According to Zhitong Finance APP, Tesla (TSLA.US) released its third-quarter financial results and held an earnings call. The financial report showed that Tesla’s third-quarter revenue was $28.05 billion, flat year-over-year and slightly below the market expectation of $28.13 billion. Adjusted earnings per share came in at $0.48, higher than the market expectation of $0.44; adjusted EBITDA was $3.9 billion, exceeding the market expectation of $3.66 billion. Free cash flow reached $1.53 billion, surpassing market expectations.
Management characterized this quarterly report as “strong demand and accelerated investment,” but the core message is an overall delay in timelines: FSD is explicitly identified as the primary driver of current demand, with over 75% of deliveries in North America equipped with FSD and more than 1.7 million paying customers globally; Robotaxi is rapidly expanding in Austin and will enter Dallas next month, but its financial contribution will not become significant until after Q2 2027; Optimus production lines have begun early manufacturing, with external customer deliveries expected by the end of next year, and achieving a scale of 20,000 units per week (an annualized million units) is “still at least a year away”; Terafab equipment costs hundreds of billions of dollars, with long-term total costs exceeding $1 trillion; the first test chip is slated for May–June next year, with mass production not expected until late 2028. Full-year capital expenditure guidance has been lowered to approximately $20 billion, but the company states this reflects “efficient capital deployment” and that long-term expansion will continue, with up to $30 billion in borrowing capacity already secured; the company declined to provide further comment on merger-related matters.
CEO Elon Musk, CFO Vaibhav Taneja, Vice President of Autopilot Ashok Elluswamy, Vice President of Vehicle Engineering Lars
Moravy, Konstantinos Laskaris, and Investor Relations Head Travis Axelrod participated in the call. Below are the details of the meeting.
I. Opening Remarks by CEO Elon Musk
Q3 performance was strong, with demand exceeding supply.
Musk described the quarter as "very strong," with demand growing rapidly and significantly outpacing production; the company is substantially increasing capacity. FSD is the primary driver of demand, and "there is no real competitor to FSD in any car currently on the market." The company is working to secure FSD approvals in more markets—multiple European countries have already approved it, and it hopes to obtain EU-wide approval within a few months, adding, "I would be surprised if it extends beyond Q1."
Semi and Roadster. Tesla production to begin in Q3.
Semi truck and demonstrated the new Roadster, which Musk called "one of the coolest product demos ever," a result of collaboration between Tesla and SpaceX teams.
Production target: approximately 5 million vehicles per year by 2030. Musk said the company is laying the groundwork for increased manufacturing capacity, “I believe we have the opportunity to achieve this goal.”
Energy: Megapack nearly sold out by 2027.
Energy demand is extremely strong; "we have nearly sold out all our Megapacks for 2027," and we are accelerating production as much as possible, with significantly higher gross margins expected.
Robotaxi: Unsupervised mileage experiencing exponential growth.
Musk said the fully autonomous mileage for Robotaxi is expanding "exponentially," with week-over-week growth that "can hardly be overstated," and added that "no company in the world can match our scale in autonomous driving." Since the Austin Cybercab fleet launched about a month ago, it has grown approximately tenfold, and the weekly chart of cumulative mileage "looks like a rocket launching into orbit."
Optimus: Early production, with a gentle slope in the initial stage of the S-curve.
The Fremont high-volume production line has begun "early production." Musk emphasized that Optimus growth follows an S-curve, but the initial phase has a shallow slope and is bumpy because "10,000 things must be done correctly simultaneously, and production speed depends on the slowest and least likely of those 10,000 tasks." Most of the "1 million units per year" production capacity is now in place at Fremont, but reaching an annual rate of 1 million units, or 20,000 units per week, will take "at least another year, possibly longer," primarily because the design is still evolving. He expects issues with AI, electromechanical design, reliability, and lifespan to be resolved "around the end of next year," at which point deliveries of Optimus to companies outside Tesla will begin. Initial customers will be industrial applications—dangerous, repetitive tasks that people are unwilling to perform and that require 24/7 operation.
Terafab: Equipment commissioning in the first half of next year, with the first test chip in May–June.
Austin’s wafer fab progress is advancing rapidly, with equipment commissioning set to begin in the first half of next year, aiming to produce its first test chip in May or June—“just to see if we can actually make a chip.” The goal is to achieve mass production of some useful chip by Q3 2028. Musk called Terafab “the most valuable building in the world,” with economic value ranking among the top 20 nations, and identified it as the critical bottleneck solution for scaling businesses like Optimus (memory, packaging, logic chips, etc.).
Regarding the potential merger with SpaceX, I said everything I needed to say yesterday—dancing the tap dance on a minefield. That’s it.
II. Statement by CFO Vaibhav Taneja
Delivery: Over 480,000 units for two consecutive quarters, the highest backlog in four years.
Q3 marked the first time in history that two consecutive quarters exceeded 480,000 deliveries. Backlog at the end of the quarter reached its highest level in four years, with deliveries primarily constrained by production capacity rather than demand. The company is focused on increasing output across all factories.
The Model Y has stood out. Preliminary data shows that the Model Y is the best-selling BEV in Europe, the best-selling passenger vehicle in South Korea, and has set delivery records in markets such as Australia, Japan, and Taiwan.
FSD: North American delivery penetration exceeds 75%, with over 1.7 million paying customers worldwide.
In Q3, over 75% of deliveries in North America included a paid FSD subscription at delivery, an increase of approximately 20 percentage points compared to Q2. Globally, paid FSD customers exceed 1.7 million, with for the first time 50% being subscription-based. Due to strong FSD performance in the U.S., the company is implementing additional measures in Europe, offering promotions on certain vehicles to expand its user base prior to full FSD approval.
Automotive gross margin (excluding credits): 16.3% → 18.1%.
The sequential improvement was primarily driven by an increased delivery mix in the Asia-Pacific region. Rising commodity prices and interest rate changes continued to present cost headwinds, which the company has partially offset by adjusting its vehicle financing programs.
Energy: Deployed 13.7 GWh, gross margin increased from 20.4% to 33.1%.
This was the second-best deployment quarter in history. The sequential improvement in gross margin primarily resulted from tariff refunds recognized this quarter and the reversal of warranty provisions following compensation agreements with suppliers regarding certain cell-related costs. Demand for data centers remains extremely strong, and the company is actively expanding production, with further improvements in ASP and gross margin expected.
Services and Other: Gross profit margin increased from 14.1% to 15.9%, a record high.
Driven by multiple businesses including ultra-fast charging and insurance. The Robotaxi business (classified under services and others) continues to grow with the launch of Cybercab in Austin, achieving an 8-fold increase in cumulative unsupervised miles this quarter and maintaining double-digit weekly growth.
Fees and net profit. The company is in a major investment cycle, and operating expenses increased sequentially, primarily due to SI R&D (depreciation of new computing power and chip development) as well as Optimus and Megapack.
3. The ramp-up costs for pre-production of new products such as Semi. This trend will continue until these products are commercialized. Net profit was impacted by a $142 million decrease in the fair value of SpaceX shares, but this was offset by approximately $290 million in gains from the fair value appreciation of Bitcoin.
Capital expenditures: Full-year guidance lowered to approximately $20 billion.
The reason is "efficient capital deployment," with the plan primarily supported by operating cash flow. Free cash flow at the end of the quarter exceeded $15 billion, and cash and investment balances exceeded $45 billion. This quarter, a revolving credit facility of up to $30 billion was secured to support anticipated capital expenditure growth; no new draws are planned for this year. Tariff refund funds are being reinvested into U.S. production capacity expansion. Capital expenditures will continue to grow over the coming years to increase the Robotaxi fleet, expand Optimus production, install solar manufacturing capacity, acquire SI computing infrastructure, and build semiconductor wafer fabs.
III. Q&A Session
1. How can Cybercab expand beyond Austin? What is the regulatory pathway under NHTSA review?
Ashok Elluswamy: Cybercab launches this quarter, primarily utilizing Model Y/Model 3.
Learning from data on other platforms doesn’t require a large fleet of vehicles. Austin is just the starting point—“there’s no fundamental reason to limit ourselves to Austin.” Next month, we’ll enter Dallas; the same software can be scaled to many other cities, alongside Model 3
Together with Y Robotaxi.
2. Is the Optimus training data the main bottleneck? Crowdsourced data vs. internal Optimus Academy?
Musk:
Crowdsourcing isn't the issue. There's still a lot of work to be done on the AI architecture. To provide training data for Optimus, approximately 10,000 to 20,000 robots need to be manufactured first, allowing them to engage in self-play, and then Optimus must be built.
The Academy enables robots to practice various tasks, creating a data flywheel. Unlike cars, Optimus doesn’t have a large fleet to collect data and must bootstrap itself by manufacturing robots.
Ashok added:
Cars have inherent data, but Optimus lacks an exoskeleton that allows humans to "wear" it like a puppet. Optimus must first learn by observing humans, then practice on its own. The company is building a large-scale reinforcement learning loop: starting with human data for initial training, then deploying it on Optimus to conduct extensive self-play in both simulation and the real world.
Musk:
The Optimus video may be released in one or two months, focusing on demonstrating non-preprogrammed tasks rather than "canned demos." Another key milestone is integrating the AI5 chip into Optimus around April next year, offering an order-of-magnitude increase in computing power over the AI4 chip, which will greatly benefit Optimus, as humanoid robots have far greater degrees of freedom than cars.
3. When will consumers be able to use unsupervised FSD?
Ashok: Cybercab and Model 3
The software for Y is the same; version V15 will release the same software to customers. An additional requirement is that customer vehicles must be able to connect to Tesla’s geofencing and fleet management. The first customer vehicles are expected to join the Robotaxi fleet around Q2 next year.
4. When will Cybercab be available for consumers to purchase?
Lars Moravy:
Plan to deliver Cybercabs to customers by the end of 2026, but currently all vehicles produced on the Texas production line are entering Tesla’s own Robotaxi fleet. As Tesla grows, it will primarily deliver fully autonomous vehicles to customers.
Designed autonomous vehicles—Cybercab is the first. This year, we’ll start small, learning about delivery, ownership, and fleet management, then scale gradually.
Musk added:
Cybercab will primarily be offered as "Transport as a Service," with the vast majority of Cybercabs being Tesla-owned Transport-as-a-Service vehicles. The model you can own is the Model 3.
3/Y, etc.
5. Robotaxi unit economics: Cost per paid mile and Optimus unit cost?
Vaibhav:
Utilization rates in Robotaxi's primary operating regions are improving, with weekly paid orders and miles driven both on the rise. While the target of $0.30 per mile has not yet been reached, a clear path forward is evident, with scaling operations being the key focus. Cybercab is currently only available in Austin, with plans to expand to other cities by year-end. As the fleet grows, the company is gaining operational experience in fleet management, pick-up and drop-off logistics, and cleaning, leading to gradual cost optimizations.
Optimus cost:
Internal tracking of design maturity is progressing well, but it is still too early to say we’ve reached our cost targets, as the design is still undergoing significant iteration. From a cost perspective, we are moving in the direction of approaching Cybercab’s final cost.
6. FSD Map Issue: Why not use your own map?
Ashok:
Agreement should not rely on external maps. The company is building its own map, but not in the way of OpenStreetMap—it’s a completely new AI-driven mapping and navigation architecture that will integrate voice, learn individual and fleet preferences, and channel collective intelligence from all vehicles into the map and navigation system. Version V15 will first implement this technology for trip origins and destinations, with full takeover of all navigation functions, parking, and destination handling expected over the coming months.
7. Current factory capacity and new factory plans?
Lars:
This quarter, the factory operated at a record pace, but capacity was constrained by the slowest process or component. Over the past year and a half, strategic global investments have been made to increase capacity at existing facilities, and results are now being seen this quarter, with continued progress expected over the next several quarters. Additionally, construction is underway at Giga Texas.
Ramp Cybercab has applied for approval to expand its Berlin facility, and the new Semi factory in Reno has been launched. The company is indeed building new factories, with a target of reaching 5 million vehicles by 2030.
8. FSD Competition and Pricing: Will other automakers launching L2+ products create pricing pressure?
Ashok:
It's easy to create a flashy demo for the first 90%, but the long-tail problems are extremely difficult, especially achieving safety levels of millions of miles without incidents. As unsupervised FSD rolls out to a broader customer fleet, "I believe no one will be able to match it for years."
Vaibhav: In the U.S. showroom, customers now ask, "Can I try FSD?" instead of asking about the Model 3.
3/Y or Cybertruck. FSD has become a key driver, restoring mobility freedom to individuals who previously faced difficulties driving.
Regarding China:
Tesla’s technological advantage over Chinese OEMs still exists today. The company is working to secure broader approval for FSD in China, with some models already capable of using FSD. Competing CEOs who tested FSD in the U.S. were astonished. Tesla is actively collaborating with Chinese regulators and has published FSD safety reports and unsupervised Robotaxi safety reports.
9. Robotaxi revenue and profitability timeline? Cybercab production ramp-up curve?
Vaibhav: From an exponential math perspective, Robotaxis will become a significant contributor to revenue and profitability by 2027.
Musk: "This quarter won't change the situation. It will likely become apparent starting in Q2, and turn into a major event in Q3 or Q4."
Regarding the Cybercab production line:
The product designed by the engineering team is highly scalable from a manufacturing perspective; capacity is not a major challenge. The key is securing regulatory approval to deploy the vehicles as soon as they are produced. Cybercab has 50% fewer components than any other vehicle, and actual production line capacity far exceeds the stated 125,000 units per year—it is currently being produced at the pace demanded by Robotaxi service needs.
10. Energy Business: What roles do distributed generation and distributed computing play? What is the Megapack product?
Musk: It's still very early, and the full picture isn't clear yet—we'll likely have a better sense around Q1 or Q2 next year. The potential is huge, but it's hard to say just how large at this point.
Is the bottleneck for robotaxi growth regulation or software?
Ashok:
Software challenges remain. Robotaxi service hours are being extended from 6 a.m. to 9 p.m., then to 10 p.m., and soon to 2 a.m., gradually increasing availability driven by AI software capabilities. The operational principle is zero safety incidents—even the slightest risk is not accepted. For the remainder of this year, 24/7 operations and higher-speed road access will be unlocked, both essential conditions for scaling. The company is expanding both in breadth (six cities) and depth (dense fleets in Austin).
Advancing further, it will expand to more weather conditions, day and night scenarios, highways, and high-speed situations, with the goal of completion by the end of this year, at which point broader urban coverage will be observed. It is not purely a regulatory issue but technology-driven; however, it is expected to quickly become a regulatory issue in Q1 and Q2.
12. How much capital does Terafab need? What are the milestones?
Musk:
The capital scale is very large, with equipment costs amounting to tens of billions of dollars; the allocation ratio between Tesla and SpaceX is still being determined. By early 2028, equipment capital will not be less than tens of billions of dollars, and the long-term total cost may exceed $100 billion.
Vaibhav:
Tesla's first semiconductor fab has placed the majority of its initial equipment orders. Mid-next year is the milestone for the fab's research phase, with full-scale production expected by the end of 2028—fast by semiconductor fab standards.
13. What is the biggest design issue with Optimus?
Konstantinos Laskaris:
The hand is one of the most challenging hardware domains. While most electro-mechanical systems have high reliability, the hand is more complex than actuator systems for locomotion from an electro-mechanical perspective, and certain specialized gears are still under development for durability and reliability.
Musk:
The main challenge lies in the precise design of the hands and fingers. Humans can self-repair; robots cannot. Fingers must be soft, with numerous sensors embedded in the skin—creating a hand that is durable, capable of grasping delicate objects, and yet does not damage the sensors is extremely difficult. Additionally, integrating the AI5 chip represents another major leap, increasing computational power by an order of magnitude. When dexterity in electromechanics, AI software, and AI chips are multiplied together, Optimus’s value will grow exponentially: “At some point in the future, the number of humanoid robots on Earth will exceed the number of humans.”
14. Is the pricing of robotaxis below that of competitors a long-term strategy?
Vaibhav: Unlike ride-hailing services with drivers, Tesla operates without drivers and manufactures its own vehicles, resulting in a different cost structure. The goal is to "take over all transportation."
Musk:
Unless blocked by governments (protecting domestic industries is possible), Tesla will complete the majority of passenger miles on Earth in any country that allows free operation. "You can come back and check this prediction—I'm confident about it."
15. Can the unsupervised mileage weekly growth rate sustain double digits? What are the growth drivers?
Ashok:
Expansion will continue indefinitely into the foreseeable future, with no obstacles in sight; the clear path is to scale to millions of vehicles over the coming months. This is the result of technological investments made over the past several years and the V15 stack, which is now delivering returns from foundational work such as end-to-end neural networks and learning from human drivers. The same technology should be able to scale to millions of vehicles in all regions where operations are permitted.
16. Is FSD ARR approaching $1 billion? What is the renewal rate?
Vaibhav:
The numbers estimated by analysts (approximately $1 billion in ARR) are accurate. As the business scales, we will begin disclosing more FSD metrics, not just revenue but other key indicators as well. We are not currently disclosing churn rates because we are offering free trials in many new countries. The 75% delivery adoption rate in North America was achieved even though every vehicle includes one month of free FSD subscription—yet users are still subscribing.
17. Can Optimus scale up production to begin training before the hand design is finalized?
Musk:
We are doing exactly that—building robots without perfect hands, while advancing Optimus training, supply chain, and production systems. The Fremont production line, supply chain, design engineering, and AI software are all progressing simultaneously “at the fastest pace possible for humans.” Deliveries of Optimus to customers are expected to begin around the end of next year. “I would be shocked if anyone could create a robot competitive with Optimus.”
Vaibhav: In addition to Fremont, Austin is also building an Optimus factory, with production capacity far exceeding that of Fremont.
18. Semiconductor equipment capacity and capital expenditure?
Musk: Currently, the impact is minimal. Capital expenditures are slightly lower than expected, but they will become significantly more pronounced starting next year, reaching hundreds of billions of dollars by 2028.