Optimus Is Now a Production Line, Not a Demo — in the Same Quarter Tesla’s Profit Fell 43%

中文版 Chinese version

All figures are as reported October 21–25, 2026 by Tesla’s Q3 earnings call coverage (TechCrunch, Electrek, TheStreet) and Bank of America’s October 22 research note as reported by TheStreet. Production targets are company goals, not output; analyst models are estimates, not facts. This is analysis, not investment advice.

What changed this week

Humanoid robots have lived in demo-land for a decade: a stage, a tether, a choreographed wave. This week Optimus left that world in one specific, checkable way — Tesla has started production on a high-volume line at Fremont, as Musk confirmed on the October 21 Q3 earnings call and as TechCrunch’s October 25 Mobility roundup summarised. The stated target is a rate near 20,000 robots per week by the end of 2027 — roughly one million units a year. Bank of America analyst Alexander Perry, in an October 22 note, put the honest caveat next to it: reaching ~1M a year “will likely take at least a year.” Reporting on the call also describes an internal Optimus Academy — a training fleet of 10,000–20,000 units that learn tasks by doing them, self-play style, before any external sales, which Musk has said could come in late 2027.

Musk has previously claimed Optimus could eventually be ~80% of Tesla’s value. Treat that as a claim about narrative, not arithmetic — but note that the company is now spending like it believes it.

The quarter underneath

The same call reported a core business under strain:

• Profit down 43% year over year, on revenue that was flat at $28.05 billion (adjusted EPS of $0.48 beat estimates, per Electrek’s reporting).
• Operating expenses up 39%; R&D up 69% to $2.76 billion in the quarter.
• Tesla cut its 2026 capital-spending target to ~$20 billion from $25 billion, per CFO Vaibhav Taneja on the call.
• And Perry’s model — from a bull — has free cash flow at roughly −$4.6 billion in 2026 and −$11.3 billion in 2027, with capex rising to ~$25.4 billion in 2027 and ~$9.5 billion of added debt, against $6.2 billion of positive FCF for all of 2025.

That is the trade, stated in the bank’s own numbers: Robotaxi, Optimus, and AI chips are funded simultaneously, and the cash flows out well before any of them flows in.

The one curve that bent

Against that, the Robotaxi data genuinely changed character this quarter. Per Perry’s note:

• Unsupervised miles: 380,000 (July 22) → 1 million (September 3) → ~2 million (September 30) — doubling in about four weeks, at ~20% weekly growth after the two-seat Cybercab joined the Austin fleet; Cybercabs are now 45% of the Texas fleet.
• CFO Taneja: cumulative unsupervised miles grew 8× in the quarter.
• Over 75% of North American buyers now take paid FSD at delivery, up from over 55% in Q2.
• Tesla plans to sell Cybercabs to customers by the end of the year; unsupervised FSD for customer cars slips to “somewhere around Q2 next year” (VP of AI software Ashok Elluswamy).
• Perry kept his Buy and $460 price objective (~19% above the $386.47 reference price) while cutting 2027 EPS to $2.26 from $2.47 — a bullish rating on a worsening near-term P&L.

And the necessary cold water, also in the note’s coverage: Waymo had driven 171 million fully autonomous miles by the end of 2025. Tesla’s 2 million is a little over 1% of that. At a sustained 20% weekly growth the gap closes arithmetically; growth from a small base is also, as TheStreet notes, the easiest kind to post.

The argument

The bear case writes itself: a car company with flat revenue and a 43% profit drop is building a humanoid-robot factory, a robotaxi fleet, and an AI chip programme at once, on modelled negative free cash flow, against a competitor (Waymo) two orders of magnitude ahead in miles. Every previous Musk timeline in this space has slipped — unsupervised FSD slipped again on this very call.

The bull case is narrower than the hype, and stronger for it. Production is a different claim from a prototype: lines force design-for-manufacture, unit costs become measurable, and the Academy fleet turns “the robot doesn’t work yet” into a data problem Tesla knows how to run. The Robotaxi curve — 8× in a quarter, with paying FSD attach above 75% — is the first autonomy metric from Tesla that looks like a business rather than a promise.

My judgment: believe the factory, discount the calendar. The Fremont line and the miles curve are real, verifiable, and new. The 20,000-a-week target and the late-2027 external deliveries are Musk-standard scheduling — assume slippage. What is not deniable after this quarter is the sequence: Tesla will spend the car business’s cash on robots and robotaxis first and show the profit later, if ever. Perry’s note is the whole story in one document — Buy rating, $460 target, and minus $11.3 billion of 2027 free cash flow. Wall Street is no longer valuing whether Tesla pivots. It is valuing whether the pivot pays before the bill does.

Sources: TechCrunch Mobility (October 25, 2026); TheStreet (October 24, 2026) reporting Bank of America analyst Alexander Perry’s October 22 note; Electrek’s Q3 reporting as cited by TheStreet; call remarks by Elon Musk, Vaibhav Taneja, Lars Moravy and Ashok Elluswamy (October 21, 2026). Related: our October 22 read of the SpaceX–Tesla merger committees — the same week, the same bet, at corporate scale.