Prepared remarks
Good afternoon, everyone, and welcome to Tesla's Second Quarter 2026 Q&A Webcast. My name is Travis Axelrod, Head of Investor Relations, and I'm joined today by Elon Musk, Vaibhav Taneja and a number of other executives. Our Q2 results were announced at about 3:00 p.m. Central Time in the update deck we published at the same link as this webcast. During this call, we will discuss our business outlook and make forward-looking statements. These comments are based on our predictions and expectations as of today. Actual events or results could differ materially due to a number of risks and uncertainties, including those mentioned in our most recent filings with the SEC. During the question-and-answer portion of today's call, please limit yourself to one question and one follow-up. Before we jump into Q&A, Elon has some opening remarks. Elon?
Thank you. So, yes, it's been a great quarter. We achieved record Q2 deliveries. Model Y, I believe it is now, I think it's the best-selling car of any kind in the world and is setting records across the board. So its popularity is increasing tremendously. And we're seeing in locations that have FSD approved, a very high take rate for FSD. In fact, for a lot of people, they're actually buying Tesla Full Self-Driving with a car attached as opposed to a car with FSD added later. They come into our stores in the U.S. and they want Full Self-Driving with whatever car they choose. So, clearly, this is a significant demand driver. As we get approval for FSD in different countries, I think we'll see a similar uptick in demand. So, yes, it's looking very good on that front. The energy business is also growing incredibly fast, and I think it will be crucial for the scale-up of artificial intelligence data centers. We're investing a lot in growing the core business and really preparing for the future. So this is a massive CapEx year, but I'm confident that all the things that we're investing in will yield incredible returns, really the best CapEx returns that we've ever seen. The Tesla team has performed incredibly well. The Cybercab has started production, and we also have started production and are ramping in — we will soon start production with Optimus. We have started production with the Tesla Semi-truck. We will soon start production with Megapack 3. We started production with the lithium refinery and the cathode refinery. We obviously are scaling up battery cell production. We're also preparing to do a massive solar cell and panel production. Actually, this is going all the way from silicon refinement to producing the solar cell and then deployment of solar because there is going to be tremendous need for electricity in the future due to electrification of transport and AI, obviously. So we're working on what we believe is the most ambitious build-out of advanced infrastructure and manufacturing capacity ever in history. It's bigger. So, with robotaxi, our goals are very ambitious, but we do need to be cautious about causing any accidents or causing any harm to anyone. Although there are, I think, 30,000 to 40,000 automotive deaths per year in the United States alone, most of those do not generate any press — you never really read about almost any of those. But if we injure even one person, it will be worldwide headline news. Regulators will immediately clamp down on our activities. We don't want to injure anyone. So we're going as fast as humanly possible in scaling robotaxi while trying to ensure that we do not harm anyone at all and ideally do not even run over a pet. So that's really the constraint: we want to grow as fast as possible with robotaxi without harm to anyone. We provide live updates for robotaxi launch as we open up in a city. If you follow the Tesla account on X, we'll post immediately when we open in a city, and we'll keep you informed as we scale to new cities. We've opened up a number of cities in Florida and in Texas and, obviously, in the Bay Area. So we'll continue to scale, I think, very, very rapidly — it looks like more than 10% a week in terms of miles driven. So it's a very high compound growth rate. Optimus, as you've heard me say before, I think Optimus will be the biggest product ever, but it is a very complex problem to solve. It's one of the hardest things to solve: making an autonomous humanoid robot that can do tasks such that if you ask it to do something or show a video, it can perform the task without any programming. No one has achieved this. There are many challenges in the electromechanical design of the robot to achieve sufficient dexterity and reliability and long wear and tear — meaning it needs to be out in the field and not break down. Otherwise, you've got a 70-kilogram robot that just flops over and you have to carry it out like a body. We don't want Optimus to go haywire. So it's a lot of work to get the design right and to scale production. I really want to emphasize that the production scaling challenge is very substantial. This will be the hardest product to scale manufacturing that we've ever made at Tesla because everything on the robot is new. The difficulty of scaling the production ramp is proportional to the newness of the parts in the robot. With electric cars, the non-electric powertrain portions had an existing supply chain, so you could go to existing suppliers for wheels, side mirrors, windshield glass, body panels. With Optimus, there is no supply chain. We've had to build up the supply chain in its entirety or produce in-house. We actually have a tremendous amount in-house. The Optimus production line we're rolling out in Fremont in place of what used to be the Model S and X production looks incredible. It's quite stunning to see. I want to calibrate people correctly: Optimus will follow the normal S-curve of a manufacturing ramp, but the initial portion of the S-curve will be quite flat and long because of the newness of the parts and the robot. You have probably seen lots of impressive demonstrations of robots on the Internet, but those demonstrations are preprogrammed or remote controlled. There is no humanoid robot that is actually able to do generalized tasks. Optimus will be the first one capable of doing that, where it's not just a demo — it's genuinely useful in day-to-day life. Optimus is designed to have full human dexterity: a hand with the same level of dexterity, if not higher, than a human hand. The human hand is incredible. The more you study it, the more amazing it is. Optimus will have that capability: human and then superhuman dexterity. It's really going to be quite something. Regarding Terafab, we expect to announce the location soon and provide more details about our plans, which we will leave to a product launch announcement rather than try to squeeze into an earnings call because this is a very big announcement that deserves its own spotlight. Terafab is going to be an amazing initiative and necessary; without it, we will be constrained in our ability to scale Optimus production because we simply won't have enough AI chips. It's crucial to solve memory, logic and packaging to scale Optimus. We've placed equipment orders for our development fab in Austin. That development fab is intended to have lithography, mask production, logic, memory, packaging and chip testing all under one roof, enabling very fast iterative cycles to try out new chip designs quickly. I don't think such a building exists anywhere on earth. This will be very helpful as we try exciting high-risk, high-payoff bets on AI chips, primarily focused on Optimus. We have Digital Optimus, which essentially drives a computer screen in the same way you drive a car: pixels or photons in, controls out. We feel confident we can adapt the same Tesla AI technology developed for self-driving cars to drive a computer screen and achieve real-time video control at high frame rates. Digital Optimus will be important for physical Optimus because physical Optimus needs to operate computers. You can't come up to a touchscreen and not know what to do. It has to have generalized touchscreen and computer use capability. This is in partnership with SpaceX. SpaceX acts as the manager of Digital Optimus, providing it with tasks and workflows. We're building out a Megapod design that has Tesla AI4 computers paired with x86. Pairing x86 with Tesla AI4 in a box runs Digital Optimus. In a Megapod, like Megapack packaging, we place many AI4+x86 combos in a giant box. These boxes can be placed anywhere in the country or outside the country. This allows us to scale AI compute using disaggregated electricity production because there are lots of places around the world, including our Superchargers, where power is available. I think we have something like 7 gigawatts of power at our Superchargers and growing. We can place Megapods at many of these Superchargers and have distributed power for AI. That could be a big deal. In conclusion, we're super excited about our autonomy and robotics roadmap. There's so much coming. It's hard to squeeze into an earnings call, but we'll have a lot of product announcements. This is going to be a great year for Tesla, one of our best years ever, and next year will be even better. I want to thank the Tesla team for their excellent execution and all of our supporters for joining us on this journey. Thank you.
Great. Thanks so much, Elon. And now Vaibhav has some opening remarks.
Thanks, Travis. So, Q2 continued the trend we saw at the end of Q1: a resurgence in demand for our vehicles across the globe. We achieved record Q2 deliveries globally with sequential growth across the Americas, APAC and EMEA of 60%, 27% and 12%, respectively. Additionally, Model Y set records in several key markets, including the Netherlands, Australia and New Zealand. The efforts of the Tesla team to create compelling products at a reasonable price with autonomous driving capability are coming to fruition. We exited Q2 with our largest order backlog since 2023. We are therefore focused on increasing production at all our factories to meet this growing demand. Production growth will be limited by our supply chain, which includes not just batteries but also electronic components. We've been here before, and like always, our Tesla team is working actively to unblock the obstacles ahead by securing strategic deals with suppliers. Like Elon mentioned, one of the key factors for vehicle demand has been FSD. Our sales data suggests one of the main reasons customers are coming and looking at the car is because of FSD. In Q2 in North America, about 55% of our deliveries had FSD subscription enabled at the time of delivery. Overall, FSD attach rates continue to improve, reaching nearly 1.5 million paid customers globally, of which 55% are upfront purchases and the remaining 45% are subscriptions. We expect the bulk of growth in FSD monetization will come from subscriptions as we removed the purchase option in most markets. Automotive margins, excluding regulatory credits, declined sequentially from 19.2% to 16.3%. As a reminder, we had highlighted in Q1 that we had a $230 million benefit from warranty true-downs and some tariff relief, which did not repeat in Q2. Controlling for the impact of those benefits from the prior quarter, our automotive gross margins, excluding credits, would have been approximately flat. This demonstrates the effective pricing and cost management done by the Tesla team as we continue to drive demand. Commodity price increases and interest rate changes continue to add to our costs. Note that interest-rate subvention costs are recognized upfront as an offset to revenue. As interest rates have risen this year, the cost of subvention has risen along with them, which had a negative impact on automotive margins. As we have previously noted, the energy business is inherently lumpy. Deployments are tied to customer timelines and largely out of our control. In Q2, we deployed 13.5 gigawatt-hours of energy storage, a 53% sequential increase and our second-largest quarter for the energy business. Despite this growth, energy gross margins declined from 39.5% to 20.4%. This margin decline resulted from several dynamics. First, there was a warranty true-up in the quarter of about $240 million related to certain vendor cell issues for our legacy deployments. Also, the tariff benefits we recognized in Q1 of more than $200 million did not repeat. Lastly, we had previously guided that ASPs for industrial storage are coming down amidst growing competition. Long term, we believe the energy business should normalize at a gross margin rate in the mid- to low-20% range. Our order backlog for this business is robust, and we are building based on both existing demand and future demand we expect from data center growth and overall electrification of the economy. Service and other margins improved sequentially from 9.2% to 14.1%, an all-time high. This was mainly driven by increased volume and better cost management of our vehicle fleet that supports businesses such as used cars, Supercharging, service centers and insurance. While marginal today, this business also includes deliberate investments we are making in infrastructure that will help scale robotaxi in the future. We continue to grow the Tesla Robotaxi fleet and have expanded to a total of seven markets in the U.S. We expect the ramp of the fleet to accelerate throughout the year, along with expansion into new U.S. markets. As previously guided, operating expenses increased sequentially. The increase came primarily from significant research and development-related activities, including preproduction ramp costs for new products like the Semi truck, Optimus, Cybercab and other AI initiatives, as well as appreciation for additional compute we brought online. Additionally, we had charges related to litigation expenses in the quarter. Note that we are in a big investment cycle and expect operating expenses driven by R&D to continue to grow in 2026 and beyond. Net income was positively impacted by a mark-to-market gain of $1 billion on our SpaceX holdings, which was offset by losses on FX of approximately $300 million and on Bitcoin of about $100 million. As previously guided, our free cash flow ended up negative for the quarter. Most of the reason for it going negative is because CapEx more than doubled sequentially, and we expect it to increase further in the second half of 2026. We continue to expect CapEx for this year will be more than $25 billion. CapEx will grow for the next two to three years as we expand our robotaxi fleet, expand production capacity for Optimus, make investments for a semiconductor fab, install solar manufacturing capacity and AI compute infrastructure, in addition to other manufacturing expansions for automotive. In addition to using our cash for such investments, we are being opportunistic in securing certain debt facilities that will give us the capacity to borrow up to $30 billion to help accelerate such investments. We believe this is the right strategy to position the company for the next era, and we'll always make such investments in a very capital-efficient manner. The path to amazing progress is ever challenging and requires making bold bets. Our progress will not be linear. The future is going to be great, and we are ready to rise to the occasion. I would like to end by thanking the Tesla team, our customers, investors and suppliers for their confidence in us on this journey.
Thanks so much, Vaibhav. As everyone on the call may have noticed, we've covered several of the say.com questions already in opening remarks, including Optimus production, Terafab and Digital Optimus. Ashok is joining us today, our VP of AI, and he's going to cover questions related to robotaxi. Ashok?
Thank you, Travis. Many of the questions on say.com relate to robotaxi deployment and scaling, so we thought we'd provide a comprehensive view of where we are now and where we're going in the future. First, I'd like to state that the robotaxi program has been operating extremely well, especially in terms of safety. The program has an impeccable safety record. We have driven more than 380,000 miles of unsupervised robotaxi now across six cities in two different states. We have had zero notable incidents. Any reports have been of other actors impacting us when we were stationary. Again, I'd like to emphasize how safe the operation has been so far: zero notable incidents over 380,000 miles. Historically, so-called experts have always claimed that you need LiDAR, radar, HD maps and the entire kitchen sink to drive safely. We show that is not true. You can have safe, comfortable and affordable autonomy with just cameras. This record should be a huge validation of Tesla's entire AI approach. Second, I'd like to discuss scaling. We started the robotaxi program roughly a year ago in Austin. We had safety monitors in the passenger seats of the cars back then. Around the end of last year, we had the first fully unsupervised robotaxis in Austin. Since the beginning of this year, we have grown at double-digit growth rates in the number of unsupervised miles the fleet drives every week. We have grown at such a high compounding rate on a week-over-week basis over the last several months, and we expect to continue growing at such a large rate through the rest of this year. Expanding to new cities has been relatively less effort on our part. We expect the time to launch in a new city will continue to trend toward zero, toward an end state where we operate in entire states as a whole instead of city by city. Such growth was only possible because we have a world-class engineering team on frontier AI development here at Tesla. The currently operating robotaxi fleet is already running early versions of the v15 FSD software we referred to in the past. For v15, we planned roughly seven major improvement tracks, and they're all happening in parallel. The early v15 builds running on robotaxi have already merged about 40% of those tracks, and that's what's running in the fleet now. As we continue to complete work on v15, the car will be extremely safe and capable. We're already seeing internal signs that our technology bets are starting to work, and we expect the payoff will be huge; that's what will help us sustain the growth rate through the rest of this year and beyond. Finally, regarding the Cybercab: it's a phenomenal product. Anyone who rides in it instantly falls in love with the experience. We have aligned our manufacturing targets to roughly match the projected growth rate of the unsupervised miles. The same v15 models that power Model Y and other platforms will also work on Cybercab. In conclusion, robotaxi growth so far has been literally exponential while keeping an impeccable safety record. Our technology bets are paying off, helping us sustain exponential growth for the foreseeable future, and the Cybercab is going to be awesome.
That's great. Thanks so much, Ashok. And there was one remaining question on data collection for Optimus training. Would you mind covering that as well?
Yes, definitely. Regarding Optimus, especially Gen 3, Optimus has been designed to not just match the appearance of humans but also the functionality and dexterity of humans. One of the main reasons for this approach is that having the human form factor and function allows us to learn from humans how to perform a wide variety of tasks. This opens up the entire world to provide data for training Optimus. Just like FSD, we have access to a broad fleet of humans giving us data from all of the workers at our factory. Optimus can learn quite a bit from observing them and performing their tasks at the factories. We also have a dedicated data collection team who can provide a relatively small amount of very high-quality demonstrations for post-training of these models. This is on top of any training we can do from data across the Internet that shows demonstrations of people doing various tasks, both in industrial settings and home settings. The second flywheel is when we have a large number of Optimus robots practicing tasks in what we call the Optimus Academy. The data from the bots experiencing the tasks themselves will be invaluable and help us close minor form-factor gaps that may exist between the bots and humans. This is when the reinforcement learning loop kicks in: the bot initially attempts tasks, fails sometimes, learns from both successes and failures, and eventually masters those tasks, perhaps at a superhuman level. Our AI strategy for Optimus is aligned with the same end-to-end strategy that drives FSD: pixels in, controls out. Just like FSD, we expect it to work broadly. In FSD, you can get in the car, type in an address, hit start, and it handles all driving from park to park. The same will be true for Optimus: you'll ask it to do anything and then it performs the entire task on its own without you intervening. It's the same team that produced FSD versions v12, v13, v14 and now v15 software that now works on Optimus as well. I'm very confident Optimus will be an extremely capable robot.
Great. Thank you so much, Ashok. So that covers the say.com questions. And now we will move over to analyst questions. The first question comes from Tom from RBC. Tom, please feel free to unmute yourself when you're ready. Tom, you're still on mute. So, please unmute yourself and go ahead and ask your question. All right. We're not getting any audio from Tom at the moment. So we will try to move on to Andrew, and we'll add Tom back to the queue to come in next. So, Andrew, as soon as you can, please unmute yourself and ask your question.
Questions and answers
So maybe just to start out on Optimus. Elon, you mentioned that there's really no supply chain today for Optimus, so you've had to mostly in-source a lot of that yourself. I'm just curious: as you've gone through this exercise, are you seeing any suppliers willing to invest alongside you to set up domestic manufacturing to potentially help you scale more quickly in the U.S.? I'm asking in the context that you want to control your own destiny to some extent and ramp as quickly as possible, but from an investment cycle standpoint, it could be more capital efficient to have some partners help you along the way. How are you thinking about that and some of the dialogue you've had with potentially external suppliers?
Elon, you may be on mute right now. We can't hear you.
Sorry. Yes, our suppliers have been great, and they are making tremendous investments in support of Optimus and robotaxi. Samsung and TSMC, in particular, are building fabs — TSMC in Arizona and Samsung in Texas — and putting in tens of billions of dollars to build AI compute for Optimus and robotaxi. Panasonic has also invested many billions in increasing battery cell production. I'm a little under the weather here, so I sound a little strained today. But yes, our partners have been great, and I'd like to thank them for their support, investment and hard work.
Great. Really appreciate that color. And maybe just my follow-up on robotaxis. You gave a lot of great information in your prepared remarks, but there seems to be some evolving regulations at the state level, potentially around sensor requirements. As you think about your rollout, what do you want to see or need to see from regulators to make sure that, while some regulation might be good, it doesn't become overregulated? What are you expecting to see and what would you like to see to make sure this is done properly?
And we've got Lars on the line. Lars, if you wouldn't mind covering this one, if you can.
Yes. Just to add to Elon's point, the Samsung fab that's going to be pretty significantly dedicated to future projects is a massive investment, multibillion-dollar. We're seeing the same level of investment going to memory and also to new specific technologies like metal injection molded parts, flexible printed circuits and all sorts of nontraditional technologies that are more based for the robot as opposed to the traditional vehicle supply chain we've had. In certain cases where we don't find a great partner, we've never hesitated to in-source it. We have a very capable manufacturing engineering team and design team that helps us scale those things.
Yes. I'd also like to thank Micron for giving us the memory allocation. They're going to make some very tough decisions on memory allocation, and we really appreciate Micron making room for Tesla in the years to come and giving us a very significant allocation on reasonable terms given the pricing environment for memory these days.
Great. Really appreciate that color. Thank you.
Sorry, Andrew. We're going to have Karn hop in here for additional color briefly.
Yes, sure. I think when we talk about regulation, particularly in the U.S., we've made a lot of headway with good initiatives from the federal government in terms of FMVSS rules, moving toward acceptance and adoption of purpose-built AVs, and I really appreciate the support from NHTSA on that. When it comes to state-to-state differences, obviously the situation in New Jersey is a little disheartening. Ultimately, as Ashok said in his opening remarks, we focus on the performance of the vehicle. Ultimately, that's what will drive regulators and the public to adopt it. I think the best regulations are those that provide companies and innovators with a goal or a task and allow us to figure out the solution. When regulations prescribe a solution before identifying the problem, we don't appreciate those. As Tesla has always done, we'll let our performance speak for itself. Ashok laid that out clearly, and that will be the foundation of our expansion moving forward.
We're going to try Tom again from RBC. Tom, can you unmute and try to speak please. Unfortunately, Tom, we see you unmuted but we're not hearing you. So we're going to send you back into the queue. Next is Alex from Bank of America.
Just a few follow-ups on robotaxi. What milestones should investors be watching from here that would cause you to accelerate fleet scaling or market deployments? Is it further safety validation like miles driven per incident or something else? And would you ever consider third-party distribution partnerships, such as with rideshare providers, to increase utilization? Or is the plan for now to keep robotaxi fully vertically integrated?
Yes. We expect to be vertically integrated with robotaxi as we are in the rest of our business. I don't think we're going to have demand challenges with robotaxi. The economics will be so compelling that I think demand will outstrip our ability to service it. In that situation, vertical integration makes sense. It's really about going through what we call the 'march of 9s' of reliability: you need very high reliability to scale. Ideally, you want 99.999% reliability. That's the march of 9s. I think that's the only thing really constraining our growth in robotaxi.
Great. Thanks, Elon. The next question is going to come from Colin at Wells Fargo. Colin, please unmute yourself when you're ready.
There's been a lot of chatter about SpaceX and Tesla combining in some way, and I know there's a lot of collaboration already today. Elon, do you see synergies eventually from combining the companies? Does that make sense from your perspective over time?
Well, as you can tell from the many collaborations on so many fronts with SpaceX, there's more and more overlap, especially with Terafab — that's going to be a gigantic project. But obviously, we can't talk about combining companies on an earnings call; that has to be done with the appropriate process. With that, I'll turn it over to Brandon, our General Counsel.
Thanks, Elon. That's exactly right. We continue to benefit from our relationship with SpaceX; they've been a great partner, and we have numerous beneficial transactions with them. Earlier this year, we deepened our relationship through an investment and a framework agreement. This will allow us to continue to work with them on projects Elon mentioned, like Terafab and Digital Optimus.
Yes. There are many other things too. Grok in the car and Grok helping drive Digital Optimus, Starlink being integrated into Cybercab and Starlink being integrated into all our vehicles — at least in markets where Starlink is active. For a robotaxi, you need coverage everywhere. There are many places, even in Silicon Valley, where cellular coverage is terrible or nonexistent. I know when I drive to work for the first 10 to 15 minutes, I sometimes can't make calls because cellular connectivity is so bad. We can't have robotaxis getting stuck in 'Bermuda triangles' of lack of connectivity. Starlink, with its ability to provide connectivity anywhere, is important so robotaxis do not go missing in action. People in the car will want to do high-productivity work or entertainment. With Starlink, you can watch 4K live sports in the car at a low cost per gigabyte of data that's not feasible via cellular. There are many other situations where this helps.
Think of it like Ashok mentioned about Cybercab: the experience is great. When you don't have to focus on driving, you have all the time in the car to do a conference call, watch a movie or anything else. That's why Cybercab design has a big screen. We've started doing Cybercab rides in our factory in Austin. When you go through that experience, soon customers will be able to do it in the not-too-distant future, and then you'll understand why connectivity becomes so important.
Great. And Colin, do you have a follow-up?
Yes. If I follow up on robotaxi: when I look at launches, you have been adding cities, but the number of units appears to be in the dozens based on media reports, rather than hundreds. Why not scale up Austin or one or two cities before adding cities? What do you need to get higher vehicle counts in a major city? What is the roadblock to adding more vehicles on the ground?
The reason we've been expanding across different cities instead of doubling down on a single city is we want to ensure our stack is very general. We want to prove to ourselves and others that it works across many cities without too much effort per city, and that's what we see internally. As Elon mentioned, the growth rate has been literally exponential — it's in the early part of the exponential, which is why it's hard for others to comprehend. In terms of miles versus vehicles, our robotaxi vehicles operate continuously, as opposed to human drivers who use vehicles a couple of hours a day. These vehicles are in mostly continuous operation, which means even a few vehicles yield many miles. That's why we track unsupervised miles rather than just number of vehicles. We aim to operate an efficient fleet, and all the work that improves efficiency means we get many miles per vehicle.
To add further, we're trying to scale in a way that identifies different kinks you might encounter. There are things not just on the software front but also on the operations front that we need to tackle. That's why we want a broader footprint to work out these issues with a smaller fleet in a controlled manner before deploying at very high scale.
One point to note about Cybercab is because it's a new chassis, we need to accumulate driving data specific to Cybercab before we can put many of them on the road. Unlike Model 3 and Model Y where we have millions of vehicles on the road, Cybercab is new. We have to accumulate miles with Cybercabs retrofitted with steering wheels and brakes and acceleration systems to calibrate to the Cybercab chassis. As we gain confidence, the number of Cybercabs in cities will increase dramatically.
Great. Our next question comes from Walt from LightShed. Walt, please unmute yourself.
Just a quick follow-up on Lars' comment: the NHTSA administrator was recently on CNBC and seemed to suggest removing pedals and perhaps even steering wheels. That seems like it would be a gating factor. Are there other federal actions you need to really unlock your ability to ramp Cybercab?
Short answer: no. We have a great relationship with NHTSA and Administrator Morrison. I think they're trying to stay ahead of what the public wants and make sure they're doing the right things. We've been open with them for the last couple of years about our plans. I wouldn't say we're in lockstep, but I feel like we have a partner there and we're working together.
Got it. Elon, roadmap question: Starlink is integrated in Cybercab; can Cybercabs become remote hotspots for Starlink mobile usage? And a roadmap question on the Semi: when do you anticipate looking at autonomy for the Semi, given the sizable market opportunity for autonomous trucking?
Yes, absolutely. There is a serious shortage of truckers, so autonomous Semi trucks are important to address that shortage. An autonomous Semi will also improve safety and comfort for truck drivers who use the Tesla Semi. Because the number of Tesla Semis is still low and will remain a small percentage of our total vehicle fleet in the near term, it makes sense to focus self-driving efforts on our high-volume vehicles: Model 3 and Y and Cybercab. We expect to get self-driving working on the Tesla Semi probably around the end of this year or early next year. But I don't want that to distract from getting generalized unsupervised self-driving working on 3/Y and Cybercab — the 'march of 9s' of safety is our priority. For communications, yes, there's an interesting angle where you could use the Starlink terminals in Tesla cars as relays or connectivity towers to provide Wi-Fi to nearby devices. We could do that with stationary Starlink terminals as well.
All right. Our next question is going to come from Will at Truist.
Elon, you teased us a little about the supply chain relative to Optimus, highlighting how difficult it's going to be. I'd be interested to understand your propensity to manufacture semiconductors for Optimus yourself. For microprocessors, microcontrollers, actuators and other semiconductors, some are commercially available and could be sourced from third-party fabs, or you could design and make them yourself. What should we think about your propensity to source these components via third parties versus designing and making them yourselves?
Optimus has a lot of specialized power electronics and circuit boards. It's all Tesla design, but fabrication is done by suppliers. For Optimus 4, which will be built in Austin, that will be a much more vertically integrated supply system. We aim for Optimus 4 production to be an order of magnitude larger than Optimus 3 — aspirationally 10 million units a year versus 1 million units for Optimus 3 — but that's intensely difficult to scale. Optimus 4 will be much more vertically integrated, so we'll probably do a lot of PCB and related work in-house for that.
As a follow-up, could you talk about timing for Optimus 4 production? Also, is there a plan to upgrade Hardware 3 to Hardware 4 to process v15 and future FSD versions? Will you do hardware upgrades on cars?
I think it'll make sense to upgrade cars that have less than Hardware 4, particularly for cars with cameras, because otherwise it would be too many modifications. Anything set up for cameras will likely be financially sensible to upgrade to the next-generation AI board. We are working on a next-generation AI chip that is a moderate improvement over AI4 that should reach production around the middle of next year. AI5 is also targeted for volume production around the middle of next year. AI will initially go into Optimus, and the Tesla chip team is making very fast progress. I'm excited about the Tesla AI chip design; I believe it will be the best edge computing chip in the world. I'd like to thank TSMC, Samsung and Micron for their support.
Great. We are coming up on the hour, but we're going to try to squeeze Dan in from Barclays. Dan, please unmute yourself.
Great. Appreciate it. I'd like to ask about the pace and path of CapEx. Elon, in the past you've said you'll spend as much as you can, but efficiency seems like the gating factor. To what extent right now in this CapEx cycle is the pace of CapEx effectively capped because spending any more would be inefficient? And to what extent is the pace of CapEx the primary determinant in unlocking different supply constraints?
Good question. I've asked the team to spend on CapEx as fast as we can without it becoming too wasteful. We're not trying to aim for extreme capital efficiency if that slows things down. There's a balance between capital efficiency and time. It's okay to be a little less capital efficient if we get things done sooner because that tends to be the higher NPV outcome for the company. Overall, I'm pretty happy with how things are going. We're doing an incredible amount of construction and production growth in many arenas simultaneously. I don't think there's ever been a company that has done this scale of industrial build-out, maybe proportionally back when Henry Ford did the Model T or during World War II, but I think this is the fastest industrial scale-up since World War II in America.
I'll add a little on that. All our CapEx is focused on productive assets — scaling factories for Optimus, Cybercab, LFP factories that went into operation earlier this year, the Semi factory, and putting in semiconductor fab capacity as well as solar manufacturing in the U.S. The amount of solar manufacturing we're bringing to the U.S. will increase by an order of magnitude. None of these things are easy because you have to essentially start from ground zero to build a factory. We end up acting as general contractor for almost all our construction facilities because we are doing so much construction. We are moving very rapidly and doing a lot of things at once.
Yes. Our CapEx efficiency is good and scaled.
And then, Dan, did you have a follow-up that we can wrap up with?
Yes. Maybe just a quick question on energy storage. To what extent is energy storage going to be supply-constrained for the foreseeable future? And to what extent are current deployments for typical peak shaving for utilities versus addressing power quality issues for data centers, given the sharp fluctuations in power use? Do you see an advantage given your software or power electronics?
It's not just peak shaving; batteries are crucial for grid balancing, especially with wind and solar. Solar-plus-battery will be how the vast majority of energy in the world is produced in the future. Power constraints are a major issue for AI: turning on AI computers requires a lot of power, and hyperscalers struggle to find and smooth that power. Training runs can cause dramatic power swings — power consumption can drop by 70% for 100 milliseconds during a run — and you need fast-acting advanced power electronics to smooth those changes, especially during training. That's why SpaceX bought many Megapacks for data centers: mostly for smoothing power during training runs. Batteries also enable tapping more power from the grid because utilities can rely on battery backups during the worst hours or days of the year, making it easier to secure more power commitments. In the U.S., there's around 1.2 to 1.3 terawatts of power production capacity, but average usage is only about 0.5 terawatts. With batteries, you could effectively increase usable output materially. We see demand for Megapack being very high in the future.
All right. That is all the time we have for today for Q&A. We appreciate everyone's questions, and we look forward to talking to you next quarter. Thank you very much.
All right. Thanks everyone.