Economics

Every Line in Tesla's Capacity Table Has a Number. The Robot Lines Have a Dash.

Tesla's Q3 update puts installed annual capacity against Model 3/Y, Cybercab, Semi, Megapack and three battery-material lines mid-ramp. Against both Optimus lines it prints a dash. The 400 megawatts of compute behind the robots is disclosed in megawatts; the robots are described in adjectives.

Duncan Galbraith

Duncan Galbraith

— 5 min read
Share

On October 21 Tesla filed its Q3 2026 update as an exhibit to a Form 8-K, and it is a more forthcoming document than the genre usually produces. Revenue by line, gross margin, operating income, free cash flow, capital expenditure. Cumulative deliveries to the tenth of a million. Active FSD subscriptions, operating-lease counts, days of inventory, storage deployed in gigawatt-hours, supercharger connectors counted to the unit.

It also carries a manufacturing capacity table: installed annual capacity and status, region by region. Model 3 and Model Y in California, more than 550,000. Shanghai, more than 950,000. Cybercab, more than 125,000. Tesla Semi in Nevada, 25,000. Megapack in Texas, 50 gigawatt-hours. LFP cells in Nevada, 7 gigawatt-hours, early ramp. Cathode material in Texas, 10, early ramp. Lithium refining, 30, early ramp.

And then, under Robotics: California, Optimus, a dash, "Early Ramp." Texas, Optimus, a dash.

A company that counts supercharger connectors to the unit does not count robots. That is the finding, and it is narrower and more useful than the general complaint about Tesla disclosure — because this is not an issuer withholding numbers. It is an issuer publishing them, row by row, and printing a dash in the two rows its humanoid robot occupies.

A dash is not evasion by itself. There are three other dashes in that table. The Next Gen Roadster's capacity is a dash, and its status reads "Design development." Berlin's 4680 cells are a dash, and their status reads "Construction." Both mean what a dash ought to mean: there is no installed capacity yet to state.

Optimus is the only line in the table that is labelled as ramping and carries no number. Every other line marked early ramp in this document carries one — 7 gigawatt-hours of LFP cells, 10 of cathode material, 30 of lithium refining. The document's own convention is that products in ramp get a figure. The robot is the exception, and the label is the reason it is interesting: "Early Ramp" is itself a claim that ramping has begun.

What Tesla discloses instead is the input.

Under supporting infrastructure, the same filing reports training compute: Cortex 1 in Texas at more than 105 megawatts, in production; Cortex 2 at more than 120, in production. "We plan to reach 400 MW of SI training compute by the end of the year," it says, "supporting the development of both vehicle and humanoid robot autonomy software."

Read that sentence twice. The capital going into the robotics programme is legible in megawatts. The capex is legible in dollars — $4.732 billion for the quarter, against $5.789 billion in the second, on free cash flow of $1.531 billion, with the spend described as covering "SI, robotics and energy" without a split. The software is legible by name: the segment itself is now called Super Intelligence.

The output is a dash.

That asymmetry is the substance of the piece, because it is the wrong way round for anyone trying to price the claim. Inputs are disclosed; volume is not. A reader can see what is being spent on robots and cannot see how many robots there are or how many the installed lines could make. Every other product in that document is auditable at the point where a production claim becomes checkable — the unit.

Tesla's own metrics table makes the same choice. It counts vehicles, FSD subscriptions, leases, inventory days, storage gigawatt-hours and charger connectors. It counts no robots. The one place robots could plausibly have appeared is a footnote, and there they do not: footnote 2 to the operational summary defines the metric tied to the 2025 CEO Performance Award as new vehicles delivered "plus any unsupervised Robotaxis placed into commercial operation." Robotaxis enter the count that governs the largest compensation package in corporate history. Robots enter nothing.

The strongest defence is real, and I want it stated before I answer it. A pre-revenue product in early ramp may genuinely have no stable capacity to publish. If Optimus comes off the same Fremont lines as vehicles — which the filing says it does, "on our high-volume lines" — then capacity is joint and any single figure misallocates it. Publishing a number early invites a milestone the company then has to hit, and no securities rule requires an issuer to guide on a product it does not yet sell. On that reading the dash is restraint: the company declining to invent a figure.

The joint-production point, though, cuts the other way, and it is the part I would press. If robot capacity is taken out of vehicle capacity, that is a cost to the vehicle business, and it is only visible if one of the two numbers moves. Keeping the robot row empty keeps both rows quiet.

And the disclosure being asked for here is not a revenue forecast. It is the same class of number Tesla prints for a cell plant that is still commissioning. The company says production began in Fremont and that "the bots coming off this line are the first production-quality bots that we have manufactured," and that it has "proven that the current manufacturing process for Optimus is suitable for a commercial product." A process that has been proven suitable for a commercial product is a process with a rate. The dash is not a refusal to guess. It is the absence of the one quantity the assertion implies.

Industrial claims have always been audited through volume, because volume is the part of a production claim that can be checked. Every other figure in this document is a volume. The robots get adjectives.

None of this says the robots do not exist. It says the document is built so that a reader cannot know, in the quarter the company converted part of a high-volume vehicle line to the programme, what came off it. The costs arrive in the same document as the claim. The count does not.

This is the third quarter running that I have written about the capital structure of this buildout — what a useful life assumption costs when it changes, what a guarantee is worth when it expires before the bond — and the same property holds. The number that determines whether the arrangement works is the one the arrangement does not publish.

What I don't know.

I read the Q3 update in full. I have not read the 10-Q filed the following day, which also references Optimus; if it carries a production figure, this piece is wrong and I will correct it. I have not read an earnings-call transcript, so I do not know whether an analyst asked for a unit count or was refused one. In the rendering I read, the Texas Optimus row shows no status label at all, which may be an extraction artefact rather than a blank in the original. And I have not read the paywalled trade reporting that has published production estimates, so I am not in a position to say whether any outside figure is well sourced. The EDGAR full-text search I ran returns Optimus in Tesla's current and periodic reports; what it does not return is a count.

Sources

Read more