Tesla's Optimus Line Has Started in Fremont; the 20,000-a-Week Target Is 14 Months Out and Has No Unit Count Behind It
Hardware / analysis
Tesla's Optimus Line Has Started in Fremont; the 20,000-a-Week Target Is 14 Months Out and Has No Unit Count Behind It
Third-quarter capex was $4.73 billion and net income $787 million, while the first production-quality robots are described as early output from a ramp Elon Musk calls 'somewhat bumpy'.
A target of 20,000 humanoid robots a week by the end of 2027 works out to 1.04 million a year, and nothing in the coverage of Tesla's letter says how many have been built. The company said on October 21 that production has begun at its Fremont, California, factory in space that once built cars, and called the units off the line "the first production-quality bots that we have manufactured". It did not say how many exist.
The photograph above is an earlier Optimus prototype shown in Stockholm in May 2024, not a unit from the new line.
What Tesla said about the Fremont line
The statement comes from Tesla's third-quarter shareholder letter, as quoted by TechCrunch reporters Kirsten Korosec and Sean O'Kane. Chief executive Elon Musk called the stage "early production" and said the ramp will be slow and "somewhat bumpy". He kept the end-of-2027 goal of 20,000 robots a week.
Musk also said the design needs changes for safety, reliability and long-duration operation, and that more A.I. work remains before the robot is useful for open-ended tasks. That is a production line turning out machines whose software is not ready for the jobs the volume target implies.
TechCrunch's account gives no weekly rate today, no customer, no price and no payload or runtime. For another hardware story that turns on a capacity figure, see Rio Tinto's 360 MWh heat battery.
The money: $4.73 billion of capex in one quarter
Tesla reported Q3 revenue of $28.04 billion, flat on $28.09 billion a year earlier, and net income of $787 million, down 43% from $1.37 billion. Expenses rose on spending for A.I., Optimus and autonomous driving, per TechCrunch.
Electrek's reading of the 10-Q supplies the spending side. Capital expenditure was $4.73 billion in the quarter, after $5.79 billion in Q2 and $2.49 billion in Q1, for $13.01 billion over nine months. The shareholder letter did not change the 2026 capex guidance; the 10-Q did, from "more than $25 billion" set in April to "to be in excess of $20 billion".
- Q12.49 $ billion
- Q25.79 $ billion
- Q34.73 $ billion
Source: Electrek analysis of Tesla's Q3 2026 10-Q, accessed 2026-10-22
The 10-Q says 2026 spending is "driven by our SI initiatives, including investments in compute infrastructure and data centers", plus factory ramps. Optimus is one of several lines inside that number, and neither source breaks out its share.
Free cash flow of $1.53 billion rests on working capital
Operating cash flow was $6.26 billion and free cash flow $1.53 billion, but Electrek calculates that a $2.59 billion working-capital benefit produced most of it; without that, free cash flow would have been about negative $1.06 billion. Unpaid equipment and construction spending reached $3.67 billion at the end of September, up from $1.63 billion a year earlier.
The cushion is financing. Tesla signed $30 billion of credit facilities a month earlier, a $20 billion delayed-draw term loan maturing in 2029 and a $10 billion revolver, with nothing drawn at quarter-end and $45.4 billion in cash and investments.
| Item | Figure |
|---|---|
| Q3 2026 revenue | $28.04 billion |
| Q3 2026 net income | $787 million |
| Nine-month capex | $13.01 billion |
| Optimus target, end of 2027 | 20,000 a week |
Reading the guidance cut against the Optimus target
The number that matters here is not the capex cut but what it leaves unexplained. At $13.01 billion spent, a $20 billion year implies about $7 billion in Q4, up from $4.73 billion in Q3; the old $25 billion guidance would have needed about $12 billion. The cut lowers this quarter's bill. It says nothing about 2027, the year in which a line that starts at an unstated rate would have to reach 20,000 a week.
That is arithmetic on two reported figures, not a Tesla statement, and the guidance language is not a plan by product. Optimus capex could be small next to data centers.
Setting 20,000 a week against Tesla's car output
Tesla delivered 486,532 vehicles in the third quarter, about 6,000 more than in Q2 but fewer than a year earlier. Spread over 13 weeks that is roughly 37,400 cars a week. A line making 20,000 robots a week would therefore match more than half of the company's entire weekly vehicle volume, in a product Musk says still needs design changes.
The call carried a second 2027 date with a track record. Ashok Elluswamy, Tesla's vice president of A.I. software, said the first customer cars could join the robotaxi fleet around Q2 2027, when owners' vehicles would receive a software update allowing unsupervised driving. In April, Musk had said the unsupervised version would "probably" arrive in Q4 2026. A schedule that slid two quarters in six months is the nearest evidence for how Tesla's software dates for Optimus might behave.
What would change my read
A disclosed weekly build rate would settle most of this. So would a named customer; neither source names one. If the fourth-quarter letter reports units built, a price or an hours-of-operation figure, the 20,000 target becomes a number with a trend behind it. Until then it is a goal set by Musk against a ramp he himself calls bumpy, and the only hard numbers in the filing are the dollars going out.
For related robotics funding coverage, see Warp Sciences' $215 million round.
Sources
More in Hardware
- 01Washington Finds Robot Imports a Security Threat and Sets No Tariff, Date or Product ListThe October 28 Section 232 proclamation orders negotiations rather than duties, so the cost of the finding sits in components nobody has priced yet.
- 02Electric Trucks Hit 50.3% of China Heavy-Truck Sales in September, in a Market Down 21%September sits below the December 2025 peak of 53.9%, and electric sales grew 37% while total sales fell, and the 2026 year-to-date figure is 38.1%.
- 03Ola Electric Registered 12,096 Scooters in October, Down 10%, as India Added 7% to 223,847Entrackr and Autocar Professional use different Vahan pulls for September, which moves the growth figure from 7.3% to about 10%.
- 04Battery Self-Tests Cause a One-Minute Dip in Grid Frequency, and Jülich Measured It on a 1.5 MW UnitA Forschungszentrum Jülich team traced a worldwide 60-second ripple in grid frequency to battery power supplies that all test themselves at the top of the minute. The preprint measured one site; the scale claims come from a press release.