.

That absence is not an oversight and it is not a cover-up. Tesla does not report a number because there is no reportable production number. On the Q4 2025 earnings call, held 28 January 2026, Musk was asked directly about robots working in Tesla's factories and said: “We have had Optimus do some basic tasks in the factory. But as we iterate on new versions of Optimus, we deprecate the old versions. It's not in usage in our factories in a material way.” He added: “Well, we are still very much at the early stages of Optimus. It's still in the R&D phase.”

That admission carries weight because of what came a year before it. On the Q4 2024 call in January 2025, Musk said “the normal internal plan calls for roughly 10,000 Optimus robots to be built this year” and, asked whether they would be doing useful things by the end of 2025, answered: “Yes, I'm confident they will do useful things.” Twelve months later the answer was that the robots were not in material use. Tesla had also stated in mid-2024 that it had two Optimus units performing tasks in the factory autonomously.

So the honest description of Fremont in September 2026 goes like this. The Model S and Model X lines were decommissioned and the last of those cars came off the line around May 2026. The building is being retooled. Tesla expects production to start “later this year”. And whatever comes off that line first is earmarked for an internal programme, not for a job on the assembly floor and certainly not for a customer.

The phrase “Optimus Academy” deserves a second look too. Tesla says the initial builds will be used for training data collection and further functionality development. The first machines off the line will not take over a job; they will generate the data used to teach themselves. That is the description of a research programme, not a product.

Where the circulating numbers come from

Four claims dominate the Optimus discourse, and all four fall apart on contact with a source. The errors are not random, either. They show, step by step, how bad robot coverage gets manufactured.

• “Lifts 45 kg.” The widely cited carrying-capacity figure for Optimus is 45 pounds, which is about 20 kg. Someone dropped the unit conversion and more than doubled the robot's strength. This is the single most common Optimus error in circulation.
• “Runs 24 hours uninterrupted.” Tesla has never published a runtime specification. What it did state at the Gen 2 reveal in December 2023 is a 2.3 kWh battery pack in the torso. A 2.3 kWh pack cannot power a walking, manipulating humanoid for 24 hours; published third-party estimates land in the hours, not the day. Any 24-hour figure is invented.
• “10,000 produced by 2027.” The 10,000 was Musk's stated internal plan for calendar year 2025, and it was missed by a wide margin. Tesla has since declined to publish any unit target for 2026.
• “Over 100 robots actively working.” Contradicted by Musk's own January 2026 statement and by three quarters of “Construction” status in Tesla's filings.

There is a related confusion worth naming, because you will run into it constantly. The figures “1 million robots a year” (Fremont) and “10 million robots a year” (Gigafactory Texas) are real Tesla statements. But they describe what the lines are designed for at full theoretical output, not what they will build. Tesla says so explicitly in the footnote under its own capacity table, which warns that installed capacity exceeds current production rate and that limitations may be discovered as rates approach capacity. Design capacity is an engineering intent. Output is a fact. Coverage that treats them as the same thing generates a large share of Optimus nonsense on its own.

Which generation is current, and why you have not seen it

The current publicly demonstrated hardware is Generation 2, revealed in December 2023, plus an interim revision Tesla itself refers to as “version 2.5” in its Q4 2025 update. Generation 3, the one Tesla calls “our first design meant for mass production”, has still not been formally unveiled.

The reveal has slipped repeatedly and publicly. Musk indicated on the Q3 2025 earnings call that V3 would probably arrive in Q1 2026. On 31 March 2026, with the quarter running out, he posted on X: “Optimus 3 is walking around, but needs some finishing touches before it's ready to be shown.” On the Q1 2026 call on 22 April he pushed it again, to roughly the middle of the year, with Tesla indicating it wanted to hold the design back until production was close so competitors would have less time to study it. As of early September 2026, no unveiling event has taken place.

Which leaves an unusual situation, and it deserves to be stated plainly: Tesla is building a factory for a product it has not yet shown anyone.

Everything published about Gen 3's appearance and capability at this point comes from patents, a silhouette shown at a conference, and executive description. There is no demonstrated machine. Anyone giving you confident Gen 3 benchmarks is extrapolating.

The specifications that are actually published

Here is what can be sourced, with the caveat attached to each. Tesla does not maintain a public spec sheet for Optimus the way it does for its cars, so parts of this table come from Tesla presentations rather than a datasheet.

Note what is missing from that table: uptime, mean time between failures, task success rate, cycle time, cost per unit. Those are the numbers that decide whether a humanoid is a product or a research programme. Tesla has published none of them.

Demo footage versus shipped reality: the teleoperation record

Optimus has a documented history of demonstrations that were less autonomous than they appeared, and this is the main reason experienced robotics people discount Tesla's clips.

At the “We, Robot” event in October 2024, Optimus units mingled with guests, poured drinks and held conversations. They were being remotely assisted by human operators, which Tesla did not disclose from the stage. Robert Scoble, who attended, said publicly that the robot bartender was not wholly AI and that “a human is remote assisting”. Morgan Stanley analyst Adam Jonas wrote in a note that the bots “relied on tele-ops (human intervention)”.

The pattern recurred. At Tesla's “Autonomy Visualized” event in Miami in December 2025, an Optimus handing out water bottles dropped them and fell backwards. Immediately before falling, its hands moved up to its face in a grasping motion that looks a great deal like a person pulling off a VR headset. The robot was not wearing one. Musk has denied that the demonstrations are teleoperated, saying of an October 2025 clip “AI, not tele-operated” and telling an earnings call that at the Tron premiere Optimus was doing kung fu with Jared Leto and “nobody was controlling it”. Tesla has published no operator logs and no independent verification either way.

Teleoperation is not fraud in itself. It is a standard and legitimate part of robot development: you drive the machine through a task to collect the demonstration data the learning system needs. The problem is disclosure. When a clip is presented as autonomous capability and is not, every subsequent claim from the same source has to be discounted. That is exactly the tax Tesla is now paying on Gen 3 announcements it has not shown.

Why it slipped: magnets, hands, and 10,000 unique parts

The delays have identifiable causes, and they are worth understanding because they apply to the whole humanoid sector, not just Tesla.

The first is supply chain. On Tesla's Q1 2025 earnings call in April 2025, Musk said Optimus production had been hit by China's restrictions on rare earth magnet exports, describing it as a “magnet issue” and noting the magnets were needed for the robot's arm. China had blocked exports of seven rare earth elements earlier that month; the licensing process runs through China's Ministry of Commerce and can take weeks to months. High-performance neodymium magnets are not a nice-to-have in a humanoid. They are what makes a compact, high-torque actuator possible at all, and a humanoid has dozens.

The second is the hand. Dexterous manipulation is the hardest unsolved problem in the field, and Tesla's own documents track it: Gen 2 shipped 11 degrees of freedom per hand, Gen 3 is built around a new hand roughly doubling that. More degrees of freedom means more actuators, more tendons, more calibration and more failure points, packed into the smallest volume on the machine.

The third is manufacturing itself. On the Q1 2026 call Musk described the challenge as roughly 10,000 unique parts on an entirely new line and said it “will move as fast as the least lucky, slowest, dumbest part in the entire 10,000”. On 1 July 2026, after walking the newly installed Fremont line, he pushed back on speculation that output had quietly begun, writing: “No, Optimus production will be extremely slow at first, as everything is new. This is not like making a car.” On the Q2 call he called Optimus the hardest product Tesla has ever had to scale, because there is no existing supply chain for it.

Those are credible engineering explanations. Taken together, they are also reason enough to treat any near-term Optimus volume forecast, Tesla's own included, as provisional.

How rivals compare on verified deployment

The most useful context for Optimus is that Tesla does not lead on the one metric that counts: robots doing repeatable paid work in someone else's building. On that measure two smaller companies are ahead.

The most telling number in that table may be Figure's 84-second cycle time. On a factory line, cycle time is how often the robot repeats the same task, and it sets the pace of everything around it. Read alongside placement accuracy above 99% per shift, it describes a production tool rather than a demonstration. Optimus has no equivalent for either figure yet.

Even those rival figures deserve care. Agility has never published an installed-base count, so unit-count estimates for Digit vary by a factor of three or more depending on who is guessing. What is verifiable is different in kind: operating hours, totes moved, parts loaded, named sites. Those are the numbers to trust, because a customer's facility manager would notice if they were wrong.

Sector-wide, the credible estimate of commercially operational humanoids worldwide in 2026 is in the single-digit thousands, and most of that is Unitree hardware sold to labs and developers rather than machines doing industrial work. Claims you will see of tens of thousands of deployed units, including cumulative Optimus figures in the 50,000 range, do not come from the companies described and do not survive a check against filings.

The credible sceptical case

The strongest criticism of the humanoid thesis does not come from Tesla short-sellers. It comes from Rodney Brooks, who co-founded iRobot, ran MIT's computer science and AI lab, and has been building commercial robots for four decades.

In an essay published on 26 September 2025 titled “Why Today's Humanoids Won't Learn Dexterity”, Brooks argued that the plan for humanoid robots to become plug-compatible replacements for human manual labour is not achievable on the timelines being sold: believing it will happen, he wrote, “any time within decades is pure fantasy thinking”. His technical argument is specific rather than dismissive. The breakthroughs in speech, vision and language all rested on carefully engineered representations of their input domains, and no equivalent exists for touch. Learning dexterity from video, without the tactile data that makes human hands work, is in his view attacking the problem from the wrong end.

Brooks is not saying humanoids fail. He expects plenty of them in fifteen years, but says “they will look like neither today's humanoid robots nor humans”: wheels instead of legs, varied numbers of arms, sensors in non-human places and task-specific end effectors instead of five-fingered hands. That is a forecast that the bipedal, anthropomorphic form factor is partly a marketing decision.

The narrower scepticism about Tesla specifically is about track record. Optimus targets have moved every year since the programme was announced in 2021: production-ready by 2023, then limited production with over a thousand units in 2025, then 5,000 for internal use in 2025, then roughly 10,000 units in 2025, then a Gen 3 reveal in Q1 2026, then mid-2026, then production start in July or August 2026, then “later this year”. Each individual slip has a reasonable explanation. The pattern is the signal.

What to watch, and how to read the next Optimus story

If you want to track this without being manipulated by either the hype or the backlash, there are a small number of reliable checkpoints.

• The capacity table in Tesla's quarterly update. When California Optimus moves from “Construction” to “Production”, that is real and Tesla is accountable for it. Nothing on X is equivalent.
• An Optimus line in the quarterly production and deliveries release. Tesla reports what it can count. The day a robot number appears there, the programme has crossed from research into manufacturing.
• A Gen 3 unveiling with a hands-on press session. A staged clip is not evidence; journalists and roboticists handling the machine is.
• An outside customer with a named site. Figure and Agility both cleared this bar. Tesla has not attempted it.
• Boring operational metrics. Operating hours, cycle time, success rate, intervention rate. Companies publish these when the numbers are good.

Specification claims are easy to filter too. Tesla keeps no public spec sheet for Optimus, so any value presented to you as an “official spec” was either lifted from a presentation or compiled by somebody else. Knowing that difference removes half of what you will read.

One more filter: if a claim about Optimus includes a specific fleet size, ask where the number came from. Tesla has never published one. Every fleet-size figure in circulation — 1,000 at Fremont, 15,000 by year-end, 50,000 cumulative — is a third-party estimate or supply-chain rumour dressed as a fact. Some may turn out to be roughly right. None is currently verifiable.

My own read, for what it is worth: Optimus is a genuine engineering programme with real hardware, a real factory conversion and a real chip behind it, and it is running roughly two to three years behind what was promised for it. Both halves of that sentence are true, and most coverage picks one.

FAQ

Are Optimus robots working in Tesla's Fremont factory right now?

No. Tesla's Q2 2026 shareholder update, published 22 July 2026, lists the California Optimus facility with status “Construction” and no installed capacity, and says the first-generation lines are still being installed. On the January 2026 earnings call Musk said Optimus is “not in usage in our factories in a material way”. Earlier prototypes did perform some basic factory tasks, but Tesla deprecates old versions as it iterates.

How much can Optimus lift?

The commonly cited figure is about 20 kg, which is 45 pounds. Claims of 45 kg come from a botched unit conversion. Tesla does not currently publish an official payload specification.

Which Optimus generation is current in September 2026?

Generation 2, revealed in December 2023, plus an interim revision Tesla calls version 2.5. Generation 3, described by Tesla as its first design intended for mass production, has been announced but not formally unveiled, after reveal dates slipped from Q1 2026 to mid-2026 and then passed without an event.

Can you buy an Optimus?

No. There is no public sale, no price, no deposit and no waitlist. Tesla stated that initial builds will go to an internal “Optimus Academy” for training data collection and functionality development. Any site offering pre-orders is not selling a Tesla product.

How many Optimus units has Tesla built?

Tesla has never disclosed a number, and no Optimus line has ever appeared in its quarterly production reports. Reporting has suggested a few hundred units against a 5,000-unit internal target for 2025, but that is not company-confirmed. Treat every specific fleet figure you see as an estimate.

Is Tesla ahead of other humanoid robot companies?

Not on verified deployment. Agility Robotics reports more than 65,000 Digit operating hours across nine customer facilities with named customers including GXO and Schaeffler, and Figure ran an eleven-month pilot at BMW's Spartanburg plant logging over 1,250 hours and 90,000 parts loaded. Tesla has no external customer and no published operating hours. Tesla's potential advantage is manufacturing scale, which is exactly the part it has not yet demonstrated.

Were the Optimus demo videos remote-controlled?

Some were. At the October 2024 “We, Robot” event the robots were remotely assisted by human operators, which Tesla did not disclose on stage; a Morgan Stanley note afterwards said the bots relied on tele-ops. Musk has denied that later demonstrations are teleoperated. Tesla has published no logs or independent verification, so treat any single clip as insufficient evidence of autonomy.

Sources

• Tesla Q2 2026 Update (shareholder deck, 22 July 2026)
• Tesla Q1 2026 Update
• Tesla Q4 2025 Update
• Tesla Q3 2025 Update
• Tesla Second Quarter 2026 Production, Deliveries & Deployments (SEC Form 8-K, 2 July 2026)
• Tesla Investor Relations: Q2 2026 financial results
• Electrek: Musk admits no Optimus robots are doing useful work at Tesla (28 January 2026)
• Electrek: Tesla pushes Optimus V3 reveal later this year, again (22 April 2026)
• Electrek: Musk shuts down Optimus “4D chess” theory (2 July 2026)
• Electrek: Optimus takes a suspicious tumble in new demo (7 December 2025)
• Rodney Brooks: Why Today's Humanoids Won't Learn Dexterity (26 September 2025)
• Humanoid Guide: Humanoid deployments in 2026 favour Figure and Agility (19 July 2026)
• The Register: Musk distracts Tesla investors with Optimus hype (29 January 2026)
• Interesting Engineering: Optimus robots remotely supported by humans at We, Robot
• The Hill: Musk says China's blocking of rare earth magnets will hit Tesla humanoid robots
• Not a Tesla App: Tesla delays Optimus Gen 3 unveil for “finishing touches” (31 March 2026)
• Wikipedia: Optimus (robot), for generation history and collected specifications