Tesla Optimus explained: what Tesla has actually filed, a 1 million-robot line in Fremont, a 10 million plan in Texas and no unit count yet

Tesla has told the US Securities and Exchange Commission that its first humanoid robot factory line is designed to build 1 million Optimus robots a year, and that a second line in Texas is being designed for 10 million. It has not told the SEC how many it has built. Through its Q2 2026 update and 10-Q, no filing gives an Optimus unit count, an Optimus revenue line or a named customer, and the first robots off the line are earmarked for Tesla's own use.

That gap between capacity and output is the most useful thing to understand about Optimus, because it is what the next filing, the Q3 2026 update due after the US market close on Oct 21, can close or widen. This explainer sets out only what Tesla has put in its SEC filings, what those figures mean next to the humanoid volumes other makers disclose, and which numbers would show the programme is real at scale.

Tesla's filings describe a factory before they describe a product

The sequence is clear from four quarterly updates. In the Q4 and FY2025 update of Jan 28, Tesla said it planned to unveil a third-generation Optimus in Q1 2026, the first design meant for mass production, and to start production on its first line before the end of 2026, with eventual capacity of 1 million robots a year. On Apr 22, the Q1 2026 update said preparations for the first large-scale Optimus factory would begin in Q2, that the first-generation line designed for 1 million a year would replace the Model S and Model X lines in Fremont, and that Gigafactory Texas was being prepared for a second-generation line designed for long-term annual capacity of 10 million.

The Q2 2026 update of Jul 22 reported progress on the building, not the robot. Tesla said it had decommissioned the S and X lines and was installing the first-generation Optimus lines, where it expected to start production soon, and that construction of the Texas building was in full swing. Its capacity table listed both the California and Texas Optimus sites as Construction, while Cybercab, which began production in Texas in the same quarter, was listed as Production.

Two details in the filed text deserve attention. The Q1 outlook said first-generation lines were being installed "in anticipation of volume production"; the Q2 outlook said "in anticipation of production in 2026", without the word volume. And the Gen 3 unveiling promised for Q1 does not appear in either the Q1 or the Q2 update as filed. Neither proves a delay, but both are what a careful reader of the Oct 21 update should check first.

The first customer is Tesla itself

The Q2 update states that the initial Optimus builds will be used in an internal Optimus Academy for training data collection and further functionality development. That makes the early robots a data asset rather than a product. Tesla's Q2 update also said its Cortex 2 training cluster in Texas supports both vehicle and humanoid robot autonomy software, the same pattern Tesla used with its car fleet: deploy hardware, collect data, train centrally.

The pay plan shows why the count of delivered robots will matter later. The CEO award approved by shareholders on Nov 6, 2025, with 1,892,235,822 votes for and 564,940,908 against (77.0% of votes cast either way), covers 423,743,904 restricted shares in 12 tranches and ties vesting to market-value steps from USD 2 trillion to USD 8.5 trillion plus operational milestones. One milestone is 1 million Bots delivered cumulatively, with a Bot defined in the proxy as any AI-driven mobile robot or physical product made by or for Tesla, excluding vehicles. Over the 10-year performance period that is an average of 100,000 a year.

What 1 million a year means next to every other humanoid maker

The International Federation of Robotics counted about 7,000 full-size humanoids, those taller than 140 cm, sold worldwide in 2025 for commercial and professional use. Tesla's Fremont design capacity is about 143 times that number, and the Texas long-term design about 1,430 times. The makers that do publish volumes are mostly Chinese: Unitree shipped more than 5,500 humanoids in 2025 according to its STAR Market prospectus, and UBTech sold 921 full-size units in the first half of 2026. In the US, Figure AI said it had delivered more than 350 Figure 03 robots by Apr 29, and 1X Technologies said its Hayward plant can build up to 10,000 NEO a year.

The comparison cuts both ways. No humanoid maker has run a line at even a hundredth of Tesla's stated design rate, which is the bear case in a sentence. Tesla, on the other hand, delivered 486,532 vehicles in Q3 2026 alone, per its Oct 2 filing, and it is converting existing car lines rather than building a start-up factory, which is the bull case. Capacity is a statement about tooling. Output depends on a working robot, on parts supply and on a task worth paying for.

The money is there, but Optimus is not broken out

Tesla does not report Optimus spending separately. What it does report is a company spending at a pace it has not seen before. Capital expenditure rose to USD 5.79 billion in Q2 from USD 2.49 billion in Q1, 2.3 times higher, and the Q2 10-Q says 2026 capex is expected to exceed USD 25 billion, driven by AI initiatives including compute, manufacturing lines and facilities. With USD 8.28 billion spent in the first half, that implies more than USD 16.72 billion in the second. Research and development rose 49% year on year to USD 2.37 billion in Q2, while operating income fell to USD 398 million, a 1.4% margin, and free cash flow was minus USD 1.09 billion. On Sept 29, Tesla signed USD 30.0 billion of unsecured credit facilities, a USD 20.0 billion delayed-draw term loan and USD 10.0 billion of revolvers, on top of USD 43.52 billion of cash and investments at the end of June. The 8-K does not tie the facilities to Optimus.