Compiled by Dolphin Research:$Tesla(TSLA.US) FY26 Q3 earnings call Trans
I. Core takeaways
1. Capex and financing
a. FY26 capex guidance was reset to approx. $20 bn, to be largely funded by operating cash flow. The company emphasized discipline on drawdowns.
b. It secured up to $30 bn in borrowing capacity this quarter to support capex growth, with no plans to draw within the calendar year. Tariff refunds will be reinvested in U.S. capacity expansion.
c. Capex will keep expanding over the next few years, driven by the Robotaxi fleet, Optimus capacity, solar manufacturing, AI compute infrastructure, and an in-house fab. Management framed these as multi-year growth pillars.
2. Opex and margin outlook
a. Opex will continue to rise QoQ as guided, mainly from AI compute depreciation, chip R&D, and pre-ramp costs for new products such as Optimus, Megapack 3, and Semi. This trend will persist until these products are commercialized.
b. Storage demand remains strong, with ASP and GPM expected to improve further. The mix and pricing environment are supportive.
3. Key financial metrics this quarter
a. Auto: ex-credits auto GPM rose from 16.3% to 18.1%, largely on higher APAC delivery mix. Commodities inflation and rates remain headwinds, partially offset by adjustments to buyer financing programs.
b. Energy storage: storage GPM increased from 20.4% to 33.1%, mainly due to tariff refunds recognized this quarter. Warranty reserves were reversed after supplier agreements on partial cell cost compensation.
c. Services & other: GPM improved from 14.1% to 15.9%, a record high. Supercharging and insurance both contributed.
d. Non-op items: SpaceX stake was marked down by $142 mn, offset by about $290 mn in Bitcoin fair value gains. Net impact was positive.
e. Cash: FCF exceeded $1.5 bn, with period-end cash and investments above $45 bn. Liquidity remains robust.
II. Call details
2.1 Key management remarks
1. Auto and FSD
a. Demand is clearly above capacity, and deliveries are capacity-constrained. Deliveries exceeded 480k for two consecutive quarters, with order backlog at a four-year high, and all plants ramping.
b. Initial data shows Model Y as the best-selling BEV in Europe and the best-selling passenger car in Korea. Sales hit new highs in Australia, Japan, and Taiwan.
c. FSD is the biggest demand driver: in North America, over 75% of deliveries subscribed to FSD at handover, about 20% higher vs. Q2. Global paid users surpassed 1.7 mn, with subscriptions reaching 50% for the first time.
d. Approvals have been obtained in multiple European countries, with a push to secure EU-wide approval within months. Management would be surprised if it slips past Q1 next year; discounts are being offered on select European models to grow fleet size ahead of approval.
e. Semi entered mass production from Q3, and a new Roadster demo was unveiled. The company targets approx. 5 mn vehicles of annual output by around 2030.
2. Robotaxi
a. Austin’s Cybercab fleet has scaled 10x since launching about a month ago. Cumulative unsupervised miles grew 8x this quarter, with weekly double-digit growth sustained throughout the quarter.
b. Austin learnings are being applied to roll out Cybercab in other markets, prioritizing network density and coverage. The long-term ambition is to cover all road transport, not just passenger rides.
3. Optimus
a. The large-scale Fremont line has begun early production, and most equipment for a 1 mn units per year line is in place. Ramps will be phased.
b. The early S-curve is shallow and bumpy, and reaching an annualized capacity of about 20k per week will take at least a year, possibly longer. Ongoing design changes are the main reason.
c. AI, mechatronics design, and reliability/lifespan are expected to be solved around end-2027, at which point deliveries to external customers will begin. The roadmap remains subject to execution.
4. Energy storage
a. Megapack is almost sold out for 2027, and expansion is in full swing. Data center demand is extremely strong.
b. Storage deployments reached 13.7 GWh this quarter, the second-highest ever. Deliveries of Megapack with in-house LFP cells have begun.
5. Terafab and other
a. Austin’s research fab is progressing fast: equipment debug starts in H1 2027, with first test chip tape-out targeted for May–Jun if all goes well. Volume-worthy chips are aimed for around Q3 2028.
b. Management called Terafab the most valuable building globally, a critical asset to resolve bottlenecks in memory, packaging, and logic, and to scale Optimus. Strategic significance was emphasized.
c. No comment on any potential merger with SpaceX. The company declined to engage.
2.2 Q&A
Q: How will Cybercab expand beyond Austin, and what is the regulatory path beyond Texas?
A: Dallas is next month, and the same software stack will scale to more cities; no specific regulatory path was discussed. Cybercab training primarily uses data from Model Y and Model 3, without needing a large proprietary fleet. Austin went first simply because the vehicles were there, not due to structural constraints.
Q: Should Optimus training data be crowdsourced, and is data the main bottleneck?
A: The bottleneck is not crowdsourcing but improving the AI architecture, and building 10–20k robots for self-practice. Unlike autos, Optimus lacks a massive fleet for data; it must first learn from human demonstrations, then scale reinforcement learning in sim and the real world via the Optimus Academy. Management said they are close to good results.
Non-scripted real capability videos will be released in 1–2 months, and the AI5 chip is slated for integration around Apr 2027, with an order-of-magnitude jump vs. AI4. Initial customers will be industrial tasks that are hazardous, repetitive, or 24/7, with household use to follow as volume scales.
Q: When will the unsupervised FSD version reach customer vehicles?
A: Around Q2 2027, the first customer vehicles can join the Robotaxi fleet. Customer Model Y and Cybercab share the same software and will receive V15, with the extra condition of being within the company’s geofenced fleet management.
Q: When will Cybercab be available for customers to purchase?
A: Target is to deliver the first customer batch by end-2026, but the primary form will remain company-operated mobility services. Output from the Texas line currently all goes to the proprietary fleet, with small-batch deliveries to build ownership and fleet management experience. Model 3/Y and Roadster are the main models for individual ownership.
Q: What is Robotaxi cost per mile?
A: It has not yet reached the $0.30 per mile target, but there is a clear path. Utilization in major operating zones is improving, and Cybercab will launch in other cities before year-end; pickup, cleaning, and other fleet ops costs are being optimized.
Q: What is unit cost for Optimus?
A: Too early to conclude, as design is still iterating. Internal cost tracking is on track, trending toward the eventual cost level of Cybercab.
Q: Why not use an in-house map for FSD issues such as speed limits, parking, and routing?
A: A brand-new AI-native mapping and navigation stack is being built. It will integrate voice and learn personal and fleet preferences; trip origin/destination will roll out with V15 first, then navigation, parking, and destination handling over the coming months.
Q: What is current plant capacity, and will you build new plants?
A: Plants delivered record output this quarter, and new plants are under construction, with a 2030 target of 5 mn units per year. Investments from the past year-plus are bearing fruit and will continue in coming quarters; Giga Texas is ramping Cybercab, Berlin expansion permits have been filed, and the new Semi plant in Reno has just started production.
Q: If other OEMs partner with NVIDIA or Qualcomm on L2+ products, will FSD pricing be pressured?
A: Management believes competitors will struggle to catch up and did not directly address pricing. It is easy to demo the first 90%, but the long tail is extremely long, and achieving millions of miles with zero fatalities is very hard. U.S. store customers now proactively request FSD test drives rather than specific models, and FSD has become a key purchase driver.
Q: With intelligent driving now standard in China, is pricing logic different?
A: Judgment is similar, and Chinese OEMs currently do not have equivalent technology. The company is actively pursuing broader FSD approvals in China, and some models are already enabled; FSD and unsupervised Robotaxi safety reports have been published, and management believes the data will be hard to refute.
Q: When will Robotaxi revenue and profit show up?
A: No material impact this season; visibility starts around Q2 2027 and becomes meaningful in Q3–Q4. On an exponential growth path, 2027 will be a significant part of revenue and profit; the required miles to turn gross profit positive were not disclosed.
Q: How will the Cybercab line ramp, given its nameplate capacity of 125k+ units per year?
A: Manufacturing expansion is not the bottleneck; securing regulatory approvals is key to investing as much as we can produce. Cybercab has 50% fewer parts than any prior model; the nameplate is just the current validated rate, while the actual line capability is far higher, and production is scheduled to match Robotaxi demand.
Q: How do you view the company’s role in distributed power and distributed compute, and how big is the Megapod opportunity?
A: It is still early, unquantified, and likely clearer in Q1–Q2 2027. Management only noted the potential is very large, but it is hard to size today.
Q: Is the bottleneck for Robotaxi expansion regulatory or software?
A: It is still tech-driven for now, with a shift toward regulatory constraints expected in Q1–Q2 2027. Service hours are being extended from 6:00–21:00 to 23:00 and soon to 2:00, depending on AI capability, with zero safety incidents as a principle.
The service now covers six cities, with high-density operations in Austin. By year-end, the plan is to unlock 24/7 operations, highways, and more weather and day/night conditions, after which city coverage will expand significantly.
Q: How much capital will Terafab require, and what are the milestones?
A: Equipment investment will be at least tens of billions of dollars by early 2028, and could exceed $100 bn over the long term. Cost sharing with SpaceX is still being discussed; most equipment for the research fab has been ordered, and mid-2027 will focus on learning the tools. True volume production is expected by end-2028, which is extremely fast by fab standards.
Q: Where are the key design changes for Optimus?
A: Mainly the hands, especially fine finger design. Other mechatronic systems have largely converged, with power and reliability of certain custom parts still under development. Hands must be both durable and soft, sensor-rich yet hard to damage; another upgrade is the move to AI5.
Management believes the value of a robot equals the product of mechatronic dexterity, AI software, and AI chips. All three are improving exponentially, and robots will ultimately be able to build robots.
Q: Is Robotaxi pricing below ride-hailing and peers a long-term share-grab strategy?
A: The cost structure is different, and the goal is to take over all forms of road transport. Ride-hailing has drivers, while the company has none and builds its own vehicles; unless blocked by governments to protect domestic industries, management expects Tesla to carry the majority of passenger miles in countries that allow free operation.
Q: Can unsupervised miles maintain weekly double-digit growth, and what drives it? Will Cybercab become the core of the fleet?
A: Expansion should be sustainable, with a clear path to scale to millions of vehicles in the coming months. Growth stems from years of tech investment and the V15 architecture, as end-to-end networks and learning from human driving are bearing fruit; the fleet mix question was not addressed.
Q: Will FSD double lifetime per-vehicle profit, and will you disclose ARR?
A: The CFO said the questioner’s math checks out, and FSD revenue and other key metrics will be disclosed at scale. The questioner cited per-vehicle sales GP of about $7,000, FSD at over $1,000 per year, and ARR near $1 bn.
Q: What is FSD subscription retention?
A: Churn is not being disclosed for now, but will be in future. New countries generally offer free trials, distorting data; the 75%+ sign-up rate in North America occurs even with a one-month free subscription bundled with every car.
Q: Can you ramp Optimus for training before finalizing the hands?
A: That is exactly the plan: use robots with non-final hands to advance training while standing up the supply chain and lines. Design, AI software, and capacity are progressing in parallel, with deliveries to customers targeted around end-2027; beyond Fremont, a larger-capacity robot factory has broken ground in Austin.
Q: Can Terafab be prebuilt more aggressively, and what is the capex impact?
A: Minimal impact for now, and capex this year is slightly below expectations. The impact becomes more pronounced from 2027, reaching tens of billions in 2028.
<End of transcript>
Risk disclosure and statements:Dolphin Research Disclaimer and General Disclosure
