30초 요약
Tesla said third-quarter 2025 free cash flow reached about $4 billion, while tariff costs exceeded $400 million. Elon Musk said the company would expand vehicle production as it pursued unsupervised driving, and planned to use both Samsung and TSMC initially for its AI5 chip. Executives described milestones and unresolved engineering work for robotaxis and the Optimus humanoid robot; the timelines and performance figures were projections made during the October 2025 call.
- 1Tesla reported record cash flow alongside tariff pressureCFO Vaibhav Taneja said quarterly free cash flow was about $4 billion and tariff costs exceeded $400 million across automotive and energy.
- 2Tesla targeted removing robotaxi safety occupants in parts of AustinMusk projected the change by the end of 2025 and said expansion to other markets depended in part on regulatory approvals.
- 3Samsung and TSMC were both to focus initially on AI5Musk revised earlier comments, saying Tesla wanted ample chip supply for vehicles, robots and potentially data centers.
- 4Optimus manufacturing would continue to evolve after production beganMusk said Tesla intended to make rolling design changes and targeted a production line capable of an eventual million units a year.
핵심 정리
- Tesla reported about $4 billion in quarterly free cash flow and more than $400 million in tariff costs.
- Musk planned to source the AI5 chip initially from both Samsung and TSMC, with excess production potentially used in data centers.
- Tesla’s robotaxi expansion and Optimus production goals depended on technology and regulatory work still ahead.
챕터별로 다듬은 전문 읽기
Autonomy ambitions and third-quarter results
Musk describes Tesla’s autonomy and energy plans. CFO Vaibhav Taneja reports deliveries, margins, cash flow and tariff costs.
Investor relations head Travis Axelrod opens Tesla’s third-quarter 2025 webcast, introduces Elon Musk and CFO Vaibhav Taneja, and directs shareholders to the company’s update and annual meeting materials. The first nearly ten minutes of the video contain music before the call begins.
Musk describes Tesla’s strategy as bringing AI into the physical world. He predicts that software updates could eventually enable many existing cars to drive without supervision, and says he now feels confident expanding vehicle production as quickly as reasonably possible.
Musk says he had held back on expansion until he saw a path to unsupervised driving. Now, he says, Tesla intends to expand future production as fast as it reasonably can.
Musk says stationary batteries such as Powerwall and Megapack can help grids use existing generation more effectively by shifting energy from periods of low demand to periods of high demand. He argues that this could expand usable output without building additional power plants.
He argues that permits and long construction timelines make new power plants difficult to build, and that storage can improve the amount of energy a grid delivers over a year.
He calls the Optimus humanoid robot a difficult project with very large potential, arguing that Tesla combines AI, engineering and manufacturing capabilities. He says the company planned to show Optimus V3 in the first quarter and describes its broader mission as “sustainable abundance.”
Musk says a Tesla car is already a physical AI system, but building a humanoid robot is not an automatic extension of making cars. He points to electrical and mechanical engineering and manufacturing scale as capabilities Tesla can apply.
He discusses the planned Megapack 4, saying its design would incorporate equipment normally found in a substation and could output at around 35 kilovolts. The aim is to make deployment less dependent on separate substation construction.
Musk imagines self-driving and robotics improving access to goods and services, including medical care. He also says the company must ensure Optimus is safe. These are his long-term expectations rather than capabilities demonstrated during the call.
Taneja says the quarter set company records for total revenue, energy gross profit and margins, and free cash flow. He attributes stronger deliveries to all regions and to the updated Model Y range, including the long-wheelbase, performance and standard versions.
Taneja reports sequential delivery growth of 33% in Greater China, 29% in Asia-Pacific, 28% in North America and 25% in Europe, the Middle East and Africa. He links the increase to the Model Y refresh and additional variants.
Taneja says Tesla was operating robotaxi services in Austin and much of the Bay Area, and had expanded its Austin coverage three times. He says paid Full Self-Driving customers represented roughly 12% of the fleet, and the company was working with regulators on approvals in China and other regions.
Taneja describes supervised FSD uptake as still small relative to the vehicle fleet, while saying the experience could eventually increase vehicle demand. He notes the company is seeking regulatory approval in additional markets.
Automotive revenue rose 29% sequentially, Taneja says, while the automotive margin excluding regulatory credits improved from 15% to 15.4%. He attributes the margin improvement to material costs and better use of fixed capacity as volumes increased.
The energy business recorded deployments, gross profit and margins, but Taneja says competition and tariffs remain headwinds. Shanghai Megafactory production was supplying non-US demand, reducing tariff exposure. Across automotive and energy, he puts the quarter’s tariff impact above $400 million, roughly evenly split.
Taneja says services improved largely because of insurance and service centers. Expenses rose from restructuring connected to chip design, legal and shareholder-meeting costs, and AI-related employee awards. He says mark-to-market gains on Bitcoin holdings fell to $80 million from $284 million in the prior quarter.
Higher employee-related spending includes performance-based equity awards for staff working on AI, Taneja says. He expects that expense to keep rising. Changes in foreign exchange also affected other income in the quarter.
Free cash flow was approximately $4 billion and cash and investments exceeded $41 billion at quarter-end, Taneja says. He expects around $9 billion of capital expenditure in 2025 and a substantial increase in 2026 to support existing businesses and AI projects including Optimus.
Robotaxi, storage and Optimus questions
Executives discuss safety occupants, energy demand and the technical and supply-chain challenges of a humanoid robot.
The first shareholder question asks for robotaxi fleet, trip, safety and intervention metrics and the obstacles to unsupervised driving. Musk expects no safety occupant in parts of Austin by year-end 2025 and forecasts operation in eight to ten metropolitan areas, subject in part to regulatory approvals.
Musk says the rollout needs to be cautious because even one accident would attract worldwide attention. He mentions Nevada, Florida and Arizona among the markets Tesla expected to enter, while noting that regulatory approvals were involved.
An executive says the Austin fleet had covered more than 250,000 miles without a person in the driver’s seat, while Bay Area vehicles with a driver had passed one million miles. He reports six billion cumulative miles using customer-supervised FSD and says the team is working toward removing the in-car person in Austin.
On energy storage, an executive says Megapack and Powerwall demand is strong into the following year. He expects the Houston-built Megablock to ship in 2026 and says data centers and utilities are seeking storage for reliability and grid constraints. He also cites stronger US residential solar demand and a planned solar lease.
The energy executive also says Tesla began producing residential solar panels in Buffalo and expected customer shipments in the first quarter of 2026. He connects a rise in solar demand with policy changes and a new lease offering.
Asked about getting Optimus to market, Musk says the challenge is substantial. He describes robots walking in Tesla’s Palo Alto offices and guiding visitors to meeting rooms, but singles out a dexterous hand and forearm as especially difficult electromechanical problems.
Musk explains that the hand’s finger lengths, range of movement and forearm actuators all affect dexterity. In his assessment, the hand and forearm present a greater electromechanical challenge than the rest of the robot combined.
Musk says a useful general-purpose robot needs capable hands, real-world AI and production at vehicle-like scale. He argues that a supply chain for such a robot does not yet exist at the needed scale, requiring Tesla to manufacture deeply into the supply chain itself.
He says a startup would struggle to build a humanoid robot at the envisioned scale without extensive manufacturing technology. Tesla, he argues, must solve the hand, the software and the manufacturing problem together.
Musk returns to his desire for voting influence at Tesla. He says he does not want to build a large fleet of robots only to be removed from the company, and frames the compensation discussion as a question of retaining influence over the technology.
AI5 chips, hardware upgrades and the Semi
Musk explains the planned AI5 chip and dual manufacturing approach. Other executives discuss older driving hardware and Semi production.
An investor asks about Tesla’s Samsung chip agreement and whether the company can secure the chips needed for autonomy. Musk clarifies that Tesla plans to focus both Samsung and TSMC initially on AI5, revising his earlier public comments. Samsung already manufactures the AI4 computer, he says.
Musk says Tesla designed AI5 around its own hardware and software requirements. He claims it could outperform AI4 by a factor of 40 on some measures and says the design removes a conventional GPU and image signal processor. He emphasizes that the chip still must be manufactured at scale to prove the projections.
Musk says Tesla’s close knowledge of how its software uses the hardware lets the chip team remove functions it does not need. He describes removing legacy graphics and image-processing components in order to simplify the design.
The resulting chip design, he says, fits within a half-reticle area while leaving room for connections among memory, accelerators and processor cores. He says both manufacturing partners would produce the design in the United States.
The planned suppliers are TSMC in Arizona and Samsung in Texas. Musk says Tesla wants an oversupply of AI5 chips so that excess capacity can go to data centers if it is not needed in vehicles or robots. Tesla already combines AI4 and Nvidia hardware for training, he says, and is not about to replace Nvidia.
Musk emphasizes that having more AI5 chips than Tesla needs in cars and robots would not necessarily waste production: the company could use the surplus in its data centers alongside other computing hardware.
Musk argues that Tesla can simplify its chip because it designs for one customer, while Nvidia must accommodate many. He forecasts advantages in performance per watt and per dollar, while acknowledging that manufacturing results remain the test.
He compares chip interconnections to highways between different parts of a system. Supporting many possible workloads can require many routes, he says, whereas designing for Tesla’s own needs permits a simpler layout.
Asked about owners of older Hardware 3 vehicles, an executive says Tesla has offered some customers the ability to transfer FSD to a new vehicle and has not abandoned the earlier computer. Ashok Elluswamy adds that a lighter version of FSD 14 for Hardware 3 was planned for the second quarter of 2026.
The executive says early Hardware 3 customers remain important and promises a way to address them once Tesla solves autonomy. He adds that he uses FSD daily in a Hardware 3 vehicle.
On the Semi truck, an executive says the factory building is complete and equipment is being installed, with larger validation builds planned and volume expected in the second half of 2026. He sees long-distance rail as efficient but expects autonomous trucks to help with shorter or last-mile routes. Elluswamy says the current focus is passenger-car autonomy, with the technology later adaptable to the Semi.
The Semi team says validation trucks were already on the road, with equipment installation underway. The plan was for initial production builds early in 2026 and a larger ramp later in the year.
Vehicle expansion, AI strategy and robot production
Analysts ask about manufacturing capacity, Tesla and xAI models, robotaxi deployment and Optimus ramp-up.
An analyst asks how a production expansion fits Tesla’s existing capacity and whether it would sacrifice margins. Musk says capacity is not yet three million vehicles a year but suggests a three-million annualized rate might be reached within 24 months, depending on the supply chain.
Musk says any increase in production must occur alongside a vast supplier network, and that Tesla would also need to decide where to build additional factories beyond existing capacity.
He describes Cybercab, then targeted for production in the second quarter of 2026, as a vehicle without a steering wheel or pedals, designed around autonomous operation and low cost per mile. Unlike a high-performance passenger vehicle, it would prioritize a comfortable ride at more modest speeds.
An analyst asks whether prioritizing volume would hurt near-term profitability. Musk responds that he expects strong demand rather than a margin sacrifice if autonomous capability meets his expectations.
Musk says he does not expect expansion to sacrifice margins and predicts demand for cars that can drive without supervision. He contrasts that anticipated capability with drivers using phones today and claims automated driving improves safety. His remarks describe a future unsupervised system; the call also distinguishes customer FSD as supervised.
Musk expresses confidence that Tesla can make unsupervised driving safer than human driving. He describes plans for reasoning, a simulation environment for training and a larger driving model, saying these changes would run on AI4 before AI5. He also speculates that idle cars might one day form a distributed computing network.
Musk describes a Tesla simulator that generates realistic driving video for reinforcement learning. He says the team also plans to increase the model’s parameter count and add other AI improvements not included in version 14.1.
As a speculative example, Musk imagines tens of millions of parked cars contributing inference computing. At a hypothetical kilowatt per car, he says, a fleet of 100 million vehicles would represent substantial distributed capacity.
Asked how Tesla chooses which markets fit its capabilities, Musk says it built competencies over time, including stationary batteries, charging infrastructure, chip design and AI software. He likens Tesla to a collection of startups and argues that its manufacturing and robotics work can expand into new markets.
Musk cites the Supercharger network and the company’s move into stationary batteries as examples of capabilities Tesla developed. He says it built its chip and AI software teams from scratch rather than beginning with all the necessary skills.
Musk speculates that robots operating continuously could produce far more work per year and describes a future in which work might be optional. Taneja says the Optimus effort draws on vehicle engineering experience, including engineers who moved from drive units to actuators; Musk adds that many new engineers have also joined.
Taneja recalls an earlier discussion in which Musk described the car as a robot on wheels. Tesla engineers from the vehicle team subsequently worked on Optimus, he says; Musk also credits new hires, including recent graduates.
Musk says Optimus engineering and manufacturing reviews run in parallel so difficult-to-make components can be redesigned. He calls Optimus 2 hard to manufacture and expects the next version to improve both function and manufacturability. He describes the regular review schedule for Optimus and AI5.
Musk points to a public Optimus 2.5 demonstration at a film premiere as an example of the robot’s progress. He expects the next version to be substantially better, while reiterating that building it at scale remains difficult.
An analyst asks how Tesla’s AI work relates to xAI. Musk says Grok is a much larger model aimed at general intelligence, while Tesla’s driving models are built for much smaller onboard computers. He says Grok can complement Tesla products in voice interaction, including for Optimus.
Musk describes xAI as pursuing a large general-purpose model, whereas Tesla’s model must fit into a vehicle computer and respond in real time. He says the two approaches can complement one another through features such as in-car voice interaction.
Asked whether new robotaxi markets require safety staff for regulatory or technical reasons, Musk says Tesla would use a cautious introduction even without a requirement. An attendant or driver could help identify local challenges before the company considers removal, he says, suggesting roughly three months as an example.
Musk says a new city may contain unusual intersections or other rare situations that justify having someone in the vehicle during an initial deployment. He presents the suggested trial period as a cautious operating approach.
An analyst asks how FSD 14 differs from the robotaxi software and whether safety or comfort drives releases. Musk says safety comes first and advises most users to wait for version 14.2 as the team smooths comfort issues. Elluswamy says the two systems share most of their algorithms and architecture, with a few customer-specific features.
Musk says early releases of a new driving architecture may feel jerky even if safety has been prioritized. Elluswamy says customer cars have a few extra features, such as choices about parking, but share the core approach with the robotaxi fleet.
Musk and Elluswamy discuss adding reasoning, such as deciding where to drop off a passenger and park. Elluswamy says the challenge is fitting real-time decision-making into the AI4 computer. Musk argues that this constraint forces the team to make its model efficient and could benefit a later move to AI5.
The team discusses a car dropping a passenger at a store entrance before searching for a space. Elluswamy says running such decisions locally on the vehicle computer is harder than sending them to a remote server because the car must respond quickly.
An analyst asks how Tesla can start Optimus production while improving the hand and establishing a supply chain. Musk says the hardware design will continue to change after production begins. He expects a production-intent prototype in early 2026 and hopes to start a line targeted eventually at one million units a year toward the end of that year.
Musk says the production ramp would take time because it depends on thousands of components and the slowest bottleneck. He speculates about later generations and much larger volumes beyond the initial million-unit annualized target.
Shareholder vote and voting control
Executives urge shareholders to support board proposals. Musk explains why he wants more influence over Tesla’s future.
Taneja urges shareholders to support the compensation proposals and reelection of three directors at the November 2025 meeting. Musk says he wants voting influence in the mid-20% range, enough for a strong voice but not so much that he could not be removed. He criticizes the influence of proxy advisory firms over index-fund voting.
Musk says Tesla lacks the special voting shares some companies established before going public. He worries that index funds can rely too heavily on recommendations from proxy advisers ISS and Glass Lewis, potentially affecting control of the company.
Taneja says the proposed plan ties rewards to substantial shareholder returns and asks shareholders to vote on the compensation package and directors. Axelrod thanks participants and closes the webcast.
같은 채널의 다른 글
24:44EN
59:56EN
23:21EN
1:09:55EN
27:27EN
1:05:20EN