Tesla (TSLA) — Q3 FY2026 Earnings Briefing
Fiscal period: Q3 FY2026 · Earnings call: October 21, 2026
Executive Summary
- A demand-led quarter: automotive gross margin ex-credits reached 18.1%, up from 16.3% last quarter, even as commodity costs and rate changes built headwinds.
- Tesla returned to positive free cash flow, generating more than $1.5 billion in the quarter while funding an expanding autonomy-and-robotics buildout.
- Deliveries were production-constrained rather than demand-constrained — the growth story is now an output-scaling story.
- Robotaxi is compounding on an exponential weekly cadence, with CyberCab newly deployed in Austin and a multi-city expansion under way.
- The long-duration bets (Optimus, TeraFab) remain execution-stage, and the company has booked its costliest capital program ever for 2028 and beyond.
Highlights and Lowlights
Highlights
- Vehicle deliveries topped 480K for a second consecutive quarter, and Tesla exited the period with its highest backlog in four years — output, not demand, is the binding constraint.
- Paid FSD reached 1.7 million customers globally, with subscriptions becoming half of that paid base for the first time — recurring revenue now anchors the attach story.
- Cumulative unsupervised robotaxi miles grew 8x in the quarter, with double-digit week-over-week growth every week; management characterized the curve as exponential.
- Energy storage deployed 13.7 GWh, the second-best quarter on record, with Megapacks effectively sold out into 2027.
- Tesla arranged credit capacity of up to $30 billion to fund the coming buildout, with no draws planned this year — near-term growth stays internally funded.
- Automotive margins ex-credits inflected higher as APAC mix outweighed commodity and financing headwinds, while services margin set an all-time high — see the P&L table below.
- The Model Y set global share records: the number-one selling BEV in Europe, the top-selling passenger car in South Korea, and fresh records in Australia, Japan and Taiwan per preliminary data.
Lowlights
- Optimus external deliveries remain about a year away, gated by an AI architecture still being worked, durability and use-life questions, and a hand design not yet converged.
- The capital program now stretches years beyond this one: TeraFab alone absorbs tens of billions through early 2028 and is headed north of $100 billion long-term.
- Energy’s sequential margin jump was flattered by tariff refunds and a vendor warranty true-down — management itself flagged both as non-underlying gains.
- Operating expenses keep climbing sequentially on research-led spend, and the trend persists until Optimus, Megapack 3 and Semi actually commercialize.
- Net income absorbed a $142 million mark-to-market write-down on the SpaceX stake, only partly offset by a roughly $290 million Bitcoin gain.
- EU-wide FSD approval timing sits outside Tesla’s control; Musk “would be surprised” if it slips past Q1 but calls that a prediction he doesn’t control.
- CyberCab economics still miss the per-paid-mile target management has set, and a direct question on NHTSA’s open safety review went unaddressed by management.
1. Demand, Backlog and Production
Underlying every other theme is a simple mechanic (see Highlights): demand outran production and has for two straight quarters. Chief Financial Officer Vaibhav Taneja called the four-year-high backlog the clearest demand signal and said Tesla is “focused on increasing production at all our factories,” warning that dialing in supply chain, manufacturing and delivery logistics takes time. That is why the growth constraint, not pricing, drives the P&L narrative this quarter.
The Model Y is the workhorse behind the share gains highlighted above, and its momentum doubles as the FSD carrier story developed in the next section. On capacity, VP of Vehicle Engineering Lars Moravy said all four existing factories ran at record rates, with earlier strategic investments now coming to fruition, plus three expansions in motion: CyberCab production ramping at Giga Texas, permit applications filed to expand Berlin, and the new Semi factory in Reno just getting under way. Musk put the ambition quantitatively: Tesla is aiming for roughly 5 million vehicles a year by approximately 2030, supported by Semi production that began in Q3. The new Roadster demo — Musk: “arguably the coolest demo of any product ever” — rounds out the own-car pipeline alongside the Model 3/Y fleet.
2. P&L Mechanics: Margins, One-Offs and OpEx
| Margin run (sequential) | Q2 2026 | Q3 2026 | Driver note |
|---|---|---|---|
| Energy gross margin | 20.4% | 33.1% | Gains flattered by tariff refunds and vendor warranty true-downs (see Lowlights); ASP and margins expected to improve further on data-center demand |
| Services and other | 14.1% | 15.9% | All-time high, driven by strength across Supercharging and insurance; robotaxi sits inside this segment |
The automotive margin itself (headline in the Executive Summary) rose on a higher APAC delivery mix; Taneja paired the good news with the countervailing trend that commodity price increases and interest-rate changes “continue to add to our cost and create further headwinds,” partially offset by adjusted vehicle-financing programs. That duality — mix tailwind today, cost headwinds accumulating — is the margin question for next quarter (see Artifacts).
Two other mechanics matter for quality of print. First, the energy margin’s jump is one-off-led (see the table and Lowlights), even though the underlying demand picture is genuinely strong — “extreme demand from data centers,” per Taneja. Second, the one-off noise in net income (SpaceX write-down versus Bitcoin gain, see Lowlights) complicates trend reads on earnings.
On operating expenses, the sequential climb continues as previously guided, and the investment cycle has a defined endpoint: OpEx grows “till we commercialize these new products” — enumerated as depreciation from additional AI compute, chip development, and pre-production ramp costs for Optimus, Megapack 3 and the Semi.
3. FSD: The Paid-Fleet Flywheel
The attach engine is quantified here for the first time on a call this cycle: over 75% of North American deliveries included a paid FSD subscription sign-up at the time of delivery — approximately 20% above Q2’s level — and every delivered North American car carries a month of free FSD subscription that buyers are paying against anyway. With the paid base (see Highlights) now majority-subscription, Taneja confirmed showroom-level demand inversion: customers who used to ask about Model 3/Y or Cybertruck now “come in and say, can we get into an FSD drive?”
New Street’s Pierre Ferragu put analyst math on the recurring-revenue value — roughly $7,000 of vehicle gross profit at sale, another $1,000-plus per year from FSD from the second year, and approaching $1 billion of ARR — and asked whether lifetime profit per car at least doubles. Taneja expressly confirmed the figures (“the numbers which you shared are correct”) and committed to disclosing more FSD metrics, including revenue and churn, as the base scales. Retention specifics are being withheld because free-trial cohorts in newly added countries muddy the read.
On competitive durability, RBC’s Tom pressed the inflation risk: he cited a 30–40% paid attach rate in North America (analyst-provided; management neither confirmed nor disputed it) and asked whether rivals partnering with NVIDIA or Qualcomm could compress FSD pricing. The answer was a moat argument, not a price argument: VP of AI Ashok Elluswamy called the long tail “so long that… anyone with their sane mind should not attempt” it, claiming no one matches Tesla’s unsupervised capability for several years. Geographically, Tesla is seeding Europe’s fleet with incentives ahead of a broader approval (EU timing — see Lowlights), and in China management said approval work is hard but ongoing with certain cars already FSD-active; Taneja added that competitor CEOs have been “blown away” by US FSD drives and that Tesla has published its unsupervised robotaxi safety report for regulators to inspect.
4. Robotaxi and CyberCab
| Dimension | Status at Q3 close | Commitment |
|---|---|---|
| Austin CyberCab fleet | Grown ~10x since launch about a month ago | Scale to millions of vehicles |
| Live cities | 6 cities now | Dallas next month, then “lots of other cities” along with Model Y robotaxis |
| Service hours | Expanded from 6 AM–9 PM to 10 PM, then 11 PM, about to reach around-the-clock at 2 AM | Full 24/7 operation unlocked before year-end |
| Road conditions | City surface streets | Higher-speed roads and highways plus wider weather/night coverage by end of 2026 |
| Unit economics | Below target | Goal of $0.30 per paid mile — not met today, but management sees “a very” clear path as operations scale |
| Customer ownership | Fleet is Tesla-operated | First customers can buy CyberCabs by end of 2026, though it remains mostly Tesla-owned transport-as-a-service |
| Customer fleet membership | Not yet | Around Q2 2027, when V15 software ships and customer cars can join geofenced fleet management |
| Revenue inflection | Not yet material | Not noticeable until roughly Q2 2027; “a big deal by Q3 or Q4 of next year” per Musk |
The scale mechanism is software reuse: CyberCab learned from Model 3/Y platform data and, per Elluswamy, “the same software will just scale to lots of other cities.” The V15 release brings that stack to customer cars broadly — and it also carries the new AI-based mapping and navigation that goes live on trip start and end “pretty soon,” then takes over navigation, parking and destination handling over coming months.
On capacity, Wolfe’s Emmanuel Rosner cited a listed CyberCab line capacity above 125,000 units a year (analyst-provided figure from Tesla’s published listing) and asked about ramp shape. Moravy said the listed figure is only the rate the line can hold today — actual capacity is “significantly higher” — noted the vehicle has 50% fewer parts than any Tesla before it, and said Tesla is “bursting” the line; production currently runs at the rate the service needs, racing the miles curve rather than the demand curve. Rosner’s triangulation of the mile math — sustaining roughly 15% week-over-week growth to cross 10 million cumulative miles by year-end (analyst-provided) — got no management mileage confirmation, only the revenue-timing commitment in the table above. Alexander Perry (Bank of America) noted Tesla prices below rideshare rivals (analyst observation consistent with the structural cost case: no driver, self-manufactured vehicle) and asked about sustained double-digit weekly growth; Elluswamy pointed to the V15 stack and years of end-to-end network work as the enabler and saw “no hurdles along the way.” Musk closed the loop with the ambition statement of the quarter: Tesla will end up capturing “a majority of all passenger miles on Earth in any country where it is allowed to operate freely.”
5. Energy Storage
Deployments were second-best on record (see Highlights), and the strategic shift this quarter is vertical integration: Tesla began deliveries of Megapacks built with in-house LFP cells. Demand is described as extreme from data centers, and the business case rests on scaling fast rather than price discovery: with Megapacks close to sold out for 2027, supply — not price — is the lever. Management expects average selling prices and margins to improve further as output scales, beyond the levels shown in the P&L table.
The forward product question is distributed compute. On the “Megapod” concept Andrew Percoco (Morgan Stanley) raised from last quarter, Musk kept it deliberately early-stage: better visibility expected around the end of Q1 or into Q2, with “a lot of potential” but no sizing yet. The countervailing margin-structure detail — what the Q3 margin number owed to one-offs — lives in the P&L table and Lowlights, not here.
6. Optimus
Production has started — “early production” on the high-volume Fremont line — with the vast majority of the ~1-million-unit-per-year line already landed there, and a second Optimus factory under construction in Austin that will exceed Fremont’s robot capacity. But Musk framed the ramp honestly around an S-curve whose first leg is shallow and bumpy: of the roughly 10,000 things that must go right, the program moves at the slowest. Reaching a 20,000-unit annualized weekly rate will take at least a year, maybe longer, and design is still churning — engineering lead Konstantinos identified the hand (including bespoke, wear-carrying gears) as the one electromechanical area unconverged while the rest has stabilized.
The learning-loop plan is explicit: a video — in roughly a month, deliberately not a canned demo — then bootstrapping via 10,000–20,000 robots doing self-play in “Optimus Academy” plus simulation, since unlike the car fleet there is no human exoskeleton to harvest data from. Hardware intelligence converges on the Tesla AI5 chip integrating into Optimus around April 2027 — an order of magnitude more compute than AI4 — which Musk sees as the step change enabling dexterity-plus-duration-plus-AI to compound. External-customer timing is the lowlight (see Lowlights), with first commercial use in industrial settings: dangerous, repetitive, 24/7 tasks before consumer availability. Musk repeated the end-state claim — humanoid robots as “the biggest product on Earth,” with volumes eventually exceeding the human population.
7. TeraFab and the Capital Program
| TeraFab milestone | Timing |
|---|---|
| Machinery commissioning, Austin research fab | First half of 2027 |
| First test chip (“just to see if we can make any chip at all”) | Around May–June 2027 |
| Useful chip at volume | Around Q3 2028, “lightning fast by fab standards” |
Taneja noted most orders for the research fab’s first equipment tranche are already placed. The cost side is the lowlight (see Lowlights): Musk sized TeraFab capital equipment at “not less than tens of billions” through early 2028 with a long-run bill “probably north of $100 billion,” and the SpaceX/Tesla cost split is still undefined. Musk’s escalator: the TeraFab becomes “the most valuable building in the world,” with economic value he compared to a top-20 national economy — framed explicitly as the bottleneck-breaker for scaling Optimus (memory, packaging, logic).
Funding architecture for the nearer term: full-year CapEx guidance was revised to approximately $20 billion for 2026, funded primarily from operating cash flow, with the tariff-refund proceeds reinvested into US production capacity — and Tesla nonetheless secured the multi-billion credit capacity highlighted above as a backstop for growth spending, which continues to expand across robotaxi fleet, Optimus plant, solar manufacturing, AI compute and the fab. Closing the question queue, Colin Rusch (Oppenheimer) tested the Semi/TeraFab supplier pipeline against CapEx; Musk’s answer was that current impact is modest, with CapEx now “a little lower than we anticipated,” “more meaningful” in 2027, and noticeable into the tens of billions in 2028. Separately, the SpaceX merger question was fenced off at the outset — IR announced no merger questions would be taken, and Musk deferred to his public statements from the previous day.
Conclusion
The quarter reads as a flywheel quarter: vehicle demand, the FSD subscription base and robotaxi miles all accelerated together, and automotive margins expanded into that acceleration rather than being bought with price cuts. The genuine negatives are structural rather than cyclical — Optimus timing, TeraFab’s capital demands of a scale no automaker has ever carried, and one-off-driven energy margin optics — but none of them contradicts the demand evidence in hand. What the quarter changes for the investment view is the locus of risk: from “does demand exist?” to “can Tesla convert an avalanche of demand into a fleet, a robot and a fab on schedule?” The next one or two prints will show whether the milestone ladder in CyberCab, Optimus and EU approval moves as fast as the miles curve did this quarter.
Artifacts and Follow-Ups
Key Artifacts — highest priority:
- EU-wide FSD approval. Matters because it gates the European fleet strategy; baseline commitment is approval across many EU countries already granted, with Musk expecting full EU approval “within a few months” and expressing surprise if it slips past Q1. Next print confirms or refutes.
- Robotaxi network cadence. Matters because the growth thesis is mile-driven; baseline commitments are Dallas launching next month, 6 cities expanding, 24/7 operation plus highway/high-speed roads by end of 2026, and first customer vehicles joining the robotaxi fleet around Q2 2027.
- Margin durability. Matters because the Q3 margin print mixes tailwind and one-off; baseline is the ex-credits automotive margin set in the Executive Summary alongside the one-off-led energy margin (see P&L table) — next quarter reveals the underlying run-rate as mix normalizes and commodity/rate headwinds persist.
- Optimus milestone ladder. Matters because the biggest-value thesis is the least-proven asset; baselines are the video in roughly a month, AI5 integration around April 2027, the 20,000-unit weekly annualized target and external deliveries a year out (see Lowlights).
- FSD metric disclosures. Matters because recurring-revenue visibility is now the anchor claim; baseline is management-confirmed analyst math of roughly $1 billion of ARR, with promised disclosures of revenue, churn and other FSD metrics as the base scales.
Secondary Tracking Items:
- Megapod / distributed-compute product definition, with visibility expected around end of Q1 into Q2.
- Capacity utilization of the CyberCab line versus its listed rate, and the race between mile growth and production.
- V15 release scope: the robotaxi software stack reaching customer cars plus navigation/parking takeover.
- 2027 Megapack order book fully closing (currently nearly sold out) and Megapack 3 ramp milestones.
- Berlin factory expansion permits and Reno Semi factory build-out progress.
- China FSD approval progress after the safety-report publication.
- Any SpaceX merger developments given the pre-empted topic on this call.