Full text | Tesla Q2 earnings call transcript: Robotaxi safety is extremely high

Tesla's total revenue in the second quarter was $28.236 billion, a year-on-year increase of 26%, driven by growth in vehicle deliveries, service and other business growth, and positive foreign exchange effects.

For details, see: Tesla Q2 revenue increased by 26% year-on-year, profit halved, and capital expenditure significantly increased.

After the earnings report was released, Tesla CEO Elon Musk, CFO Vaibhav Taneja, Vice President of Supply Chain Karn Budhiraj, Vice President of Automotive Engineering Lars Moravy, Director of Autonomous Driving Software Ashok Elluswamy, General Counsel Brandon Ehrhart, and Head of Investor Relations Travis Axelrod interpreted the earnings report and answered analysts' questions.

In the Q&A session with investors: mainly about Robotaxi, Cybertruck, and Optimus.

Ashok Elluswamy: First of all, I want to say that the Robotaxi program is running very well. Especially in terms of safety, the project maintains an impeccable safety record. Currently, our autonomous taxis have driven over 380,000 miles unsupervised in six cities across two states. During this time, we have not had any major accidents. All reports are about situations where we were affected by other vehicles while stationary. I want to emphasize again that so far, our operations have been very safe, with no major accidents during over 380,000 miles of driving.

Historically, so-called experts have always claimed that to achieve safe driving, you need lidar, radar, high-definition maps, and all imaginable technologies. Here, we prove that this is not the case. With just cameras, you can achieve safe, comfortable, and affordable autonomous driving. This record is a huge validation of Tesla's entire artificial intelligence (AI) approach.

Secondly, I want to discuss the issue of scaling up. About a year ago, we launched the autonomous taxi project in Austin. At that time, we installed safety monitors in the passenger seats. Around the end of last year, we launched the first fully driverless autonomous taxis in Austin. Since the beginning of this year, the unsupervised mileage driven by the fleet has been growing at a double-digit rate weekly. I say again, in the past few months, our weekly growth rate has reached such a high compound growth rate. Moreover, we expect this growth rate to continue to remain high for the remainder of this year.

Another noteworthy point is that our efforts to expand into new cities have been relatively minimal. We expect the time to expand into new cities to continue approaching zero until we can operate statewide, rather than advancing city by city.

The reason we can achieve such growth is that we have a world-class AI frontier development engineering team at Tesla. The currently operating fleet of autonomous taxis is running an early version of the v15 FSD software we mentioned earlier For version v15, we have roughly planned seven major improvement directions, all of which are being pursued in parallel. Currently, in the early v15 version running in the autonomous taxi fleet, 40% of the improvement directions have already been integrated, which is the system currently operating in the fleet.

As we continue to complete the work on version v15, we will see that this vehicle will have amazing safety and powerful performance. We have seen from internal indications that all the efforts we have made for technological improvements are continuously paying off, and we expect the returns to be substantial, which will help us maintain the same growth rate for the remainder of this year and into the future.

Finally, regarding Cybercab. First of all, it is an extraordinary product. Anyone who has tried it will immediately fall in love with the experience. We have adjusted our manufacturing targets to roughly match the expected growth rate of unsupervised driving mileage. The v15 model, which powers the Model Y and other platforms, will also be applicable to Cybercab.

In summary, so far, the growth rate of autonomous taxis (Robotaxi) is exponential while maintaining an impeccable safety record. Our technological investments are paying off, helping us sustain this exponential growth in the foreseeable future, and Cybercab will be fantastic.

Regarding the Optimus humanoid robot, especially the third-generation model of Optimus, its design must not only match the appearance of humans but also align with human functions and dexterity.

One of the main reasons for this approach is that having a human shape and function allows us to learn from humans how to perform a wide variety of tasks. This provides a wealth of data for training Optimus. Just like the fire department, we can access a large group of people who provide us with data from all the workers in the factory.

Optimus can learn a lot from observing them performing their tasks in the factory. We also have a dedicated data collection team that can provide a relatively small but very high-quality demonstration for further training of these models. All of this is built on any training we can do from data across the entire internet, which shows demonstrations of people performing various tasks in industrial and home environments.

The second phase is to have a large number of Optimus robots practice tasks in a place we call "Optimus Academy." The data from the robots themselves performing tasks will be invaluable and help us narrow any small gaps that may exist between robots and humans. This is also when the reinforcement learning loop begins, where robots initially attempt some tasks, sometimes fail, learn from the successes and failures of these tasks, and ultimately master these tasks at a superhuman level.

The AI strategy we have developed for Optimus aligns with this. It is the same end-to-end strategy that drives FSD, pixel input, and control output. Just like the fire department, we hope it can play a broad role. In FSD, you can get in the car, input the address, then click start, and it handles all the driving from one park to another Optimus is the same. You just need to let it do anything, and then you can complete the entire task yourself without having to do anything along the way. It is this team that developed FSD, v12, 13, 14, and now v15 software, which can also run on Optimus. I am very confident that Optimus will be a very capable robot.

Analyst Q&A session:

Morgan Stanley analyst Andrew Percoco: Thank you for the opportunity to ask a question. The first question revolves around the Optimus humanoid robot. Elon, you previously mentioned that the complete supply chain for Optimus currently almost does not exist, and the vast majority of components need to be developed and produced by Tesla itself. I would like to ask, in the process of promoting in-house development and production, are there external suppliers willing to invest simultaneously, build factories in the U.S., and assist Tesla in expanding production more quickly? We understand that Tesla will control core technologies, but from the perspective of investment efficiency, supply chain partners sharing capacity construction can optimize capital expenditures. I would like to understand your overall thoughts on external supplier cooperation and the current progress of communication with suppliers.

Musk: Overall, supplier cooperation is going very smoothly, with major manufacturers continuously investing huge amounts of money to support the implementation of projects like Optimus and Robotaxi. Among them, Samsung and TSMC are particularly prominent. TSMC's Arizona factory and Samsung's Texas factory have invested hundreds of billions of dollars specifically to produce digital AI computing chips suitable for humanoid robots and autonomous taxis. Amazon has also invested tens of billions of dollars to expand battery production lines. Although I am feeling a bit unwell today and my voice is not in great shape, I still want to thank all our partners for their financial investment and support for implementation.

Kahn Budilag: To add to Elon’s remarks, Samsung's new wafer fab capacity will heavily tilt towards Tesla's future projects, with an investment scale reaching the tens of billions of dollars. New robot-specific industrial chains, including storage chip manufacturers, new metal injection molded parts, and flexible printed circuit boards, are all simultaneously increasing their investments, completely distinguishing themselves from the traditional vehicle supply chain. If we cannot find high-quality suppliers that match our needs, Tesla will not hesitate to develop and produce in-house. We have top-notch manufacturing engineering and product design teams, fully capable of independently expanding component production capacity.

Musk: Additionally, I want to thank Micron for reserving storage chip capacity for Tesla. The current market for storage chips is extremely tight, and Micron has made significant internal capacity adjustments to provide Tesla with a large long-term capacity allocation at reasonable prices, for which we are very grateful.

Morgan Stanley analyst Andrew Percoco: Thank you for the addition. My second question is related to autonomous taxis. Currently, the regulatory policies regarding hard requirements for autonomous driving sensors are continuously changing across various states in the U.S. I would like to ask what regulatory policy requests Tesla has when implementing Robotaxi? Moderate regulation is beneficial for industry development, but excessive hard regulations will limit innovation. How does Tesla hope the regulatory framework will be reasonably established?

Kahn Budilag: Let’s first talk about federal-level regulation in the U.S. We are making significant progress; the National Highway Traffic Safety Administration (NHTSA) has issued a new version of the Federal Motor Vehicle Safety Standards (FMVSS), recognizing pure vision autonomous driving solutions, and we are very grateful for NHTSA's support The differentiation in state-level regulation is evident, and the stringent policies introduced by New Jersey are disappointing. As Ahsok mentioned earlier, we primarily rely on the actual safety performance of vehicles to persuade regulators and the public.

The optimal regulatory model is to set safety goals and allow companies to independently develop paths to achieve them. If regulations directly impose hardware solutions without focusing on safety outcomes, we do not accept that. Tesla has always used actual operational data to prove product safety, which is also the core support for our subsequent cross-regional expansion.

Bank of America analyst Alexander Perry: Thank you for the opportunity to ask questions. I have two questions regarding autonomous taxi services. First, what key milestones should investors track that will drive Tesla to accelerate fleet deployment and expand into new cities? Is it safety data (accident rates per million miles) or other metrics? Secondly, will Tesla collaborate with ride-hailing platforms to enhance vehicle utilization? Or will it continue to insist on vertically integrating and self-developing the Robotaxi business across the entire industry chain in the long term?

Musk: Yes. We hope to achieve vertical integration in the autonomous taxi business, just as we have done in other business areas. So I believe we will not encounter any demand challenges in the autonomous taxi business. Therefore, from an economic perspective, I think the appeal of autonomous taxi services will be tremendous, and we will genuinely be more willing to use this service to the extent that demand will exceed our service capacity. So clearly, in this case, you can only choose vertical integration.

So, this is essentially going through what we call the "5 nines reliability journey." In this process, you need to consider what level of reliability you need to achieve in order to scale. Ideally, you want to reach 99.999% reliability. So, this is the "5 nines" journey of reliability. I believe this is the only real constraint on the development of our autonomous taxi services.

Wells Fargo analyst Colin Langan: Thank you for the opportunity to ask a question. There are numerous rumors in the market that Tesla and SpaceX will merge, and both sides are already deeply collaborating in multiple fields. I would like to ask Elon if there is long-term synergistic value in a potential merger between the two companies and whether such a merger is feasible.

Musk: I mean, you can see from our numerous collaborations with SpaceX across various fields that our scope of cooperation is expanding, especially with the collaboration on Terafab, which will really be a massive project.

But obviously, we cannot discuss company mergers during a conference call. This must be done through the appropriate procedures. On this matter, I will hand the microphone over to our General Counsel, Brandon.

Brandon Elhardt: Thank you, Elon. We will continue to benefit from our collaboration with SpaceX, and they have always been a great partner. We have had many beneficial transactions with them. Earlier this year, we deepened our partnership through an investment and a framework agreement. This will enable us to continue working with them on the projects Elon mentioned, such as Terafab and Digital Optimus Musk: Yes. And there are many other things. Obviously, your car is equipped with Grok. So—Grok is helping to drive Digital Optimus. You also have Starlink integrated into the Cybercab, and Starlink will be integrated into all of our vehicles, at least in markets where Starlink is active. Because in the case of autonomous taxis, you need ubiquitous coverage. Even in Silicon Valley, there are many places where cellular network coverage is poor, and sometimes even nonexistent, which is surprising for Silicon Valley. But I know that during the first 10 to 15 minutes of my drive to work, I actually can't make calls because the cellular network connection is very poor.

Therefore, we can't let autonomous taxis get stuck in areas like the Bermuda Triangle where there is a lack of cellular network connection. So, having Starlink, which can provide connectivity anywhere, is actually very important. This way, our autonomous taxis won't go missing while performing tasks. Obviously, if people are sitting in the car, they will want to do something productive or entertaining. With Starlink, you can watch 4K live sports events in the car, and the cost per GB of data is very low, which is fundamentally unfeasible in cellular network systems. Additionally, there are many other scenarios.

Vaibhav Tania: To add to the Cybercab product logic. Just like Ashok mentioned, the experience of Cybercab is really great. In a world where you don't have to focus on driving, you can have conference calls, watch movies, or do anything else in the car. That's why when you see the design of the Cybercab, you'll find it has a large screen. We have already started producing Cybercabs at our factory in Austin, and customers will be able to experience it in the near future, and then you'll understand why this connectivity becomes so important.

Wells Fargo analyst Colin Langan: The second question is about the pace of deploying autonomous taxis. Currently, Tesla chooses to expand synchronously across multiple cities, but the number of vehicles deployed in each city is only in the dozens, rather than hundreds. Why not focus on a single city like Austin first, deploy a large fleet, and then expand outward? What is the core obstacle to large-scale deployment in a single city?

Ashok Elluswamy: The reason we are expanding in different cities rather than doubling down in just one city is that we want to ensure our business portfolio has broad applicability. It does have broad applicability. We just want to prove to ourselves and others that it can work in many different cities without requiring too much effort in each city. That's what we see internally. Then, as Elon mentioned earlier, the growth rate is actually exponential. It's just that we are still in the early stages of that exponential growth. That's why it's hard for others to understand.

In terms of mileage and the number of vehicles, since the vehicles in the autonomous taxi fleet are essentially running continuously, while human drivers may only use their vehicles for a few hours a day, these vehicles are mostly in continuous operation, which means that even with just a few cars, they can accumulate a lot of mileage This is why we track unsupervised driving mileage, not just the number of vehicles or other similar metrics. We want the fleet to operate efficiently. Therefore, all efforts to improve efficiency allow each vehicle in the fleet to drive more mileage.

Vaibhav Tania: Yes. I would like to add further, for example—think about it, we are trying to scale up. We are working hard to ensure that there are different bottlenecks. This is not only reflected in software but also in operations, where we are working to solve some issues. Therefore, we are trying to expand our business scope to ensure that before a truly large-scale deployment, we can address these issues in a controlled manner within smaller fleets.

Musk: Yes. Regarding Cybercab, one point to note is that since it is a new type of vehicle chassis, we need to accumulate Cybercab-specific driving data before it is widely deployed. Therefore, unlike Model 3, Model Y, and our other models, which have a large number of vehicles on the road, in the millions, Cybercab has not yet reached that scale. So, we actually have to accumulate mileage for Cybercab by retrofitting components like the steering wheel, brakes, and acceleration to calibrate the Cybercab chassis. Therefore, you will see—we are very confident that the number of Cybercabs in the city will increase significantly.

Lars Moravy: So, I would like to add a thought here. Another reason I want to point out is that the situation with traffic control (TCO) varies by city and state, and there is no federal framework to unify management. Therefore, the reason we are expanding city by city is to ensure that we can meet all these requirements one by one and do our best to meet the various needs of each city. We must expand to each city to meet these requirements, and we will continue to do so.

LightShed Partners analyst Walt Piecyk: I want to ask about the regulatory topic Lars mentioned. The head of NHTSA recently appeared on CNBC and stated plans to eliminate the mandatory steering wheel regulation for vehicles. Will the removal of the steering wheel become a core bottleneck for the scaling of Cybercab? Besides the steering wheel regulation, what other policy gaps at the federal level will restrict the expansion of driverless dedicated vehicles?

Lars Moravy: The short answer is "no." I mean, I think we have a good relationship with the National Highway Traffic Safety Administration (NHTSA) and especially with Director Morrison. I think they are just responding to public expectations because the world knows what is about to happen, and they are trying to stay ahead to ensure they are taking the right steps. So, over the past few years, we have been open with them, telling them what our plans are and what we are doing. I don’t want to say we are in sync, but I feel we have a partnership and we are working together Walter Pichik: Understood. Then, Elon, I want to ask a question about the roadmap. Obviously, Starlink has already been integrated into Cybercab, and you just talked about the connectivity for watching movies, but what about the other side? Can these network taxis become remote hotspots for Starlink Mobile? Similarly, regarding the Semi, I want to ask a question about the roadmap; when do you expect to consider the autonomous driving capabilities of the Semi? Because in terms of autonomous driving trucks, this market seems quite large, and given the current situation in our country, there is also a significant demand for autonomous driving trucks.

Musk: Yes, indeed. As you mentioned, there is a severe shortage of truck drivers. Therefore, there are not enough people willing to drive trucks, and trucks are crucial for transportation across the United States. So, an autonomous Semi truck will be very important in addressing the shortage of truck drivers. Obviously, this will greatly enhance safety and make the job of truck drivers easier. I believe an autonomous Semi truck will significantly improve the safety and comfort of drivers using Tesla Semi trucks.

Since the production of the Tesla Semi is still relatively low, even by the end of this year, its share in our total fleet will be very small, so we are focusing on the mass production of vehicles, namely Model 3 and Model Y, and are committed to achieving true autonomous driving for these vehicles and Cybercab, which is a very reasonable goal for achieving general unsupervised autonomous driving.

We expect to enable the Tesla Semi to achieve autonomous driving by the end of this year or early next year. However, I just don’t want this to distract us from our focus on achieving the "nine milestones" for autonomous driving safety in 3Y and Cybercab. So, for the next six months or so, the autonomous driving Semi truck project will take a bit of a backseat. But next year, as the mass production of the Tesla Semi progresses, it will definitely receive some attention.

In terms of communication, yes, I think there is an interesting angle here, that you can use the Starlink terminal in Tesla vehicles as a sort of cellular base station or connection tower, essentially providing relay services for mobile phones on the ground and anyone needing WiFi. But, I mean, that’s a different topic, but obviously, we can also achieve this through fixed Stall terminals.

Truist Securities analyst Will Stein: Elon, you previously teased us a bit about the supply chain issues with Optimus, emphasizing the difficulties involved. But we really want to know, do you lean towards producing these semiconductors in-house? When we think about the microprocessors, microcontrollers, actuators, and other semiconductors that will be used, it suddenly occurs to me that many of them are off-the-shelf. You could choose to design and produce some of these in third-party factories. You could also choose to design and produce them yourself. How should we view your tendency to procure these components in these different ways? Elon Musk: Yes. Optimus has a lot of very specialized power electronics and circuit boards. These are all designed by Tesla, and they are design-related, but they are manufactured by suppliers.

For the Optimus 4, which will be produced in Austin, this will be a more vertically integrated supply system. Our goal is to make the production of Optimus 4 at least an order of magnitude higher than that of Optimus 3. That is to say, we hope that the annual production of Optimus 4 can reach 10 million units, while the annual production of Optimus 3 is 1 million units. But there is an important issue here, which is that scaling up production is extremely difficult. So—yes, Optimus 4 will achieve a more vertically integrated production model. Therefore, we may complete a lot of PCV work internally.

Will Stein: As a follow-up question, first, perhaps you can talk about the expectations for Optimus 4, such as when production is expected to start. But the real follow-up question I want to ask is about the plan to upgrade from Hardware 3 to Hardware 4 in order to handle FSD version 15 and future versions of FSD. Is the plan to upgrade the hardware of the cars still in place?

Elon Musk: I think it makes sense to upgrade those cars with lower hardware configurations for almost all cars equipped with cameras; otherwise, too many modifications may be required. But for any car designed for cameras, it is economically wise to upgrade them at some point. I think we want to upgrade them to the next generation AI board, so we can have an upgraded AI chip, which is a moderate improvement over AI4, and it may go into production around the middle of next year. Then there is AI5, which we hope can achieve mass production around the middle of next year.

Artificial Intelligence (AI) will first be applied to Optimus, and in fact, progress is going very smoothly. The work of Tesla's chip team has been outstanding, making incredible rapid progress in AI. We are very excited about the design of Tesla's AI chip, and I think it will become the best edge computing chip in the world. If anyone has a better one, I hope that person can shake my hand because this AI chip is really amazing. So I think we are making great progress in chips. Thanks again to TSMC, Samsung, and Micron Technology for their support.

Barclays analyst Dan Levy: I want to ask about the pace and path of capital expenditures. Elon, you have said in the past that you would spend as much as possible, but in reality, efficiency seems to be the key factor. So, to what extent is your pace of capital expenditures in this capital expenditure cycle actually constrained simply because spending more would be done inefficiently? To what extent is the pace of capital expenditures indeed the main determining factor in releasing the different supply constraints you are currently facing? Musk: This question is very valuable. In fact, what I ask the team to do is to invest in capital expenditures as soon as possible while avoiding waste. Therefore, we are not pursuing extremely high capital expenditure efficiency, as that would slow down progress. So, it requires finding a balance between capital efficiency and time. If we can complete tasks faster, then it's okay if capital efficiency is slightly lower, because that will actually bring higher net present value (NPV) returns to the company.

So overall, I am very satisfied with the current situation; we have achieved amazing construction and production growth simultaneously in so many different areas. I believe no company has ever been able to do this on such a scale. Perhaps when Henry Ford was producing the Model T or similar products, the company scaled production proportionally, but that was during World War II when companies had to scale up production. I think this might be the fastest expansion of industrial scale in the U.S. since World War II.

Vaibhav Tania: Yes. On this point, I would like to add a little. Basically, if you look closely, all our capital expenditures are aimed at building productive assets. Correctly, we are essentially scaling up factories, whether it's for Optimus, Cybercab, or the LFP factory that started operations earlier this year, or the semiconductor factory, and we even plan to build a semiconductor manufacturing plant there. In fact, we are working hard to conduct a large amount of solar manufacturing in the U.S.

Musk: By the way, the expansion potential of domestic photovoltaic manufacturing capacity in the U.S. is severely underestimated by the market; we will increase domestic photovoltaic capacity by an order of magnitude. All expansion projects are extremely challenging, and with multiple businesses being launched simultaneously, we can only push forward at full speed to validate the real market performance of the products. At the same time, I believe the return on investment efficiency of current capital expenditures is extremely high.

Dan Levy: The second question is about energy storage. First, will the energy storage business continue to be constrained by supply chain capacity in the short term? Secondly, the current demand for energy storage installations is divided into two categories: utility-side peak shaving and valley filling; and data center-side stable power fluctuations (data centers experience significant fluctuations in computing load, and Tesla's energy storage hardware and software have unique advantages). What is the current demand proportion for these two categories?

Musk: The application scenarios for energy storage are not just peak shaving; there is also grid balancing. Therefore, this is not just a simple fact, but to balance the power generated by wind and solar energy, batteries are fantastic. I mean, solar battery combinations will be the way the vast majority of energy will be produced in the future world. Of course, all Starlink satellites are powered by solar panels and batteries.

So, you can think of us as a giant satellite. So far, the power provided by the sun far exceeds anything else. We believe that power limitations will be a major issue for artificial intelligence. The demand for AI computing is so high that even people with super large-scale resources encounter difficulties in opening AI computing and finding power, and then smoothing power with smooth power, especially during training runs, where power can fluctuate sharply in a very short time; during training runs, power consumption can drop by 70% in 100 milliseconds. You really need advanced power electronics that can respond quickly to smooth these huge power changes, especially during training runs This is why SpaceX has purchased so many giant data packets for data centers. In fact, this is mainly to smooth out the power for training runs. If you tell utility companies that they don't need to supply power during the worst times of the year or the worst days because batteries can handle it, it will make it easier for them to supply power to you.

In fact, what I mean is that the best way to expand total energy production in the United States may be to use batteries, because I believe the power generation in the U.S. is about 1.2 or 1.3 terawatts, while the average consumption is only 0.5 terawatts. So, your power generation is 2.5 times the average consumption. This means that, with just batteries, you could potentially double the energy output of the United States. So really, we see a very, very high demand for Megapack in the future