After the market closed on Wednesday Eastern Time, electric vehicle giant Tesla... Tesla released its Q3 2025 financial report. Despite record-high revenue driven by a surge in electric vehicle purchases by US consumers, earnings fell short of Wall Street expectations.
The profit decline reflects both a price reduction in Tesla's electric vehicles and a roughly 50% increase in operating costs, which the company stated was related to artificial intelligence. (AI) and "other R&D projects".
After the earnings report was released, Tesla During the earnings call, CEO Elon Musk and others released significant information regarding capacity expansion, core new products, and technology development. Musk stated that Tesla and its future strategy are at a critical turning point as they apply artificial intelligence to the real world. "I think it's important to emphasize that Tesla is indeed a leader in real-world AI."
However, Musk has become more cautious about the progress of new products and businesses such as Robotaxi and Optimus. For example, in the second quarter earnings call, Musk set an ambitious goal of covering half of the US population with Robotaxi service by the end of the year, but three months later he changed his statement to expanding the service to 8 to 10 cities by the end of the year; regarding Optimus, Musk postponed the start of its production line from the beginning of next year to the end of next year.
Overall, the market is not buying into Musk's grand vision for AI, and the downward revision of his targets for Robotaxi and Optimus seems to indicate that these new businesses are not progressing as smoothly as previously anticipated. Consequently, the company's stock price remained weak in after-hours trading, falling nearly 5% at one point.
Key Takeaways from Tesla's Earnings Conference
Business Strategy :
AI drives breakthroughs across multiple fields, anchoring "sustainable wealth." The conference clarified that Tesla's core strategy is to deeply integrate AI into the real world, covering automobiles, energy, and robotics. The three main components share the ultimate goal of achieving "sustainable prosperity," namely, eliminating poverty and making top-tier healthcare accessible through autonomous driving and Optimus. Musk emphasized that Tesla is a "real-world AI leader," boasting the highest AI intelligence density in the industry for its vehicles, and that its AI technology can be reused across different scenarios (such as using automotive AI to power Optimus), creating a unique technological synergy advantage.
Automotive business:
Regarding capacity expansion driven by both production capacity and autonomous driving, the company plans to achieve an annualized production capacity of 3 million vehicles within 24 months (or less). The core increase will come from the Cybercab, which will begin mass production in the second quarter of next year—a steering wheel/pedal-less vehicle optimized for autonomous driving, aiming for the lowest operating cost per mile. Strong demand is expected without sacrificing profit margins. Significant progress has been made in autonomous driving, with FSD 14.1 already released. By the end of the year, "reasoning capabilities" (such as autonomously identifying parking spaces) will be added. Austin has already achieved over 250,000 miles of driverless Robotaxi driving. By the end of this year, the company plans to enter 8-10 new markets, initially deploying safety drivers in each new market to assess risks before removing them.
Optimus humanoid robot :
With technology approaching "human-like" design and a clear mass production roadmap, the current Optimus 2.5 can perform kung fu moves, achieving such a high degree of anthropomorphism that passersby might mistake it for a human, and it can guide visitors 24/7 at the headquarters campus. A core breakthrough is achieved in "hand dexterity," with the new hand described as an "engineering masterpiece." The mass production plan is clear: the Optimus V3 mass production prototype will be showcased in the first quarter of next year, with a production line aiming for an annualized output of 1 million units by the end of the year, and a long-term goal of reaching a scale of hundreds of millions. The team leverages its experience in automotive electromechanical engineering and manufacturing, while simultaneously adding new engineers to accelerate R&D.
Core technologies:
AI chip performance leaps forward, energy business continues to lead the chip industry; AI5 performance is 40 times better than AI4, eliminating redundant functions to achieve high intelligence density, manufactured by TSMC. Samsung's US factories manufacture the products, and excess capacity can be used for data centers. In the energy business, demand for Megapack and Powerwall is strong, and the Shanghai factory is easing tariff pressure in non-US markets; US residential solar power... Due to surging policy demand, new leasing products will be launched in the first quarter of next year, and the Buffalo factory will ship new high-efficiency solar panels.
Operations and Finance :
Key metrics hit record highs, with increased R&D investment. Q3 deliveries, total revenue, energy business gross profit, and free cash flow (approximately $4 billion) all reached record highs. Automotive revenue increased by 29% quarter-over-quarter, and the profit margin, excluding credits, rose slightly to 15.4%. Capital expenditures are projected to reach approximately $9 billion in 2025, with a significant increase expected in 2026, primarily focused on AI and Optimus. R&D expenses continued to rise due to AI employee stock incentives. Despite facing tariffs (over $400 million) and competitive pressure, the company ensured business resilience through supply chain optimization and cost control.
The following is a transcript of the earnings call (translated with AI assistance).
Attendees:
CEO Elon Musk
Chief Financial Officer – Vaibhav Taneja
Ashok Elluswamy, Vice President of Vehicle Software
Lars Moravy, Vice President of Vehicle Engineering
Vice President of Energy — Mike Snyder
Travis Axelrod, Vice President of Investor Relations
Travis Axelrod, Vice President of Investor Relations
Good afternoon, everyone, and welcome to Tesla's Q3 2025 earnings webinar. I'm Travis Axelrod, Head of Investor Relations. Joining me today are Elon Musk, Vayibaf Tania, and several other executives. We announced our Q3 results at approximately 3 p.m. Central Time in the latest presentation available at the same link as this webinar.
In this meeting, we will discuss the company's business prospects and make forward-looking statements. These comments are based on our forecasts and expectations as of today. Due to various risks and uncertainties (including our recent filing with the U.S. Securities and Exchange Commission) , (As mentioned in the filings with the Transaction Commission, there are risks and uncertainties), and actual events or outcomes may differ materially from expectations. We urge shareholders to read our final proxy statement, which contains important information regarding matters to be voted on at the 2025 Annual General Meeting.
Travis Axelrod, Vice President of Investor Relations
During today's Q&A session, please allow each person to ask only one question and one follow-up question. Please click the "Raise Hand" button to join the question queue. Before we begin the Q&A session, Elon will give some opening remarks. Elon, please go ahead.
CEO – Elon Musk
Thank you. Tesla and its future strategy are at a critical turning point as we apply artificial intelligence (AI) to the real world. I think it's necessary to emphasize that Tesla is truly a leader in real-world AI. No company can match us in terms of real-world AI applications. Overall, I have a fairly deep understanding of the AI field. I believe that Tesla has the highest intelligence density of all AI applications in the automotive sector, and this advantage will continue to grow. We are only just beginning to roll out Full Self-Driving (FSD) and robotaxi on a large scale and fundamentally change the nature of transportation. I think people haven't fully grasped the extent of this transformation. Frankly, its impact will be like a shockwave. Related vehicles are already everywhere.
As you know, we currently have millions of cars on the road, all of which can be upgraded to fully autonomous vehicles with just one software update. In fact, we produce around two million cars annually. And, given that we now have a clear prospect of achieving fully autonomous driving—more precisely, unsupervised fully autonomous driving—I am confident in scaling up Tesla's production capacity. Our goal is to scale up future production as quickly as possible. Before we had a clear prospect of achieving unsupervised fully autonomous driving, I was cautious about expanding production. Now that we have that goal, it is wise to scale up production as quickly as possible at a reasonable pace.
At the same time, we use batteries Energy storage Our business is also having a significant impact on the energy sector. With Powerwall (residential energy storage batteries ) and especially Megapack (large-scale commercial energy storage batteries ), we are significantly enhancing the energy production capacity of the power grid.
Let me elaborate on this. Take the total energy supply capacity of the United States as an example. The current continuous power supply capacity in the US is approximately 1 terawatt. However, due to the significant difference in electricity demand between day and night, the average electricity consumption over a 24-hour period is only 0.5 terawatts. If we use batteries to buffer and store energy, we can effectively double the energy output of the United States without building any new power plants. Building new power plants is extremely difficult, not only time-consuming but also requiring complex approval procedures, and the industry itself is not adept at rapidly advancing projects. We believe that Tesla battery packs have the potential to significantly increase the annual energy output of both the US power grid and other regional power grids.
We are also on the verge of a major breakthrough with the Optimus (humanoid robot ) project, which I believe has the potential to become the largest product ever made. It's an extremely challenging project, and it's important to note that success is not automatic. As far as I know, Ford and General Motors... No other American car company has undertaken a similar robotics development project. I imagine people may still see Tesla as a car company—we currently primarily manufacture cars and battery packs. But developing Optimus was no easy feat. However, we possess all the elements necessary to develop real-world AI, exceptional electromechanical engineering capabilities, and the capacity for large-scale production, and I believe no other company currently possesses all of these simultaneously.
Regarding Autopilot version 14, you can see the reactions from netizens; they are all amazed. In fact, in the United States, any Tesla owner can upgrade to Autopilot version 14 simply by selecting "I want advanced software" in their car. If you are listening to this conference and would like to experience this system, simply go to the settings menu, select "I want advanced software," and you can complete the upgrade.
Regarding Megapack, we launched Megablock (large-scale energy storage module) and Megapack 3 (third-generation Megapack). We also have exciting plans for Megapack 4 (fourth-generation Megapack). Megapack 4 will integrate many functions typically found in substations and is expected to directly output 35 kV. This improvement will greatly enhance our ability to deploy Megapack, as deploying Megapack 4 no longer relies on building 335 kV substations. This will be a key engineering focus for the Megapack project.
We expect to launch the Optimus V3 (the third generation of the Optimus) in the first quarter of next year , at which time it should be ready for demonstration. I believe the third-generation Optimus will be amazing.
It doesn't even look like a robot; it looks more like a person in a robotic suit—which aligns with our original design philosophy for Optimus. It's so realistic that you'll want to poke it to confirm it's indeed a robot. Clearly, most of the real-world AI technologies we've developed for automobiles can be applied to Optimus, providing a solid foundation for its development .
Finally, we are thrilled with Tesla's updated mission—"sustainable abundance." Our mission is no longer limited to "sustainable energy," but has been upgraded to "sustainable abundance." We believe that, with Optimus and Autopilot technology, we can create a world free of poverty where everyone has access to top-tier healthcare. Imagine a world where everyone has access to top-notch surgeons. For example, Optimus could become an excellent surgeon. Imagine what the world would be like if everyone had access to the care of top surgeons.
Of course, we need to ensure that the Optimus meets safety and performance standards in all aspects. But I firmly believe that we are moving towards a world of sustainable abundance. I am very excited to work with the Tesla team to achieve this goal.
Travis Axelrod, Vice President of Investor Relations
Okay, thank you very much, Elon, for sharing. Vaibaf will also be giving some opening remarks.
Chief Financial Officer – Vaybaf Tania
Thank you, Travis. The third quarter was significant on multiple levels. We not only set new records in vehicle deliveries and product deployments, but also in a range of financial metrics including total revenue, energy gross margin, energy profit margin, and new free cash flow. This achievement is a testament to both the continued trust our customers have in our products and the tireless efforts of the Tesla team. The strong delivery growth was driven by outstanding performance across regions: deliveries in Greater China and Asia Pacific increased by 33% and 29% sequentially, respectively ; North America by 28%; and Europe, the Middle East, and Africa (EMEA) by 25% . This growth was also fueled by continued market enthusiasm for the new Model Y.
We previously stated that 2025 would be the "Year of the Model Y," and we have delivered on that promise: launching the new Model Y in the first quarter of this year, followed by the long-wheelbase and performance versions, and most recently the standard Model Y in North America, Europe, the Middle East, and Africa. Currently, our autonomous taxis are operational in two markets: Austin and most of the San Francisco Bay Area. Since the initial launch of our autonomous taxi service in Austin, we have expanded our service coverage three times and plan to continue expanding further.
Unlike our competitors, our fleet of autonomous taxis can blend seamlessly into the market because these vehicles do not have additional sensors . The integration of the vehicle with external devices won't feel out of place. This is an advantage of our current vehicle lineup that hasn't received enough attention—all our vehicles are designed with autonomous driving capabilities in mind from the outset. We believe, as Elon has stated, that as more people experience supervised fully autonomous driving (FSD) services at scale, market demand for our vehicles will increase significantly.
We have made good progress in the adoption of fully autonomous driving systems. However, it's important to note that the current customer base for paying for fully autonomous driving systems remains relatively small, representing only about 12% of our existing vehicle fleet. We are working with regulatory agencies in markets such as China, Europe, the Middle East, and Africa to obtain the necessary approvals to launch fully autonomous driving systems in these regions. Now, let me briefly outline the financials: Automotive revenue increased by 29% sequentially, driven by increased deliveries. While regulatory credit revenue declined sequentially, we secured new contracts and continued to fulfill existing ones. Excluding regulatory credits, our automotive profit margin improved slightly from 15% to 15.4%, primarily due to lower raw material costs and improved fixed cost allocation resulting from increased production. The energy storage business continued its strong performance, achieving record deployments, gross profit, and profit margins.
As we discussed previously, the energy storage business is more affected by tariffs as a percentage of cost of goods sold (COGS) because all products in this business are currently sourced from China, although we are exploring alternative sourcing channels. However, this situation is improving as the Shanghai Gigafactory's capacity gradually increases—we are using this factory to meet demand in non-US markets, thereby avoiding tariffs. As Elon Musk stated, in the field of grid-scale energy storage, the only way to quickly optimize power supply is through energy storage technology. Furthermore, it's worth noting that the energy storage business currently faces some headwinds due to increased competition and tariff impacts. In the third quarter, the combined tariff impact on the two businesses (Automotive and Energy Storage) exceeded $400 million, with roughly equal impacts on both. Services and other businesses improved significantly sequentially, primarily due to our insurance... The business and service center operations have improved.
It should be noted that, despite its smaller scale, the costs of our autonomous taxi business are included in the "Services and Other Businesses" category along with other business costs such as paid supercharging, used car sales, and sales of parts and related products. Our operating expenses increased sequentially, with the largest increase being in restructuring and other related expenses, which relates to certain initiatives we have undertaken to reduce costs and improve efficiency by integrating our AI chip design business. In addition, we incurred legal costs related to certain legal proceedings and additional expenses incurred in preparing for the shareholders' meeting, all of which were included in selling, general and administrative expenses (SG&A). Furthermore, our employee-related expenses are also increasing, particularly in research and development (R&D) – we recently granted several performance-based equity awards to employees working on AI projects, and therefore, these expenses are expected to continue to increase in the future.
Our other revenue declined sequentially, primarily due to a decrease in market capitalization-adjusted gains from Bitcoin holdings—$80 million in the third quarter, significantly lower than the $284 million in the second quarter. Other changes were attributed to currency fluctuations this quarter. Our free cash flow reached a record high of approximately $4 billion this quarter. As of the end of the quarter, our total cash and investments exceeded $41 billion. Regarding capital expenditures (CapEx), while we project approximately $9 billion this year, we anticipate a significant increase in this figure by 2026 as we prepare for the next phase of growth for the company—this involves not only expanding our existing businesses but also investing in AI projects, including Optimus. Finally, it's important to emphasize that applying AI to the real world is no easy feat, but we are never afraid of challenges.
We are optimistic about the future and are laying the foundation for the company's long-term development. Over the next few years, we will gradually reap the rewards of these efforts. Finally, I would like to thank the Tesla team, our customers, investors, and supporters for their unwavering trust in us.
Travis Axelrod, Vice President of Investor Relations
Thank you very much, Vaibaf, for sharing. Now, let's answer some investor questions. The first question is: What are the latest data on autonomous taxis, including fleet size, total mileage, number of completed trips, and human intervention rate? When can safety drivers be eliminated? What are the current obstacles preventing the deployment of unsupervised fully autonomous driving (FSD) to customer vehicles?
CEO Elon Musk
I'll give a brief answer to this question first, and Ashok will provide further details later. We anticipate that by the end of this year, autonomous taxis will no longer require safety drivers in at least most areas of Austin. In the coming months, we expect to achieve completely driverless operation in at least parts of Austin. Obviously, we will be very cautious during the deployment process. Our goal is to maintain a "cautious" approach during deployment because it's clear that even a single incident could make global headlines. Therefore, a cautious deployment strategy is more prudent for us.
We do anticipate that in the coming months, Austin's self-driving taxis will be operating without safety drivers—I think that's probably the most important data point. We also expect to have self-driving taxis operating in approximately 8 to 10 metropolitan areas by the end of this year , the exact number depending on the progress of various regulatory approvals. In fact, most of our regulatory applications are publicly available online. We anticipate that by the end of this year, self-driving taxis will be operating in Nevada, Florida, and Arizona. Ashok, can you add something?
Ashok Eluswamy, Vice President of Vehicle Software
Okay. Currently, our autonomous taxi fleet operating in Austin has achieved driverless operation, accumulating over 250,000 miles. In the San Francisco Bay Area, due to local regulations, our autonomous taxis still require a driver, but the cumulative mileage has exceeded 1 million miles. Practice shows that our autonomous taxi fleet is operating well, with positive customer feedback and no major issues. At the customer level, as of yesterday, customers using the supervised fully autonomous driving system have accumulated 6 billion miles, a significant milestone. Overall, the system has demonstrated excellent safety performance. As Elon stated, we are progressing according to plan, first achieving fully driverless autonomous taxi operation in Austin.
Travis Axelrod, Vice President of Investor Relations
Okay, thank you very much Ashok for the addition. The next question is: What is the demand and order backlog situation for Megapack, Powerwall, solar products, or energy storage systems? With the current AI boom, does Tesla plan to provide power support for other hyperscale data center operators?
Mike Snyder, Vice President of Energy Business
Thank you for your question. Demand for Megapack and Powerwall remains strong, and this momentum is expected to continue into next year. We have received positive feedback from customers regarding the Megapack product, which will begin shipping from our Houston facility next year. This is driven by the growing demand from hyperscale data center operators and utilities. As the company has come to recognize the versatility of Megapack products in improving reliability and alleviating grid stress (as Elon previously mentioned), we have observed a significant increase in demand for energy storage products in the AI and data center sectors.
In addition, affected by policy adjustments, residential solar power in the United States... Market demand is surging, and we expect this growth trend to continue into the first half of 2026, with the launch of new solar leasing products in the first quarter of next year . We have begun production of Tesla residential solar panels at our Buffalo factory and plan to begin shipping to customers in the first quarter of next year. These solar panels are industry-leading in both design and shading resistance, reflecting our continued commitment to U.S. manufacturing.
Travis Axelrod, Vice President of Investor Relations
Okay, thank you very much, Mike, for your answer. Unfortunately, the next question concerns future products, and this isn't the appropriate time to discuss that topic, so we'll skip it. The next question is: considering application control software, engineering hardware, general mobile model training, task-specific model training, speech model training, manufacturing implementation, and supply chain development , what are the current challenges in bringing Optimus to market?
CEO – Elon Musk
Clearly, bringing Optimus to market was an extremely challenging task, no easy feat. However, from a technical standpoint, Optimus is already capable of navigating within the campus. At our engineering headquarters in Palo Alto, California, the Optimus robot is active almost 24/7 within the office building. Any visitor can call out to an Optimus robot and ask it to take them somewhere; it will actually lead them to the designated meeting room or location within the building. But I don't want to downplay the difficulty of developing Optimus; it was indeed an incredibly arduous task. Developing a robotic hand that is as dexterous and powerful as a human hand is exceptionally difficult—the human hand is nothing short of a marvel.
The deeper you delve into the study of the human hand, the more astonished you become by its intricate details: why humans have four fingers and a thumb, why fingers possess specific degrees of freedom, why different muscles vary in strength, and why fingers differ in length. It turns out that these features all have their reasons. Developing a robot's hand and forearm is a massive engineering challenge—similar to the human hand, the muscles controlling hand movement are primarily concentrated in the forearm, and this is also true for Optimus's hand and forearm. From a mechatronics perspective, developing the forearm and hand is far more difficult than developing any other part of the robot. In fact, to create a practical, general-purpose robot, it must possess a highly flexible hand, real-world AI capabilities, and be mass-producible to generate real value—producing only a few hundred robots is meaningless. We need to achieve mass production of Optimus, with output comparable to, or even higher than, that of automobiles. If we plan to produce one million Optimus units annually, the manufacturing challenges will be enormous, because currently, there is simply no suitable material for humanoid robots. The supply chain is crucial. While the automotive and computer industries have mature supply chains, the humanoid robot industry lacks one entirely. Therefore, to produce Optimus, Tesla must achieve a high degree of vertical integration, deeply integrating into the downstream of the supply chain to manufacture its own components, as there is currently no readily available supply chain.
I've imagined myself as a startup trying to develop humanoid robots , and I've found that without advanced manufacturing technology, this work is simply impossible. That's why I believe Tesla is almost unique in its combination of manufacturing technology, scalability, real-world AI capabilities, and the ability to develop highly flexible robotic hands. Other robotics projects often lack at least one of these three key elements. I believe that through relentless effort, we can overcome all these challenges—that's our plan. My biggest concern about my voting control at Tesla is: if I lead the creation of this massive robotic "army," will I be ousted someday?
This is my biggest concern, and the only problem I'm currently trying to solve. It may seem like a salary issue, but I'm not trying to squander the money. Ultimately, the question is: if we build this robot "army," will I have enough influence over it? Not absolute control, but significant influence. In short, without sufficient influence over this robot "army," I cannot confidently advance this project.
Travis Axelrod, Vice President of Investor Relations
Okay, thank you very much for Elon's reply. We have already discussed the expansion plans for autonomous taxis. Unfortunately, the next question also concerns future products, so we will skip that one. The next question is: Could you provide an update on the $16.5 billion chip collaboration agreement with Samsung in Taylor, Texas? Given the semiconductor... Crucial to Tesla's Autopilot technology and AI-driven future, what makes you believe that Samsung can deliver the AI5 chip on time and has an advantage over TSMC in terms of yield and cost?
CEO – Elon Musk
Okay, I will answer this question in detail, because I need to break it down first and then respond to each point.
First, I have nothing but praise for Samsung—it's an exceptional company. It's worth mentioning that Samsung is currently manufacturing our AI4 chip, and it's performing exceptionally well. Regarding the AI5 chip, I need to clarify some comments I've made publicly before: Initially, we will have TSMC and Samsung jointly focus on AI5 chip production. I believe Tesla's AI5 chip design is a masterpiece. For the past few months, I've spent almost every weekend working with the chip design team on the AI5 chip. I don't readily praise others, but I must say that Tesla's chip team is indeed designing an extraordinary chip.
In some metrics, the AI5 chip will be 40 times more powerful than the AI4 chip, not 40%. Because of our deep understanding of the entire software and hardware architecture, we were able to precisely address all the pain points at the software level when designing the hardware. I believe no other company currently has achieved this – the deep integration of the entire architecture and calibration with real-world data (how our cars and robots operate in the real world). We know exactly what the chip needs to do, and equally importantly, we know what it doesn't need. For example, in the design of the AI5 chip, we removed the traditional graphics processing unit (GPU) found in the AI4 chip; furthermore, we removed the image signal processor (ISP).
We removed many features, and these cuts were crucial. These adjustments allowed us to integrate the AI5 chip into a semi-modular unit, while reserving ample space for wiring connections between memory and the chip, the Tesla chip accelerator, the ARM CPU core, and the PCI interface module. This is a beautifully designed chip, and I personally invested a great deal of effort in its development. I am confident that this chip will be a breakthrough product, leading the industry into a new era. Therefore, it is a logical choice for TSMC and Samsung to jointly focus on the production of the AI5 chip. Technically, Samsung's factory equipment is slightly more advanced than TSMC's. Both companies will produce their AI5 chips in the United States: TSMC's production base is located in Arizona, and Samsung's is located in Texas. Our clear goal is to achieve an oversupply of AI5 chips, and initial production will be driven by this goal.
Even if we produce more AI5 chips than are needed for automobiles and robotics, we can still use the surplus chips in data centers. Currently, our data centers are already using AI4 chips in their training processes, and these AI4 chips are being integrated with NVIDIA chips. This involves using NVIDIA hardware in conjunction with AI4 chips. It's important to clarify that we are not aiming to replace NVIDIA , but rather to integrate the AI4 chip with NVIDIA hardware. This allows us to utilize surplus AI5 chips in data centers. While NVIDIA's technology continues to advance, their challenge lies in meeting the diverse needs of numerous customers. Tesla, on the other hand, only needs to meet the needs of its own customer, significantly reducing the complexity of chip design and allowing us to eliminate many unnecessary and complex functionalities from the chip. The importance of this cannot be overstated.
Imagine the various logic modules within a chip: as the number of logic modules increases, so does the number of interconnecting lines between them. This is analogous to highways in a city—how many highways would you need to build to connect all parts of the chip? Especially when you can't determine the amount of data transfer between the logic modules, it could result in the chip's interior being filled with complex "highways," making chip design virtually an impossible task. NVIDIA excels at handling these extremely complex requirements. We, on the other hand, strive for ultimate simplicity.
Ultimately, I believe the AI5 chip may outperform its competitors by 2 to 3 times in performance per watt and by 10 times in AI performance per dollar. Of course, practice is the sole criterion for testing truth. Clearly, we need to actually produce this chip and achieve mass production—and we are currently working towards that goal.
Travis Axelrod, Vice President of Investor Relations
Okay, thank you very much, Elon, for your detailed explanation. We have already discussed the issues related to unsupervised fully autonomous driving systems. The next question is: why not encourage customers to trade in their old vehicles (trading in vehicles equipped with Hardware 3 for new cars) instead of going through the trouble of upgrading from Hardware 3 (third-generation hardware) to Hardware 4 (fourth-generation hardware)?
Lars Moravi, Vice President of Vehicle Engineering
Yes, we haven't completely abandoned Hardware 3. However, over the past year, we've offered customers the option to transfer Full Self-Driving (FSD) functionality to their new vehicles, sometimes with promotional offers. Customers who previously purchased FSD could enjoy better prices when trading in their new cars. We've been handling this issue carefully, but our primary goal is to resolve the technical challenges of FSD. Afterward, we'll launch solutions to support these customers. These customers are crucial to us; they are our early adopters. As an aside, I commute daily in a car equipped with Hardware 3, and I use the FSD system every day. We are confident we can provide satisfactory solutions for these customers.
Travis Axelrod, Vice President of Investor Relations
Okay, thank you very much for your answer.
Ashok Eluswamy, Vice President of Vehicle Software
Once the full rollout of the 14 version of the autonomous driving system is complete, we plan to develop a simplified version of Hardware 3 (V14 Lite), which is expected to be launched in the second quarter of next year.
Travis Axelrod, Vice President of Investor Relations
太棒了，谢谢阿肖克。好的，最后一个问题是：多久之后我们能看到具备自动驾驶功能的特斯拉半挂卡车（Semi）？你认为这项技术有可能取代火车吗？
首席执行官—— 埃隆·马斯克
好的，我先从半挂卡车的生产计划和时间表说起。半挂卡车工厂的建设进展顺利，厂房已完工，目前正在安装设备。我们的验证车队已在道路上进行测试。今年年底，我们将扩大半挂卡车的生产规模；明年上半年，我们将实现半挂卡车的线上量产；到明年下半年，产量将大幅提升。目前，各项工作进展顺利—— 这是实现自动驾驶半挂卡车上路的第一步。至于火车，它们在长距离点对点运输方面确实表现出色，效率极高。但在“最后一英里” 的装卸货环节，以及短途运输场景中，自动驾驶半挂卡车的表现会更具优势，这将为物流运输带来极大便利。
正如我此前所说，随着我们彻底改变人们对交通方式的认知，运输行业的格局有望发生转变。我们对这一变革的时间表充满期待。阿肖克，关于完全自动驾驶技术在半挂卡车上的应用，你可以补充说明一下。
目前，我们的团队正全力攻克乘用车的自动驾驶技术。不过，一旦我们获取了足够的半挂卡车运行数据，这些自动驾驶技术将很容易应用到半挂卡车上。
投资者关系副总裁—— 特拉维斯·阿克塞尔罗德
好的，接下来进入分析师提问环节。第一个问题来自沃尔夫公司（Wolfe）的伊曼纽尔。伊曼纽尔，请解除静音发言。
伊曼纽尔·沃尔夫
非常感谢。大家好。埃隆，您之前提到，既然对无监督自动驾驶技术有了信心，就会尽快扩大汽车产量。结合特斯拉目前300万辆的产能，我们该如何理解这一扩产计划？您是否希望将产量提升到这一水平？大概会有怎样的时间规划？这是否需要通过一定措施刺激需求？考虑到长期机遇，特斯拉是否会在短期内优先追求产量而非盈利能力？
首席执行官—— 埃隆·马斯克
我们目前的产能尚未达到300 万辆，但未来终将实现这一目标。理想情况下，我们或许能在24 个月内达到年化300 万辆的产能，甚至可能用不了24 个月。但要注意的是，这需要庞大的供应链体系同步跟进。我们会在自身能力及供应商供货能力允许的范围内，以最快速度扩大产能。此外，我们还在考虑未来新建更多工厂以进一步提升产能。其中，产能扩张最关键的项目是Cybercab（自动驾驶出租车），该车将于明年第二季度启动生产。这款车是专为完全自动驾驶优化设计的车型。
事实上，Cybercab 没有方向盘和踏板，其工程设计核心是将每英里运营成本降至最低，全方位考量了运营中的单位里程成本。而我们其他车型的设计仍带有“无马马车” 的影子—— 显然，如果车辆配备方向盘和踏板，且设计初衷是满足用户对高速加速、急转弯等高性能的需求，那么其设计思路会与主打舒适驾乘（最高时速仅需85-90 英里）的车型截然不同。Cybercab 的定位就是全程提供平稳舒适的驾乘体验。至于扩产是否会牺牲利润率，我认为不会，因为市场对这类车型的需求将会非常旺盛。
真正的“杀手锏” 功能其实很简单：用户在车内能否放心地发短信？如果告诉用户“现在这款车足够智能，你在车内全程都能玩手机、发短信”，那么只要有购买能力的人都会选择这款车，事情就是这么简单。这正是所有人真正想要的功能。事实上，很多人不仅“想” 这么做，现在就已经在这么做了。全球范围内交通事故率上升的原因之一，就是人们在驾驶时发短信。而自动驾驶辅助系统（Autopilot）能显著提升安全性，因为当驾驶员低头看手机时，驾驶操作会大幅变形，自动驾驶辅助系统恰好能弥补这一问题—— 这才是真正改变游戏规则的关键。
目前，我可以100% 确定地说，我们能攻克无监督完全自动驾驶技术，而且其安全水平将远超人类驾驶。我们已经发布了14.1 版本的自动驾驶系统，后续的技术路线图也极具前景。我们将为车辆加入“推理能力”，而用于强化学习的世界模拟器性能也十分出色。当你看到特斯拉实景模拟器生成的画面时，你根本无法区分它与真实路况视频的差别—— 两者看起来完全一致。这一模拟器能为我们构建强大的强化学习循环，进一步提升特斯拉AI 的性能。我们计划将AI 模型的参数数量提升一个数量级（这一改进不会体现在14.1 版本中），同时还会对AI 进行多项重大优化。
搭载AI4 芯片的车辆，其AI 性能会让你感觉它像一个“活物”。而这还只是AI4 芯片的水平，AI5 芯片的性能会更加强大—— 正如我之前所说，从部分指标来看，AI5 的性能将是AI4 的40 倍，保守估计也能达到10 倍提升。甚至可能出现一种情况：汽车的智能程度过高，以至于它可能会“感到无聊”。我曾设想过，如果大量车辆处于闲置“无聊” 状态，我们可以将它们组成一个庞大的分布式推理集群—— 在车辆不行驶时，让它们参与分布式推理运算。
未来，若我们的车队规模达到数千万辆，甚至某一天突破1 亿辆，假设每辆车具备1 千瓦的高性能推理能力，那么整个集群的总推理能力将达到100 吉瓦，而且电力供应和散热问题都已得到解决。这无疑将成为一项极具价值的资产。
投资者关系副总裁—— 特拉维斯·阿克塞尔罗德
好的，谢谢埃隆。下一个问题来自摩根士丹利（Morgan Stanley）的亚当。亚当，请解除静音提问。（稍作停顿）似乎亚当的音频出现了问题，我们稍后再请他提问。接下来请巴克莱bank （Barclays）的丹提问。
巴克莱银行分析师丹
晚上好，感谢各位接受我的提问。埃隆，我知道特斯拉在“第四篇章宏图计划”（Master Plan 4）中重点推进AI 与物理世界的结合。过去一段时间，我们也看到特斯拉积极开拓新市场、探索新目标市场规模（TAM）。在您看来，特斯拉的核心竞争力应如何界定？在AI 应用或市场拓展方面，特斯拉的边界又在哪里？哪些领域属于非核心竞争力范畴？
首席执行官—— 埃隆·马斯克
其实，我不太理解你所说的“特斯拉核心竞争力的AI 应用” 具体指什么。要知道，特斯拉创立之初，根本没有所谓的“核心竞争力”—— 当时我们的竞争力几乎为零。你可以把特斯拉看作是“一家公司里装着十几家初创企业”，而每一家“初创企业” 都是由我牵头启动的。比如，我们以前并不生产储能电池组，但现在不仅生产，还推出了面向家庭的Powerwall 和面向公用事业级市场的Megapack；我们搭建了全球范围内的超级充电网络，这是其他任何企业都未能实现的成就—— 北美地区的超级充电网络甚至吸引了该地区几乎所有其他汽车制造商采用我们的充电标准。
如果搭建超级充电网络真的那么容易，为什么其他企业不自己做呢？再比如，我们的芯片设计团队是从零组建的，特斯拉AI 软件团队也是如此。我的做法通常是：在X 平台（原Twitter）上发文宣布“我们要启动某个项目”，然后邀请感兴趣的人加入。事实上，阿肖克是我为特斯拉自动驾驶团队（现更名为特斯拉AI 软件团队，因其核心是AI软件开发）面试的第一人。可以说，我们的核心竞争力是“按需创造” 出来的。
规模化生产Optimus 人形机器人，将是实现“无限财富密码” 的关键。很难用语言形容其潜力—— 比如，Optimus 的年生产力可能达到人类的5倍，因为它可以24小时不间断工作，甚至无需充电（可通过有线供电持续运行）。这也是我所说的“可持续富足” 的核心：未来，工作将成为一种选择，而非必需。AI 对人类生产力的提升存在上限，但实体化AI（如Optimus）的潜力是无限的—— 这就是我称其为“无限财富密码” 的原因。
我还想补充一点：很多人忘记了，特斯拉自动驾驶系统的第一个版本早在10年前就已推出，我从那时起就开始推进这项技术了。我们有实实在在的成果可以证明这一点。
车辆工程副总裁—— 拉尔斯·莫拉维
完全正确。即便是Optimus 项目，很多人以为它是“新项目”，但我还记得，四年多前的一次财务会议上，埃隆就说过“我们的汽车是轮式机器人”—— 从那时起，我们就开始为后续的人形机器人研发积累技术。事实上，目前Optimus 团队的大部分工程师都来自汽车研发领域。这也是为什么我们在制造环节能快速推进：当年研发驱动单元的工程师，现在正在研发Optimus 的执行器。如果说有哪家公司能实现人形机器人的规模化生产，那一定是特斯拉。
首席执行官—— 埃隆·马斯克
我们也为Optimus 团队招募了大量新工程师，其中很多是刚毕业的大学生，他们为项目做出了重要贡献。Optimus 团队的工程师们才华横溢，每次看到他们的成果我都会感叹“太出色了”。目前，我们对Optimus 的研发和制造进行同步评审，形成“设计- 制造” 的迭代循环：设计出原型后，若发现制造难度过大，就立即调整设计方案，确保其具备可量产性。通过这一过程，我们在提升Optimus 功能的同时，大幅优化了其制造工艺。坦白说，Optimus 2 的设计几乎不具备量产可能性。
回顾Optimus 的发展历程：从最初“像人穿着机器人外壳” 的原型，到Optimus 2.5 能完成功夫动作—— 它还在《创：战神》（Tron）的首映礼上与杰瑞德·莱托（Jared Leto）一起表演功夫，全程无人操控，相关视频在网上就能看到。有趣的是，很多人路过时误以为它是真人—— 尽管Optimus 2.5 的腰部只有3英寸宽（明显不是人类的生理结构），但其动作的拟人化程度极高，以至于很多人没意识到它是机器人。而Optimus 3 将在这一基础上实现巨大突破，同时具备规模化生产能力—— 这无疑是一项艰巨的任务。
目前，我们每周五晚上会召开Optimus 专项会议（有时会持续到午夜），每周六下午则与AI5 芯片设计团队开会。这两项工作对公司的未来至关重要。
投资者关系副总裁—— 特拉维斯·阿克塞尔罗德
好的，丹，您有后续问题吗？
巴克莱分析师丹
是的，我想追问一个相关问题：特斯拉的AI 团队与xAI（埃隆旗下另一家AI 公司）的工作是互补关系，还是各自专注于不同类型的AI？能否为听众解释两者的区别？谢谢。
首席执行官—— 埃隆·马斯克
两者专注于不同类型的AI。以xAI 的Grok 模型为例，它是一个规模极其庞大的模型，根本无法在汽车上运行—— 它就像一个“庞然大物”。Grok 的目标是通过大规模训练和算力支持，实现通用人工智能（AGI）。比如，Grok 5 模型只能在GV300 算力平台上有效运行，其规模可见一斑。而特斯拉的AI 模型规模要小得多，可能还不到Grok 的10%，甚至接近5%。两者解决问题的路径截然不同：xAI 的Grok 主要与谷歌Gemini、OpenAI ChatGPT 等模型竞争，不过在某些领域，两者也存在互补性。
例如，在车内与Grok 进行语音交互会是很有趣的功能；Optimus 的语音识别和语音生成功能也会用到Grok 技术—— 这些都是互补的体现。但总体而言，特斯拉AI 和xAI 的Grok 是从AI 光谱的两个极端出发解决问题。
投资者关系副总裁—— 特拉维斯·阿克塞尔罗德
好的。亚当，我们再尝试一次，请您解除静音准备下一个问题。（稍作停顿）很遗憾，音频问题仍未解决。我们接下来请莱特希德公司（Lightshed）的沃尔特提问。沃尔特，请解除静音。
莱特希德分析师沃尔特
现在能听到我说话吗？
投资者关系副总裁—— 特拉维斯·阿克塞尔罗德
可以听到。
莱特希德分析师沃尔特
太好了，谢谢。我想回到奥斯汀自动驾驶出租车的话题：既然你们已能在奥斯汀取消安全驾驶员，那么旧金山湾区仍需配备安全驾驶员，是否完全是出于监管要求？还是说，你们需要针对每个市场进行逐步学习？另外，您之前提到计划新增8-10 个运营市场，在这些市场中决定是否配备安全驾驶员（或安全乘坐人员），是你们开拓新市场的标准步骤，还是单纯由当地监管规定决定？我个人认为，即便监管不强制要求，谨慎起见，你们可能仍会选择配备安全人员。
首席执行官—— 埃隆·马斯克
出于对安全的“极端谨慎”，我认为在新市场初期配备安全驾驶员或安全乘坐人员是合理的，目的是排查潜在风险。要知道，哪怕万分之一的行程出现问题，都可能引发严重后果。我们需要确认：目标城市是否存在特殊路况（比如极其复杂的交叉路口），或其他意想不到的挑战，以应对那万分之一的极端情况。理论上，我们或许可以直接在这些城市启动无安全员运营，但我们不愿冒这个风险。
我们的计划是：在新市场的前三个月配备安全驾驶员，确认系统运行稳定后，再取消安全员—— 整个过程大概就是这样。
莱特希德分析师沃尔特
好的。关于完全自动驾驶14 版本（FSD 14），它的体验与13 版本有明显区别，与奥斯汀自动驾驶出租车的系统体验也有所不同。请问，自动驾驶出租车的技术研发路线，与面向早期用户推送的版本是否存在差异？在推送新版本时，你们的核心目标是降低人工干预率，还是说干预率问题已基本解决，当前重点转向新增功能（如自动泊车、驾驶模式选择）或提升整体驾乘舒适度？
首席执行官—— 埃隆·马斯克
当我们发布自动驾驶系统（Autopilot）的重大软件架构更新时，首要优先级始终是“安全”—— 先确保安全，再优化舒适度。这也是我不建议大家立即升级初始版本的原因：我会建议大多数用户等到14.2 版本发布后再下载14 系列系统，因为到那时，我们已解决了大部分舒适度问题。安全优先，舒适度优化紧随其后—— 这就是为什么很多人会觉得“早期版本虽然安全，但行驶过程可能有些颠簸”。对于自动驾驶系统架构的重大更新，我们需要时间打磨细节，在确保安全的基础上提升舒适度。
我对特斯拉现实世界AI 的技术路线图有非常细致的了解（阿肖克是这一领域的负责人）。我花了大量时间与团队深入讨论AI 的优化方向，正如我之前所说，搭载AI4 芯片的车辆会让你感觉它像一个“活物”—— 而这还未用到AI5 芯片。
这份技术路线图令人极其振奋，我们迫不及待地想将正在研发的功能推向市场。至于面向普通用户的版本与自动驾驶出租车版本，两者核心技术基本一致。当然，普通用户版本会有一些专属功能（比如选择在停车位还是车道内泊车），这些功能对自动驾驶出租车而言并不重要，除此之外仅有少量细微差异。两者的算法和架构主体是完全相同的。
正如我之前提到的，我们计划为系统加入“推理能力”。阿肖克，这项功能大概会在14.3 或14.4 版本中推出，对吗？
车辆软件副总裁—— 阿肖克·埃卢斯瓦米
是的，无论是搭载AI4 还是AI5 芯片的车型，今年年底前肯定能实现这一功能。
首席执行官—— 埃隆·马斯克
没错。具备推理能力后，车辆会自主判断“该选择哪个停车位”。比如，它会分析：“这是商场入口，但如果停车场车位紧张，入口附近有空位的概率很低”，进而做出决策—— 先将乘客送到商场入口，再自行寻找停车位。它还能更精准地识别空车位，甚至比人类更高效，因为它拥有360 度全景视野。正如我所说，推理能力将帮助车辆解决更多复杂场景问题。
但真正的挑战在于：如何在搭载AI4 芯片的车载计算机中实现这一功能。因为在服务器端实现推理可能需要很长时间，但车载系统必须实时做出决策—— 如何在有限的车载算力中集成这些功能，是团队当前面临的核心任务。
车辆软件副总裁—— 阿肖克·埃卢斯瓦米
这也是为什么我敢说，在深入了解Grok 这类大型模型和特斯拉AI 模型后，我可以确定：特斯拉AI 的“智能密度” 是最高的。如果以“每千兆字节（GB）承载的智能” 为衡量标准，特斯拉AI 可能比其他任何AI 高出一个数量级。这其实是“被逼出来的”—— 因为它必须适配AI4 芯片的算力限制。而这种对“智能密度” 的极致追求，将在AI5 芯片（算力提升一个数量级）上释放巨大价值：在保持同等智能密度的前提下，AI5 芯片的性能将实现10 倍飞跃。
投资者关系副总裁—— 特拉维斯·阿克塞尔罗德
好的。下一个问题来自奥本海默公司（Oppenheimer）的科林。科林，请准备好后解除静音提问。
奥本海默分析师科林
非常感谢各位。您之前提到了人形机器人手部灵活性、供应链复杂性以及特斯拉正在推进的垂直整合策略，我想进一步了解：明年启动Optimus 量产的计划，与当前供应链现状是否匹配？此外，手部灵活性的研发似乎仍有大量工作要做，在硬件设计最终定型并启动规模化生产前，还需解决哪些问题？
首席执行官—— 埃隆·马斯克
事实上，即便Optimus 启动量产，其硬件设计也不会完全“冻结”—— 我们会持续迭代优化。因为很多制造层面的问题，只有在量产后期才会暴露，所以我们会对Optimus 的设计进行“滚动式调整”，即便量产启动后也不例外。不过，我必须说，新版机器人手部的工程设计堪称杰作。
正如我之前所说，明年第一季度（可能在2-3 月），我们将展示具备量产意向的Optimus 原型机。我们计划搭建一条年化产能100 万台的Optimus 生产线，希望能在明年年底前启动生产。不过，要达到年化100 万台的产能需要一定时间—— 因为生产过程涉及10000 个独特零部件，整个产能爬坡速度将受限于“最慢、最容易出问题” 的环节。但最终我们一定会实现这一目标。未来，我们还会推出Optimus 4（目标产能1000 万台）、Optimus 5（目标产能5000 万至1 亿台）—— 这个规模确实非常庞大。
投资者关系副总裁—— 特拉维斯·阿克塞尔罗德
好的。很遗憾，今天的问答环节到此结束。在会议结束前，瓦伊巴夫将发表闭幕致辞。
首席财务官—— 瓦伊巴夫·塔内贾
谢谢特拉维斯。我想借此机会谈谈11 月6 日即将举行的一场至关重要的投票—— 这场投票将决定特斯拉的未来。我们恳请各位股东支持两项与埃隆薪酬相关的提案，并支持艾拉（Ira）、凯瑟琳（Kathleen）和乔（Joe）重新当选董事会成员。
特斯拉的成功是“团队协作” 的结果，董事会是这支胜利团队不可或缺的一部分。股东始终是我们一切工作的核心。此前，一个特别委员会已制定了一套薪酬方案—— 正如埃隆所说，我们甚至不愿将其称为“薪酬方案”。
首席执行官—— 埃隆·马斯克
是的，核心诉求很简单：我需要足够的投票权以确保对公司的“重大影响力”，但又不能拥有绝对控制权（以防我未来做出不理智决策时无法被罢免）。我认为，25% 左右的投票权是一个合理区间。作为一家已上市的公司，我们已探索了所有可能提升投票权的方式，但遗憾的是，上市后无法再设置“超级投票权股票”（Super Voting Stock）。谷歌、Meta 等公司之所以拥有这类股票，是因为它们在上市前就已设立，属于“ grandfathered（祖父条款）” 范畴，而特斯拉没有这样的基础。
正如我之前所说，如果我无法确保对公司的重大影响力，我不会安心推进机器人军团的研发—— 万一我因为ISS（机构股东服务公司）或Glass Lewis（格拉斯刘易斯公司）这类机构的荒谬建议而被罢免，那将是灾难性的。说实话，这些机构就是“企业恐怖分子”。
我来解释一下核心问题：如今，大量指数基金、被动型基金的投票决策完全依赖ISS 和Glass Lewis 的建议。这些机构过去曾提出过许多糟糕的建议，若这些建议被采纳，将对公司未来造成极大破坏。当被动型基金将投票权拱手让给ISS 和Glass Lewis 时，上市公司可能会面临灾难性后果。
投资者关系副总裁—— 特拉维斯·阿克塞尔罗德
当被动型基金控制了上市公司的大部分股权时，公司实际上已被ISS 和Glass Lewis 间接控制。这是公司治理领域的一个根本性问题—— 因为它们的投票决策并非基于股东利益最大化。这就是关键所在：ISS 和Glass Lewis 就是“企业恐怖分子”。
首席财务官—— 瓦伊巴夫·塔内贾
我想说的是，特别委员会制定的这套方案完全以股东利益为核心—— 只有当股东获得可观回报后，相关薪酬才会兑现。因此，我恳请各位股东不仅支持这套薪酬方案，也支持三位董事的连任—— 他们拥有卓越的专业知识和丰富经验。在特斯拉，我们与董事们保持着日常紧密沟通，几乎每天都会与某位董事或其团队交流。我们的董事绝非只会“念PPT”，他们会深度参与公司日常运营。再次恳请各位股东按照董事会的建议进行投票。谢谢大家。
投资者关系副总裁—— 特拉维斯·阿克塞尔罗德
好的，谢谢瓦伊巴夫。感谢各位今天的提问，我们期待下一季度与大家再会。非常感谢，再见。
Further Reading
(Article source: CLS)
