Good afternoon, everyone. Good afternoon. My name is Zach Kirkhorn. I lead finance here at Tesla, and welcome to our Investor Day. For those of you here in the audience at our global headquarters in Austin, Texas, we welcome you. Thank you for being here, and thank you for traveling in. For those joining us virtually, thank you for being a part of the day today. As we reflect back on the history of the company, there's been distinct phases of product advancement, technology innovation, and rapid volume growth, the most recent of which has been the global expansion and localization of the Model Y program. Today, we wanna talk about the future. We don't wanna talk about this quarter or next quarter, we wanna go further out into the future.
We've divided today's presentation into three parts, the first of which we're going to go macro. What does it take to convert Earth to sustainable energy generation and use? The second, we want to talk about Tesla's contribution to that global need, and we're going to go function by function through the company, and you'll meet our entire leadership team, and we're going to get into the details of what those teams are doing as part of the broader goal. In the third part, we're going to bring it back up and talk about what this all means for the company as a whole. Before we get started, statements made in this presentation are forward-looking statements that are subject to risks and uncertainties. More details can be found in our written materials. With that, let's get started. Part one, Elon Musk and Drew Baglino.
All right. Master Plan Part 3. As Zach Kirkhorn was mentioning, the thing that I think is we wanted to convey probably more importantly than anything else that we talk about here is that there is a clear path to a sustainable energy Earth. It's not, it doesn't require destroying natural habitats. It doesn't require us to be austere and stop using electricity and sort of be in the cold or anything. The story, and I think that this holds together quite well, and we'll be actually publishing a detailed white paper with all of our assumptions and calculations, is that there is a clear path to a fully sustainable Earth with abundance.
In fact, you could support a civilization much bigger than Earth, than much more than the 8 billion humans could actually be supported sustainably on Earth. I'm just often shocked and surprised by how few people realize this. Most of the smart people I know actually don't see a clear path. They think that there's not a path to a sustainable energy future, or at least there's not one that is sustainable at our current population, or that we'd have to resort to extreme measures. None of this is true. We're gonna walk through the calculations for how to create a sustainable energy civilization. Yeah.
To set the stage, today, our energy economy, it's, let's be honest, it's dirty and it's wasteful. Over 80% of global energy, primary energy comes from fossil fuels, and only 1/3 of that global energy actually ends up delivering useful work or heat. This is the problem statement, but we're here to talk about the solution.
Yeah. It's like if, Some of this I'm gonna elaborate because there's a very wide range of technical expertise, out there from people who are like, you know, whatever, level nine wizards in this subject to people who do not do engineering at all. Like when, if you have a gasoline car, you're converting less than a third, often maybe only 25% of the energy in the gasoline is converted into motion. The rest is turned into waste heat. That doesn't do any good at all. There's a lot of energy required even to get the oil out of the ground, to refine the oil, and to transport the gasoline to the gas station.
When you look at all that for a typical gasoline car is actually going to be using less than 20%, fully considered, of the energy from the oil actually goes into motion. When I see people or when we see people doing calculations for what does it take to create a sustainable energy Earth, they assume that the same energy amount is required for an electrified civilization versus a combustion civilization. This is not true, the because most of the energy of combustion is waste heat.
Even to get the fuel to combust in the first place and get it to the end use, there's a lot lost along the way. I mean, this is the primary energy consumption, 165 PWh a year. Petawatt-hour is a trillion terawatt, trillion kWh . It's a large amount of energy, but the nice thing about electrified economy, there's a better way, we're gonna talk about it, is that. Through end use efficiency and through efficiency along every step of the way, actually the total energy use, it halves. This is one of the most enabling aspects of electrifying everything, is that the sustainable energy economy is that much easier to accomplish. It's actually half the problem statement of the fossil fuel economy.
Yeah, we're being conservative here, so it could be better than half. We're trying to have assumptions that are reasonable, not overly optimistic, in fact, slightly pessimistic. It's really better than half, but it's easy to make the argument that we need half as much energy with an electric economy versus a combustion economy.
Yep. how the master plan works.
Yeah. The thing that is needed is at very large scale that is not currently present is a vast amount of battery energy storage. Our rough calculations are that this is about 240 TWh or 240,000 gigawatt-hours. This is a lot of batteries, but it is actually a very achievable amount. We'll go into details on that.
Yep.
That's a combination of electric vehicles and stationary storage. If you've got solar or wind, you've got to store the energy when the wind is not blowing and the sun is not shining. we're assuming sort of an eight to one ratio of stored energy to power, so 30 TWh of power. That's 30 TW of power. Our actual capital expenditure calculation for manufacturing investment is more like $6 trillion, but we, you know, we made it higher to make it $10 trillion.
This is across mining, refining, you know, battery factories, recycling, vehicle factories, all the things that we're gonna talk about needing to invest in to build this sustainable energy economy.
Yeah. Now, if you look at the total world economy, it's just under 100 trillion. If this was spread out, say over 10 years, it would be 1% of the global economy. Over 20 years, it would be half a percent.
Very doable.
Of the global economy. This is not a big number relative to the global economy. As Drew mentioned, you need about half as much energy with an electric economy versus a combustion economy. In terms of wind and solar, how much land would be used, it's less than 0.2% of the land area of Earth. Like, generally people don't realize quite how much energy is reaching us from the sun. It's roughly a gigawatt per square kilometer. You know, the sun doesn't shine all the time, but it's if you multiply that by, say, four to get the continuous power, four or five , then that gives you the land area of solar. You can put wind and solar often in the same place.
A lot of places that currently have wind, you could have solar there, and you double your energy.
You can also put wind offshore. It doesn't even need-.
Yeah
to be on land, wind is even more flexible.
You could put solar offshore too.
Yep.
Earth is 70% water. Anyway, the point is that with a really remarkably small amount of Earth's land area, we can go fully sustainable. Yeah.
Do the resources and raw materials exist to support this transition? We'll go through that in detail, but we do not see any insurmountable resource challenges at all. In fact, in the end, we should be mining less ore to accomplish this economy than we currently do with the fossil fuel economy, and we're gonna talk through that.
Yeah. Just to emphasize that again, the electrified economy will require less mining than the current economy does.
Yes.
Less, not more. Okay.
This is the plan, and now we'll get into a little bit more of the details of the plan. Basically, five areas of work. first, renewable power, the existing grid. second, switch to electric vehicles. third, switch homes, businesses, and industry heating to heat pumps. fourth, high temp heat delivery and storage for high temp industrial and chemical processes, and a little bit of green hydrogen in there for chemical processes that need hydrogen. Finally, sustainable, sustainably fuel planes and boats. These are the five areas, and we're gonna go into detail on all of them.
Yeah, I mean, my personal opinion is that as we improve the energy density of batteries, you'll see all transportation go fully electric with the exception of rockets. That's awkward. You can make the fuel with CO2 and water. You can make methane with CO2 and water. In fact, that's-
You can do that with just electricity.
Yes, exactly.
Yeah.
In fact, on Mars, if we hopefully get there at some point, the atmosphere is CO2, and there's water ice throughout Mars, so you can take the CO2 and H2O and turn that into CH4, which is methane and oxygen. Ultimately, even rockets can be electrified.
First, repowering the existing grid with renewables, and this is gonna be a consistent theme. You'll see our estimates for the number of TWh, terawatts, and trillions of investment at the bottom of the page. You know, this is already actively occurring in front of us. 60% of the generation added to the U.S. grid was solar in 2022, and actually, on a year-on-year basis, solar deployment is growing 50% year-on-year as of 2022. this is a serious upswing, and if we continue this trend, this is gonna be behind us before we even know it.
Yeah.
Second, switching to electric vehicles. Again, 21% reduction in fossil fuel use by doing this alone. Obviously, Tesla is heavily engaged in this activity as, along with many others. overall EV production grew 59% year-over-year in 2022, and EVs hit an amazing 10% market share. I mean, it's an awesome milestone. I'm super excited to see that.
Yeah. I mean, this is obviously happening very rapidly, and I mean, I think really all cars will go to fully electric and autonomous. Riding a non-autonomous gasoline car is gonna be analogous to riding a horse and using a flip phone. That's basically gonna be the situation.
We actually took a somewhat conservative assumption here in terms of how many batteries are required, 'cause the more the fleet is autonomous, the fewer the smaller the fleet needs to be, just from a utility basis. We're not accounting for all of those benefits or really much of those benefits at all in this number. What does this fleet look like? You know, just rough view from our perspective. Of course, we could be wrong, but, you know, you can see the sort of breakdown of the fleet by millions of vehicles. You know, our goal is to do 20 million electric vehicles a year.
Yeah. Fewer vehicles will be needed, at least passenger vehicles, with autonomy. There's some debate as to what that number is, but it's some number less than the number of vehicles needed today. There's roughly 2 billion cars and trucks in operation in the world today.
Yeah, what we show here is actually, I think, only $1.4 million or so.
Yeah
We're $1.4 billion, I mean, or so. A smaller fleet, and you know, the numbers here in this presentation are around 85 million vehicles a year produced, just to give you a sense of how we're thinking about this. Again, we're gonna put all these assumptions up online and you know, encourage people's thoughts.
Yeah. We're basically heading rapidly towards an electric or autonomous future.
Exciting.
Yeah.
One of the reasons why EVs are so enabling is this end-use efficiency point. Tesla Model 3, it's 4 x more efficient from the well to wheel than a Toyota Corolla, that's all about the efficiency of getting the electricity to the car in a sustainable energy economy, and then how efficient the car is in transferring that stored energy to motion on the road. When compared to the engine in the Toyota Corolla and all the, you know, extraction, refining, transmission, distribution of the gasoline to the Toyota Corolla. Just for, this is a fun reference. Model 3 can drive over a mile on the energy it takes to boil a pot of water for pasta, and then it can drive another mile on the energy it took to cook the pasta.
That pasta's one pound. The Model 3 is, like, 4,000 pounds. Just to give you a sense of just, like, what, like, it really doesn't use a lot of energy to move a Model 3, that 4,000-pound object, down the road.
Heat is a lot more energy than motion.
Yeah. Yeah. but people, you know, you just boil a pot of water, you don't even think about it. You know? It's just interesting how efficient these cars are. Next, switching to heat pumps in homes, businesses, and industry. you know, right now, heat pumps meet 10% of building heating needs. Install rates growing 10% year-on-year. It really needs to accelerate. Heat pumps can serve, you know, heat applications up to 200 degrees Celsius in businesses and industry. From an investment perspective, as you can see on this page, this is actually the lowest hanging fruit in terms of displacing fossil fuels. you might be saying, "Well, what exactly is a heat pump?" heat pumps don't create heat, they move heat.
When you think about, like, the natural gas furnace in your house, like, it's, it is generating heat itself, but what the heat pump is doing is actually moving heat from outside of your house into your house. They're just an air conditioner or refrigerator in reverse. We're surrounded by heat pumps. There's, like, you know, they're all over this factory, they're in your house, and all this really is about bringing them to displace all the fossil fuel heating in all the homes, businesses, and in the industry that we can. From an end-use efficiency perspective, this is a 3 x reduction in the total energy required to heat these buildings. A real obvious thing to do.
Yeah, heat pumps. They're in our cars, yeah, now as default.
Yep
And at some point, we might make a heat pump for our home, so.
Yeah, maybe. Maybe.
Maybe.
Next, a little bit more detail on electrifying high temp, sort of industrial and chemical processes. Over 50% of industrial heat is greater than 400 C. We're, you know, cement, steel, fertilizers, chemicals, plastics, metals refining all need, like, 1,500 C, so we need a solution here. Ultimately, it's purpose-built equipment that enables electrification. You know, carbon, graphite is stable up to 2,000, 2,800 C. There's other options in the 1,500 C range, like silicon carbide, other materials. The idea here is you create and store heat when renewable power is available. If this is a sustainable energy economy, you know, renewable energy is intermittent.
Peak of the day, you've got more generation than you need, you make a bunch of heat then, you transfer that heat into the industrial process 24 hours a day using the stored heat you created when the sun or the wind was blowing. That's the concept here. On the hydrogen side, we also need green hydrogen to decarbonize metals and chemical refining processes. This is things like ammonia, making steel. You know, there's roughly 120 million tons of hydrogen sourced from fossil fuel today to do these things. Hydrogen can also directly replace coal, which is currently used in a ton of steel production through a process called directly reducing iron. You can replace blast furnaces with a hydrogen reducing direct reduced iron furnace.
This is the way to eliminate fossil fuels from these aspects of the economy and the CO2 associated with them.
I mean, some of this, there's like room to disagree, but some amount of hydrogen is needed for industrial processes. My personal opinion is that hydrogen will not be used meaningfully in transport, but it shouldn't be. If you're going to use a chemical fuel, you should use CH4 not H2. Nonetheless, it is needed for industrial processes and can be produced just by splitting water essentially.
I mean, something that's been done for decades and decades.
Yeah.
This is not, you know, rocket technology.
Yeah.
Lastly, a small part of the pie, but a necessary part of the pie is sustainably fueling planes and boats. Shipping accounts for 3% of global CO2. It's ripe for electrification. Even with a lithium iron phosphate, long-haul ships can be fully battery-powered, that's a great opportunity to electrify. Energy density is a little bit harder for planes, but short haul is doable today. With some improvements, we'll get long haul underway. Even in the meantime, we can leverage sustainable aviation fuels produced and stored using excess renewable electricity. There's a lot of work going on in this space, it's Yeah.
Yeah. I mean, it to really get long-range aircraft and long-range shipping to use lithium ion, you need to redesign the ship and not just.
The plane.
The plane. To take advantage of the fact that it is a new source of energy. It's. That's it. It's a different architecture. Just like with an electric car, you wouldn't just, you know, take a gasoline car and stick a battery in it. That's very suboptimal. It's much more efficient to have the battery be the structure of the car, and you know, make it mass efficient and optimized for batteries. If that's done with the aircraft, then you can get long-range aircraft with sales at around 450 watt-hours per kilogram, which you can buy it right now, actually. They're expensive, but I think that price will come down.
When we stack up all of these efforts, we end up with the numbers we shared at the beginning of the presentation: 30 terawatts, 240 TWh, $10 trillion. You're, you may be saying, like, "I need some context. Is this feasible?" Spoiler alert, it's entirely feasible. Just looking at it from a growth rate perspective, how much do we need to grow the deployment of these technologies? We're talking about only a 3x growth rate in solar and wind deployment. Solar is already growing at a breakneck pace, as is wind. This gap is gonna be closed really quickly. When we look at the electric vehicles, they have to grow 11x. They grew 60% year-on-year last year.
That gap is gonna close pretty quickly as well. Lastly, storage. You know, Tesla's energy storage business has grown at 65% CAGR since 2016. The global, you know, energy storage business is accelerating pace as well. I mean, all these gaps are gonna close, especially as this momentum of the transition to sustainable energy accelerates. Of course, our goal on this page is 20 million EVs per year and 1 terawatt-hour of stationary storage per year, basically as soon as we can. You know, what about this investment? How do I have a reference point on this investment? You know, Elon mentioned it's 10% of, you know, 1 year's world GDP.
Another way of thinking about it is how does it compare to like what we invested last year in the fossil fuel infrastructure. It's 60% of that investment. Actually building the sustainable energy economy is less than extending the fossil fuel economy from a year-over-year investment basis. Very doable. When we look towards, does this fit on the planet? Absolutely. Less than 0.2% of land. As a reference point, the total land area intensively farmed today is 12.5% of all land. I mean, you drive around, you see some farms, but you don't see them everywhere. This is an order of magnitude, more than an order of magnitude difference between farming and what we're talking about for sustainable energy land.
Yeah. It doesn't need to displace farmland.
Of course not, no.
You know, forests or jungle or any kind of ecological preserve. It can be used in very sparsely populated desert regions.
Barren areas.
Yeah.
Areas that are just not really fit for development or otherwise used, so.
Yeah.
I mean, 0.2% can fit into a lot of places.
Yeah. There's essentially no meaningful ecological impact. In fact, transition to a sustainable energy economy would result in a substantial reduction in current ecological impact.
That's a great way to put it.
Yeah.
What about on the mineral extraction side? This is a cartoon that sort of gives you a sense for all the ore and the, like, extracted minerals that are coming out of the earth every year. It's about 68 gigatons. Each truck is a gigaton. What does this look like when we're in a sustainable energy economy? Looks like that. The fossil fuel extraction disappears. We replace it with the materials required to fulfill the sustainable energy economy. It actually reduces. Now, it's not to say that we don't need to continue to explore, you know, bring on mining and refining for the sort of specific materials for the sustainable energy economy. We do. The investment and mass flows are all very achievable just looking at what is already happening on the planet.
Like this is nothing out of scale of what has been done and is already being done. And then we calculate it on a sort of, you know, element by element basis. The resources are there to support the transition. You know, this is cumulative demand to move in a sustainable energy economy direction until 2050 relative to USGS resources today. You know, we're not breaking the resource bank for any of these materials. And then when we look at what really happens as we move forward, history teaches the more we look, the more we find. What people think happens is, oh, there's this many resources, next year there's gonna be less 'cause we're gonna extract them. What actually happens is, as we extract resources, we find more.
You can see on the right what has actually occurred with the key materials to the sustainable energy economy. Since 2000, as the sustainable energy economy has been growing, and Tesla's been growing, and all the industries around us have been growing, the actual resource availability has increased, not decreased.
Yes. There still seems to be quite a bit of confusion about lithium. Lithium is extremely common. It's one of the most common elements on Earth. There's no country that has a monopoly on lithium.
Yep
There's enough lithium ore in the United States to electrify all of Earth if the United States was the only place producing lithium. There's enough domestic material to electrify Earth. It's very common. The limiting factor is the refining of the lithium into battery-grade lithium hydroxide or lithium carbonate. That's the actual limiting factor.
The same is true for these other materials.
Yes.
This is, again, these are not like crazy technologies. It's just the investments need to be made, and the investments, they're not gigantic, they just need to happen.
Right. Nickel's maybe the of them all the trickiest one to solve, but as we showed with the graph there, maybe need like 30% of the world's nickel, known nickel reserves.
The nickel reserves have actually grown since 2000.
Exactly. You only need nickel for basically aircraft, long range boats, and very long range cars or trucks. The vast majority of the heavy lifting for electrification will be iron-based cells. Iron is actually literally the most common element on Earth. Little trivia point, if you say like, "What is Earth made of?" By mass, it is most, it is made of iron more than anything else, and second, oxygen, and then everything else after that. Basically, we're a muddy rust ball is what Earth is. An iron cathode is sort of, we're definitely not gonna run out of iron. There's so much iron, it's insane. That's it, yeah.
Ultimately, you know, this resource extraction, we go through this effort, we build these batteries, and then we recycle these batteries. Ultimately, we're doing this to build this sustainable energy economy, but the maintenance amount of ore that we require is really an order of magnitude or more less because of recycling. In the end, a sustainable energy economy is within our reach, and we should accelerate it.
Yeah. I mean, this is really the main message of today. I really wanted today to be not just about Tesla investors who own stock, but really anyone who is an investor in Earth. The, what we're trying to convey is a message of hope and optimism. Hope, optimism that is based on actual physics and, and real calculations, not, it's not wishful thinking. Earth can and will move to a sustainable energy economy and will do so in your lifetime. Thank you.
Yes. Thank you. I just wanna welcome up Lars and Franz to the stage.
Hi.
Hi, I'm Franz. I lead design at Tesla.
I'm Lars. I've been doing vehicles with Franz for almost 13 years now.
I joined Tesla in 2008 to vertically integrate design into the company. It didn't exist before. It was a pretty small team. I was tasked with designing the most beautiful, innovative, and well-engineered vehicles on the planet. No small task in 2008. There was not a lot going on in the EV sector then. Since then, we've focused on constant improvement in cost, efficiency, innovation, things that you'll continue to hear about today, while continuing to design the most desirable cars. Today, we produce cars differently than we did 10 years ago, but the end result is always an exciting, futuristic, and desirable set of vehicles.
Back then, we only had a handful of designers and engineers like myself, but we had a great vision to radically change transportation.
Back in 2008, when we were designing Model S, we didn't have a factory. In fact, we had a really small engineering team and a tiny design team. That allowed design to lead all the conversations. It let us innovate forward-thinking ideas, like how do you fit seven people into a sedan? How do you make door handles disappear into the doors? Putting a huge touchscreen into the center of the vehicle, something that had never been done before.
We won Motor Trend Car of the Year.
Yes, we won Motor Trend Car of the Year in 2013. Our first, you know, great award, first car. Pretty good start. Kind of a home run, I think. That whole process resulted in a linear process that you see on the screen. We designed first, then we engineered, and then we figured out how to manufacture it.
Yeah, I mean, I think that's really important. When we were designing the car together, we didn't even know where we were gonna build it. We came up with.
We didn't even know Lars when we were designing the car.
That's true. Once we got Fremont, we were very fortunate, and we figured out the manufacturing solutions sort of like we were flying a plane and putting the wings and building the engine at the same time. We knew we had to do better.
Yeah, we knew we had to be better in order to scale. As part of the master plan that you've read, Model 3 needed to be a smaller, more efficient, and more affordable version of Model S, but had to be equally great. It had to have all the things that people loved in their Model S or Model X, but be much more affordable. So we approached the process a little bit differently than the first time around. Now we had teams that we all worked together, so we were able to combine design, engineering, and manufacturing process all at the same time. Somewhere along the way, we changed the manufacturing process to be fully automated.
We leaned into this whole new way of manufacturing a car, but we had already engineered it, so things didn't quite go as well as planned.
It was an amazing product, it landed us in production hell. Many of us who lived through that carry those battle scars. It was a great idea, it wasn't the right timing. Like Franz said, automating something that we designed to be built manually is super hard, and we have many failed examples of that at the Fremont factory that we ripped out. Some of them eventually still work. This is one I actually worked on with a small team of engineers. It's still running today, and some of the engineers that came by said we couldn't do it are no longer with the company, but it's running, and it's running faster than ever. We kind of self-imposed constraints on the design when we were doing it to be built manually, and we really didn't think about it.
Despite all that, Model 3 is the best-selling EV ever.
Yeah, Model Y, which is derived from Model 3, is about to pass that. We knew we had to improve the process further. With Cybertruck, we designed a vehicle around a vision that actually started with the manufacturing process, and in this case, the materials dictated the design.
Forming full hard stainless steel isn't rocket science, but it sure isn't easy, and it limited the way we could do it.
Absolutely. It really forced us to think about designing something in a way that you couldn't normally stamp panels, you couldn't form them in a traditional way. You ended up with very linear bending processes that are just not in automotive kind of language of manufacturing today. It actually created a very efficient and process and one of the most dynamic designs ever, I believe. It's definitely something that's gonna change the road landscape.
Hopefully, you guys saw it down there and you experienced it. It's definitely real. Those are real trucks. We're on our way to build them. What that stainless steel opportunity did for us, it has let us rethink the factory footprint. We don't stamp those. That's a huge part of it. We don't even paint them. Our footprint got smaller, and we started to think about innovative ways to take those constraints and make great products.
That constraint didn't really change the end result of the truck. It's a super dynamic truck, and it has all the functionality you would expect out of any of the other competitive trucks. The best thing about it's coming this year.
Ideally, after all that, we would design, engineer, manufacture, and plan for automation happening together. It gives us the opportunity to question requirements. This is something that is fundamentally only available at Tesla. The places I used to work and the top manufacturing companies in the world, they don't sit together.
Yeah, nowhere.
We are one team.
Nowhere I know has all these teams together thinking about these processes from the very beginning.
In fact, all of those engineering teams, manufacturing, design, automation, they're all in one org. They all report to one person. We can't point fingers at each other. We have to solve them together, which is the best way to innovate. The traditional way of making a vehicle is this: You stamp it, you build a body in white, you paint it, and you do final assembly. What's interesting is these shops are dictated by the organizational structures that exist, and they're dictated by the boundaries that exist in the factories that are laid out. If something goes wrong in final assembly, you block the whole line, and you end up with buffering in between. This is at the tail end of its manufacturing optimization. Henry Ford first invented this assembly line in 1922.
It's been 100 years, and it's really hard to make a change after 100 years. When you watch it happen, it's actually really silly to a guy like me. You take all these stamped panels, you put them together, then you put them in a framing station, you build a body that looks something like a car, you put the doors on, and then you paint them. Once you get the color, you take the doors off, and then you start putting the interior inside the car. It comes in through the openings that already exist. I wish it went in like this big piece like this-
Yeah
There's actually people coming in and out of the car. There's awkward, you know, movements. We lift the car up, we put stuff from underneath it, we put it down, then we put the seats in the car, and finally, we close it all up with glass, and we bring those doors that went away for a trip, and we put them back in the car. Most of the time, we're doing this with a big, giant car and moving it and doing really nothing to it at all.
What's funny, though, in this kind of whole process is that just recently Toyota just called this an engineering work of art.
True.
The Model Y.
That was humbling. At Tesla, it's not good enough. If we're going to scale the way we want to do, we have to rethink manufacturing again. As part of the Master Plan, we have to make another step change in cost. We started this on Model Y when we made these huge Giga Castings, and we deleted hundreds of parts. We simplified assembly with the Model Y structural battery, where we decided the floor should be a part of the car. The battery is the floor. We put the seats and the interior on the battery, and we bring it up through a big open hole, and we assemble it. This allowed us to do things in parallel, fully rethinking the process and reducing the final assembly line by about 10%. We thought maybe we could do this other places.
Yeah, I mean, in a way, the constraints become part of the solution rather than a problem.
When you think about what I'm trying to say, I really wanna hammer this home. When you have a car that's about 5 meters long, and you have people working around it like we did in Model 3, and we change that to this process where we take different parts of the car, and we can do more at the same time like we did with the Model Y structural battery pack. What you see here is us doing that on the front part of the vehicle or the rear part of the vehicle. That means we can get more people working on the car, or robots, working on it at the same time. That means we have better operator density, less time doing nothing. I call that space-time efficiency. It has nothing to do with quantum mechanics. We can have that conversation later.
We get 44% more operator density, which means more work, less time walking back to the station, 30% improvement in space-time efficiency. Because we're not building it in and out of the car with those slow movements of those robots I showed you from production hell, when we go to automate it's gonna be a lot easier. In the end, that will probably look something like this, where we balance parallel and series manufacturing in a way where we only do things that are necessary, with a much shorter final line blocking a lot less of the entire rest of the factory, so we can optimize material flow using the best practices.
What that means, it's going to look something like this, where we build all the sides of the cars independently, we only paint what we need to, and then we assemble the parts of the car once and only once. We put them where they need to go. The interior is attached from a bottom up or a top-down strategy, so there's more access for those robots and people. We aren't moving heavy objects around and doing nothing to it. It means we're doing more work on the car more of the time.
When we take it, all of these tested sub-assemblies and we put them together, we finally assemble the car only one time, putting the sides on with all of their parts to a front and rear that was already assembled, carrying the floor in with the seats and finally boxing it out with the doors one time, just like Cybertruck. In the end, you get the same car, but it's not gonna be a Model Y.
This is a Model Y for illustration, not the next-gen vehicle.
In the end, what does that mean? To increase the scale and adoption of electric vehicles on the orders of magnitude that we just showed you, we have to make constraints part of the solution. It leads us to greater than 40% reduction in footprint, which means we can build factories faster with less CapEx and more output per unit. Faster, less CapEx, more output per unit dollar. Zach's gonna go into more details on this later, but it also means through this innovation and some of what my other engineering colleagues are gonna talk to you about in the future, we'll reduce costs as much as 50%. This is the two-for-one concept you hear me and Elon talking about on earnings calls.
Yeah. I think our track record proves that we can deliver the best cars, and we deliver the best cars in spite of, because of these constraints. I'd love to really show you what I mean and unveil the next-gen car, but you're gonna have to trust me on that until a later date. I promise we'll always be delivering exciting, compelling, and desirable vehicles, like, as we always have.
Have we ever not?
Yeah.
We always do.
What else is Tesla known for?
Speed. With that, I wanna bring on Colin, who's in charge of making our cars go fast.
Thank you, Lars. Thank you, Franz. My name is Colin Campbell, I have the real privilege of leading powertrain engineering here at Tesla. We make the fastest cars that you can buy for the money, whether they're electric or gas. The Model S Plaid that we're looking at here, it has more than 1,000 horsepower, and pound per pound, the motors in that car are as powerful as jet engines. Our cars are super fun. People absolutely love driving in them. The other thing that you probably all know about our powertrains is that they're efficient. Our cars go 25%-30% further than other EVs in our class for the same amount of efficiency. At Tesla, efficiency means more than just reducing how much energy the cars use.
It's about how we develop, how we manufacture, how we refine, and how we scale the powertrain. The Model 3 and Model Y powertrain is a great example of this broader meaning of efficiency. Since we launched it back in 2017, we've continuously improved that powertrain and the factory that builds it. The drive unit, the engine of the car, is 20% lighter for the same power. We use 25% less heavy rare earths than when we started. The powertrain, the powertrain factory, which is behind me today, is 75% smaller and 65% cheaper than the one that we originally built. What I really wanna emphasize is that we did all of this without compromise. Our cars are just as powerful, they go just as far, they cost the same or less, and the factories have the same output.
How did we do that? We did it by designing the entire vehicle and the entire factory together as one company. This sets Tesla apart. We have small and highly capable teams. To make a critical decision, we can have the battery cell chemists, the mechanical engineers, the manufacturing engineers, the supply chain team, the automation designers, the software programmers, all in one room working together in real time. That allows us to make decisions that are best for the whole car and to make them really fast. That approach is unlike traditional automotive engineering, which is really fractured. If you were to go buy, like, a premium German electric car, the engineers who designed the drive inverter in that car, they did not work for that car company. They worked for a contractor.
At Tesla, we design the entire car and the factory that builds it. I wanna highlight a few examples of what we've been able to do in-house, thanks to that unique approach. Inside the charger of your Tesla are transistor packages, that's at the top of your screen here. Every electron that moves you down the road flows through one of these packages. We designed our own custom package, which is what you're seeing here, we can extract twice as much heat out of that package as what we could buy off the shelf. What does that mean? It means that the silicon carbide wafer that's inside those packages can be much smaller. Silicon carbide, it's an amazing semiconductor, but it's also expensive and it's really hard to scale.
Using less of it is a big win for us. On top of that, orchestrating all of these transistors and making them switch in the right ways is computationally extremely intensive. It used to require four microprocessors, which are shown here on the bottom left. We have developed our own custom microprocessor. It's purpose-built for high-power electronics. It's half the cost, and it does in just one the job of all those four. These are just two examples of many that I could use to showcase our expertise in high-power electronics. That expertise has allowed us to take the cost of the chargers that were in our Model S when we launched it in 2012, both the cost and the mass, and cut both of those in half. Even more important, power electronics are central not just to our cars.
They're also central to our Superchargers and to our energy storage products, and Rebecca and Mike will be talking more about that. In addition to the work that we've done in software and hardware, we've also done a lot of work in-house on software. This is the drive unit for Model 3 and Y, and if we take a cross-section, we see the stator and the rotor, and they're responsible for the core function of the drive unit, which is converting electricity into motion. Our custom software lets us simulate the rotating magnetic field that is responsible for that conversion. Getting that simulation exactly right, it's central to the cost, the weight, the size, and even the sound of the drive unit.
Now, you can buy software that will do all of this, but our tools are faster, and they are more accurate. That was not easy to do. That allows us to quickly iterate through millions of possible drive unit designs to find the best one. Wanna highlight one more area where Tesla really excels because we integrate work that is often farmed out. When you are making a new product, it's not enough to think about the product itself. You have to think about how you're gonna make it at scale. At Tesla, our powertrain and our powertrain manufacturing equipment is both designed under one roof. The engineers who are designing the motor, they are in the same room as the engineers who are designing the machine that's gonna put that motor together.
That collaboration pushes us from day one to design products that are not only high performance, but are really easy to assemble. All of this expertise that we have in the powertrain team, in hardware, in software, in manufacturing, it's gonna have a major impact on our next platform. In our next powertrain, those silicon carbide transistors that I mentioned that are key component but expensive, we figured out a way to use 75% less without compromising the performance or the efficiency of the car. Of course, we know that battery cell supply is one of the constraints on the scalability of EVs right now. Our new powertrain is compatible with any battery chemistry. That will give us great flexibility in battery sourcing. If we wanna make EVs more accessible to people, they have to be cheaper.
We've reduced the drive unit cost to about $1,000. We don't think any other automaker is even close to that number. Finally, the bigger a factory is, the longer it takes to build. If we can build the same number of cars from a smaller factory, we are gonna be able to scale EV production faster. Our next powertrain factory is 50% smaller than the one that's behind me today, even though it has the same capacity. All of these improvements, I think, are gonna be transformative for the adoption of EVs and our ability to scale them. There's one more thing that I wanna highlight. I talked about how we had reduced the amount of rare earth in our powertrains. As the world transitions to clean energy, the demand for rare earth is really increasing dramatically.
Not only is it gonna be a little hard to meet that demand, but mining that rare earth, it has environmental and health risks. We wanna do even better than this. We have designed our next drive unit, which uses a permanent magnet motor to not use any rare earth materials at all. How does all this fit into the master plan? We can make lower cost products that are still efficient and compelling, and we can make them at scale. We can use less of constrained commodities, silicon carbide, rare earths. We're gonna build them all in compact and high output factories that are easy for us to build quickly. We're gonna make that easy to scale powertrains all the way up to the levels that Drew and Elon mentioned at the beginning.
This achievement, like all of the achievements that I mentioned today, is only possible because of the incredible people on our powertrain teams.They are absolutely committed to the cause of sustainable energy, and that is why we can do what no other company can do. Thank you. Next, I'd like to welcome my colleagues, Pete Bannon and David Lau.
Thanks, Colin. Good afternoon, everybody, and welcome to Texas. Pictured here is the low voltage system of a Model S from 2012. This car was primarily designed with vendor source controllers integrated by Tesla. There are over 300 low voltage devices that encompass everything from the mundane, like the light in the glove box, to the complicated, like the infotainment computer, to the safety critical, like the airbags, steer-by-wire, and brake-by-wire. The low voltage harness is built from individual wires cut to length, crimped, and inserted into connectors, a manual process that is tedious, error-prone, and doesn't scale well. Going forward, we want to reduce the size and complexity of the harness and enable automated manufacturing.
These wire harnesses introduce extraordinary complications, especially in the early stages of development and bring up of a vehicle, because when we're trying to bring up this entire system and we see that it's not working properly, we don't know whether it's a problem with the software, with the controller, the processor, any of the endpoints in this entanglement of wires. We have to go and debug everything all at once. Pete's gonna tell us about how we're gonna make it better.
We're gonna talk briefly about a few things that we've changed over the years since the Model S. With the Model 3, Tesla started designing an increasing share of our controllers by merging controllers together, and we were able to simplify the design with a significant reduction in wire count and weight. Going from S - 3, we reduced the wire harness by 17 kilograms compared to Model S. This is a pretty big deal, and just to put that into context, our VP of Engineering, Lars, will deliver a bottle of your favorite spirit to your desk if you manage to save 1 kilogram of weight from the car. This improvement cost him quite a lot. Did we skip one?
Yeah.
The original Model 3 controllers were enhanced and then built into Model Y. Those improved controllers were then moved back into Model 3 when we introduced the heat pump. At that point, we had shared controllers across both cars, which helped simplify our supply chain. Those Model 3 and Y controller designs were then updated and enhanced for the new versions of Model S and X that we introduced in 2021. For Cybertruck, we are now designing 85% of the controllers in the car. For the next gen platform, we're gonna finish the job and be designing 100% of the controllers to give us full control of the design and the supply chain at the component level.
Having full control of the supply chain at the component level has been critical to Tesla over the last few years as the supply constraints have hampered our ability to build cars.
With that control comes control over all the software, which enables us to develop features and functionality that we never even dreamed of at the time we designed the hardware. That's why you see software in your cars getting better and better over time in ways that we didn't even think of when we designed the hardware in the first place. Mm-hmm. Yeah. Sentry Mode is one of my favorite last-minute changes.
One of my favorites.
With the introduction of Model 3 in 2017, we deleted relays and fuses from the car in favor of eFuses. eFuses replace moving parts with solid-state transistors that provide fine-grained control of the power system to software and allow software to do advanced things like load shedding in the car under adverse conditions. eFuses also allow for software-controlled retries for transient faults, they allow for detailed monitoring of the power system over time.
The theme you hear a lot about today is software-controlled hardware, and when I think about that in my team, software-controlled hardware is fundamentally about being able to sidestep what otherwise would have been static trade-offs between one attribute or another in a piece of hardware. We are able, with software, to instill intelligence, context awareness, and context-specific behavior into what otherwise would have been a piece of hardware that had to get optimized for one type of scenario. We get more of everything.
In 2022, we completed the transition from lead-acid batteries to lithium-ion batteries and which use a new toolless connector. We expect lithium-ion batteries to last for the life of the car, so don't expect anyone to wake up to a dead battery ever again. This eliminates a major source of failures for our car, and the toolless connector makes it easier to service the car, and it also includes a software feature to allow validation that the connector has been properly installed at the end of a service event, which removes another source of failures in our cars.
Obviously, the mass and volume savings are also significant, which is super helpful. At Tesla we're always trying to improve every single component in the car, and a nice example of that is the 15-inch display that was originally shipped in the 2017 Model 3. Over time, the cost of the display has gone down 24%. We've been able to reduce the weight by 12%, and we've reduced the power by 33%. At the same time, we've increased the brightness of the display by 50% and improved the color accuracy. This is one of our favorite things at Tesla, is to make a component cheaper and better. At Tesla, the drive to improve efficiency in the car is never-ending. I think of it as like peeling a carrot.
You just take a swipe at it, a little bit falls off, doesn't really make much of a difference. If you just keep at it, over time it accumulates into a pretty significant improvement. One of the changes that we're looking to make right now is to change something that's been steady for the last 60 years: 12 volts. 12 volts has been, as I said, for 60 years. The demand for power in the car has been steadily increasing to the point where we now have to have pretty large wires to drive over 200 amps of current around the car, which increases the mass and cost. With Cybertruck and all future Tesla platforms, we will be moving to 48 volts. This reduces the current needed by a factor of 4.
Since power loss in the harness is resistance times the square of the current, a 4x reduction in current leads to a 16x reduction in lost power while distributing energy in the car. That allows for smaller wires, smaller eFuses, and smaller controllers. It also allows us to make those heat sinks smaller, or in many cases, remove it completely, all benefiting the car in terms of mass and weight. 48 volts is the future for low voltage design at Tesla, and likely the rest of the industry in due course.
We welcome and encourage other OEMs and the entire supplier network to join us on this evolution.
You bet. The number of wires in the car is driven by the number of endpoints that need to be powered and controlled. In the past, centralized control has led for wires spanning the entire car. For Cybertruck design, we have moved to a local controller where the wire is connected to the nearest controller, and those controllers are connected over Ethernet. Wires are routed to the nearest controller, where the data is converted into a network packet for transmission to the correct location in the car. To be effective, the network must be reliable, have low latency and low jitter. These are all attributes that we've been able to achieve with the current design. The design has eliminated most of the cross-car wires in Cybertruck, with the next-gen platform, we're gonna finish the job and eliminate all of them.
This consolidated vehicle network allows us also to make a whole lot more dynamic changes on the fly to how components in the vehicle talk to each other, rather than the traditional approach of separate CAN buses that are spread throughout the vehicle.
Yeah.
fixed in hardware.
One of the nice things is that you can see the entire vehicle through a single connection, which wasn't possible in the past for debug. For the next-gen platform, we're looking to optimize the controller design across the entire car and across all of the organizations, not just for a given subsystem. Simplifying the wire harness will enable automation. 48 V will allow us to reduce the size, mass, and cost of the low voltage system, and is one of the key things to enabling us to scale production of the low voltage systems to for the Master Plan 3. David?
Cool. I joined Tesla in 2012, even in those very early days of the Model S, we already had the ability to update software over the air in all the controllers throughout the entire vehicle at a time when other cars were just starting to update software in only their infotainment units. Furthermore, we've had sophisticated, anonymized data logging and telemetry capabilities that allow us to understand how the fleet is doing, what customers are doing with it, and how it's responding in the field to that usage. These two capabilities combined, OTA updates and data insights, give us the ability to iterate quickly on our software and to maximize the amount that we learn and proceed and the progress that we make in every one of those iterations.
We've used these capabilities to inform countless design decisions in both software and hardware.
Yeah, one example is that we were able to monitor and track the use of the sunroof in our cars and found that our customers never use their sunroof, so we made this easy decision to remove it.
Very barely, yeah. We haven't had sunroofs for a while. As another example, we are able to use our collision data to design our crash safety systems for what happens in the real world, far beyond what regulatory and consumer ratings tests prescribe. Furthermore, we're able to use that data to recreate, in simulation, all of those crashes every single time we make a change to the vehicle's design or to the software that controls the airbags and other restraint systems. Here on this graph, the gray dots that show up first are specific crash tests that are prescribed by regulatory and consumer ratings agencies throughout the whole world. The rest of the color that fills in around and on top of it, those represent the crashes that we've seen happen in the real world. Those are what we design for and what we test to.
Right. Last year, this data was used to change the algorithm for seatbelt tensioning to reduce injuries in the field, which was OTA to all of our cars.
Here are a few statistics that demonstrate how staggeringly quickly we gain new insights from the anonymized data that we've received from the fleet every day. Every day, 123 million miles driven, 1.9 million charge sessions, and those rates are only increasing as we deliver more and more cars into the world. As another example, we've analyzed patterns in the way the fleet drives and charges to optimally size battery packs for our next generations of vehicles. You'll hear lots of other examples throughout the rest of the day about how we're using data to inform decisions in the product and in manufacturing. You've heard a lot from Pete, Colin, and others about how important vertical integration is for us in hardware, but it's especially important for us in software.
This is extremely difficult from the way the rest of the traditional automotive supply chain is set up. In most situations, all the controllers in the car are delivered by different tier one suppliers, whose software is written by tier two suppliers, who farm some of that software out to different tier three suppliers, and so on and so on. Making a change that spans multiple components takes months of coordination before any work can even start. Furthermore, integration of software in the server side with what's happening in our vehicles has always been a core part of our DNA and enables us to do things that nobody else can. Most engineers think about the vehicle as a complete, consolidated, fully contained system.
In the software team, we think about the system as including the vehicle as well as all of our back-end server-side applications and infrastructure and the resultant feedback loop from the entire fleet that informs all the decisions we make as an engineering team. For example, we recently released a feature in Model S and X that automatically, predictively adjusts and raises the suspension for ride comfort before the car hits a section of rough road. We do that by leveraging the fleet to generate a map of road roughness everywhere our cars drive. Sounds pretty simple, requires coordination of software across a number of different components inside and outside of the car.
The restraint control module, whose inertial measurement unit senses the road roughness, the Autopilot computer, whose GPS module and localizer figure out where the car is in the world, our navigation server, which aggregates the anonymized telemetry from the fleet and annotates our map with things like updated speed limits, lane topology, and now road roughness. The onboard navigation engine, which looks ahead of the route that the car is traveling on and determines whether things are about to get rough. Finally, the air suspension controller, which takes all of those factors into account and decides continuously whether it is appropriate to adjust the suspension for ride comfort.
Before we released this feature to the world, we sent prototype versions of it to the entire fleet that ran passively in the background, sending us anonymous data about every time it would have engaged, which taught us exactly how it was going to behave in the wild, in every environment, in every circumstance.
Yeah. Being able to test new algorithms in the background without impacting the car is really critical ability for us, especially for safety critical things like automatic emergency braking.
We've used it to iterate on early versions of our stability control algorithms, which were introduced to the entire Model 3 fleet in 2017 with extremely high confidence, even on a very foundational and safety-critical function. So far, you've heard me talk a lot about how we leverage our software systems to iterate quickly on the customer-facing product. What I haven't talked about, and what we haven't said much about publicly in general, is how we leverage these capabilities in our own internal operations. For example, in the early days of manufacturing Model S, we very quickly realized that there are a lot of different ways you could misassemble a car. You could forget to plug in a wire, you could not fully set a connector, you could pull a part from the wrong bin.
When we designed our manufacturing processes for Model 3 and all of our other vehicles going forward, we took a page from the software playbook, which says, "Test early and test often." We applied that mantra to every single car that we build at every step of the assembly line. Now, when a production associate plugs something into the car, the central car computer sees that connection, confirms that it's the correct part for the type of car that's supposed to be built, installs a software update if necessary, checks configurations and calibrations, and runs a barrage of tests on that thing and all the other things that are connected to it.
If it finds an anomaly that can't be fixed purely with software, it throws an alert immediately that is displayed prominently on the vehicle's display as well as sent to back-end command and control systems so that a human can come over and fix the problem before the rest of the car gets built on top of it. We're taking a similar approach in service. Through a combination of onboard vehicle diagnostics, natural language processing applied to customer narratives when a customer schedules a service appointment, and a suite of internally developed tools, we are successfully diagnosing, scheduling, and ordering parts for over 33% of customer concerns in service.
Yep. We also use testers built around our own controllers to test subassemblies before they get to the car, which will be even more important as we go to Lars' unboxed assembly strategy.
100%. We'll know with extremely high confidence that before those boxes get put together, that they are assembled correctly. Okay. You've heard at lots of other events about all the awesome things that our Autopilot and AI teams are doing to make the car drive itself, bringing us into the future of autonomy. We've also been thinking for years about all of the other pieces that we are going to need to manage a network of autonomous vehicles. A lot of this has been happening behind the scenes in the form of platform-level functionality that we'll leverage later. You've seen some of it surface already in terms of features that our customers and our internal operations teams can benefit from.
Like in 2021, we built on top of our mobile app Phone Key and gave our customers the ability to share their car with anyone by sending them an invitation in a text message or an email. Last year, we introduced profile synchronization, which synchronizes your seat, steering, and mirror positions, as well as your settings, media favorites, and stored navigation locations across all vehicles in your account. Internally, we have an app that allows our personnel in engineering, manufacturing, delivery, and logistics to view, locate, and drive all Tesla-owned vehicles at their site. When customers bring their car into service, when they receive a Tesla-owned service loaner vehicle, we are starting at some locations to automatically add that vehicle to the customer's mobile app account, so that combined with cloud-synchronized profiles and Phone Key, it's a completely seamless experience.
As soon as the customer walks up to that loaner, it behaves and feels exactly like their own car. All this is built on top of end-to-end encryption and cryptographically signed commands so that customer data remains private and obscure, and the fleet only trusts commands from authorized parties. That's a bit of a preview of all the pieces that we're building looking ahead to our future of autonomous fleets: synchronization, permissions management, security, and privacy. That has been a whirlwind tour through some of the things that we're doing behind the scenes to enable the efficiency, cost reduction, and speed that we are going to need for their next phase of growth and for the next phase of the Master Plan. Now, to tell us about the latest in Full Self-Driving, I'd like to introduce Ashok.
Thanks, David and Pete. Hey, everyone. My name is Ashok Elluswamy. I work on Autopilot and self-driving at Tesla. I joined Tesla back in 2014, so I've been working on this problem for almost 9 years now. Yeah. Some of you might be wondering, "Hey, what's the self-driving got to do with the plan to a sustainable future?" It's actually a critical part of this plan, and here's why. Currently, when the car is not being used, it is sitting idly in parking lots, not doing anything. When autonomy is truly unlocked, this car, instead of being idle, can go serve other customers. This fundamentally reduces the need to scale manufacturing to extreme levels because each car is being used way more. This is no trivial problem. Building a scalable self-driving system is, I think, one of the hardest real-world AI problems out there right now.
Nonetheless, we at Tesla have made significant strides in making one of the most general systems at solving this problem. There are three main parts to get right to build a scalable problem. First is the architecture, the architecture of the AI system. Second is the data, third is the compute. We'll start with the architecture. At Tesla, we are betting on AI machine learning neural networks to help us build a general vision and planning system. In the early days, we used to have single camera, single frame neural networks that produced some outputs. These were stitched together in some post-processing steps for the planner, this was very brittle and was not leading to great success. What we did in the last few years is transition most of our stack into this multi-camera video neural networks.
These neural networks take in the live feed from the car in real-time of the eight cameras in the car and produce a single unified 3D output space. There are many tasks that we produce, such as the presence of obstacles, their motion, lanes, roads, traffic lights, what have you. This is one example output that you're seeing here. This is from our occupancy network that predicts the positions of obstacles and their motion. You can see that it precisely captures the swervy, violent motion of this truck next to us, and this helps the planning system to avoid a collision with this object. Some of the tasks, such as lane connectivity, are more complicated to model using naive methods in computer vision.
We don't stop at the computer vision techniques, but we reach out to techniques in other areas such as language modeling, reinforcement learning, to model this task. This is similar, these networks use similar techniques such as transformers, attention modules, autoregressive modeling of tokens, similar to what large language models like ChatGPT do out there. With such an end-to-end system of solving perception, we have really removed the brittle post-processing steps and produced high-quality outputs for the planning system. Even the planning system is not stuck in the old ways. It is now starting to use more and more AI systems to solve this problem. The neural network-based planners are needed especially in complicated urban planning when there's a lot of other objects interacting with us.
This is, for example, an intersection where we have to turn left while yielding to crossing objects and to the pedestrians crossing the road. We have to do this both safely and smoothly while respecting everyone's right of way and preferences. If this was done naively, each configuration would take 10 milliseconds of compute, and there are easily thousands of configurations to reason about. This would not be feasible using traditional compute. By using AI, we have packaged all of this into a 50-millisecond compute budget so it can run in real-time. The second big piece of this puzzle is the data. This is where Tesla has a unique advantage because it can tap into the fleet to access the exact data that can fix the problems. Raw data is not sufficient. We...
In order to train these networks, you need label data. You need labels to supervise the networks. If you only depended on the human labelers, this data would be too tiny to train these large multi-camera video modules. We need lots and lots of data to train these networks. Hence, we have built a sophisticated auto-labeling pipeline that collects data from the fleet, runs computational algorithms in our data center, and then produces the labels to train these networks. Here, you are seeing a 3D reconstruction that is happening by collecting various clips from different cars in the fleet and assembling them, all of them together into a single unified representation of the world around the car. You can see all the lanes, the road boundaries, curb, crosswalks, even the text on the road being accurately reconstructed by these algorithms.
We don't have to stop at this kind of reconstruction. Once we have this base reconstruction, we can build various simulations on top of it to produce an infinite variety of data to train for all the corner cases. We have a capable simulator that can synthesize adversarial weather, lighting conditions, and even the motion of other objects to test all the corner cases that might even be rare in the real world. Here's an example of why this data is critical and how we can solve problems using data. Back in the days, for example, in this case, we had some false braking where we thought this car that's actually parked there was gonna move into our path, and hence we are precautiously braking. It is unnecessary because the car is truly parked. How can we solve this?
What we did was we mined the fleet for similar cases where the car had false braking due to some parked car. We added 14,000 videos to our training set, once we trained the networks again with this new data, it now correctly understands that, okay, there is no driver in this car, hence it must be parked, there is no need to brake. This is a systematic way to solve problems. On the chart on the right side, you can see that every time we add data, the performance improves. We can do this for every kind of task that we have in our system. This is what we call as data engine. We identify challenging cases, such as the one you saw earlier. There could be other different types of challenging cases too.
We mine the fleet for such data, put it through our auto-labeling system and produce the labels, add it to the training set, and once we have the newly trained models, we deploy them to the fleet. If we rinse and repeat this process, everything gets better and better. The final critical piece is the compute. In order to train these large models in a reasonable amount of time, you need lots of compute. In addition, compute is also required to produce the labels automatically. This is just compute in our data centers. In addition, we also have high compute computers in the car which can run up to, like, 150 TOPS of compute. On the back end, we have a 14,000 GPU cluster, and roughly 30% is used for auto-labeling, and the remaining 70% is used for training.
We also have a 30 petabyte of video cache. This is growing to 200 petabytes. All of this is gonna significantly increase once we bring Dojo, which is our training computer, on board into this. Just to give you a reference, just the occupancy networks that you saw earlier use 1.4 billion frames to train these networks. We have already shipped our FSD Beta software to pretty much everyone that has bought it in United States and Canada. This amounts to roughly 400,000 customers who can turn it on anywhere. The car would attempt to drive to the destination. It's still supervised, it can already handle turns, stop at traffic lights, yield with other objects, and generally get to the destination.
We have observed that people who use FSD are already 5 to 6x safer than the U.S. national average. Like I mentioned earlier, the solution to scalable FSD is getting the AI architecture, the data, and the compute just right, and we have assembled a world-class team to execute on this. We are pushing the frontier on these three items, and as we improve the safety, the reliability, and the comfort of our system, we can then unlock driverless operations, which then makes the car be used way more than what it is used right now. With that, I'd like to introduce Elon back on stage to give some more updates.
Play the video?
Yeah, this is the next latest video.
Okay. It's kind of weird seeing the arms and legs just separate. We have a whole lab full of arms and legs. Worth bearing in mind that, when we did AI Day, this version of Optimus didn't work, walk at all. The rate of improvement here I think is quite significant. It's obviously not doing parkour, but it is walking around, and we have multiple copies, I suppose, of Optimus. The thing that's, I think Tesla brings to the table that others don't have is that we have the real world AI. We're the most advanced in real world AI. The same AI that drives the car, which you can think of the car really as a robot on wheels, and this is a robot on legs.
As we solve real world AI, and I don't think there's anyone even close to Tesla on solving real world AI, that same computer and software goes into Optimus. It's not that helpful to have a humanoid robot if you have to program every individual action. It needs to be able to walk around autonomously and solve tasks. You should be able to instruct it in simple things by showing visually what you're, what the robot needs to do or just telling it what to do. I think that's a key advantage that we have. We also are good at designing things for manufacturing and then manufacturing itself. The actuators in Optimus are all custom designed Tesla actuators.
We designed the electric motor, the gearbox, the power electronics, obviously the battery pack and everything else that goes into Optimus. We were quite surprised to find how little was available off the shelf. 'Cause there's a lot of a vast number of electric motors, gearboxes and whatnot that are available in the world, and we found none of them were useful in a, in a humanoid robot, literally none. You have to custom design the actuators for a humanoid robot. The same team that designed the groundbreaking electric motors that are in the, say, the Model S Plaid, designed the actuators in the robot.
So I mean, for practical purposes, what this means is that we should be able to bring an actual product to market at scale that is useful, far faster than anyone else. You know, assuming the things I'm saying are true, or at least you can put it, I think they are true, it's just a question of the timing. You start getting into interesting questions of, like, what's the ratio of humans to humanoid robots? I think it might be greater than one to one. You know, 'cause you could, you could sort of see a use, a home use for robots, certainly industrial uses for robots, humanoid robots. I think we might exceed a one-to-one ratio of humanoid robots to humans.
It's not even clear what an economy means at that point, you know, since an economy is output per person times persons, but if output is much higher and there's no limit on persons, then what's the actual limit on the economy, you know? We're still pretty far from Kardashev Scale here, but we're getting there. It's probably the least understood or appreciated part of what we're doing at Tesla, but will probably be worth significantly more than the car side of things long term.
I think Rebecca's next.
All right, great.
for talking about charging.
Charge.
Thanks. Hi there. My name is Rebecca Tinucci, I lead our charging infrastructure teams here at Tesla. At Tesla Charging, we have understood since day one that a great charging experience is the linchpin to electric vehicle adoption. That understanding has meant that we've always taken a holistic approach to charging. That's a word that you've heard a lot here today, but what it means for charging is that we've considered every use case. We think a lot about what it means to charge at home, even if that home is an apartment or condo, and we spend a lot of time thinking about what does it mean to charge away from home, including if that's for daily commuting or if you're going on a road trip. This holistic approach has led to some pretty incredible results.
In 2022, we provided 9 TWh of charging across our various charging methods. Over 50% of that was supplied via convenient AC home charging. When our customers are away from home, they can visit one of our 80,000 charging points. That includes 40,000 of our beloved Superchargers. Getting here has meant that we've spent 10 years building charging infrastructure when basically no one else in the industry would do it. While we certainly have a lot of areas that we want to improve, those 10 years have afforded us the opportunity to get pretty good at charging. First, we have the industry's lowest deployment costs. Our costs are often 20%, if not 70% lower than alternatives, and that goes for both our Supercharging hardware and deployments and our AC charging product lines.
A lot of reasons that we're able to achieve this, you've heard a lot of them here today. We're vertically integrated, we manufacture and engineer all of our own charging equipment, we share components across our different product lines, and on the Supercharger side, we also install and operate all of our own sites. That's led us to be pretty obsessed with finding new ways to innovate around installation. As an example, we have recently extended our excellence in manufacturing to how we build Supercharger sites. We are pre-building four-post Supercharger units at our factory in New York. We load them on a truck, we truck them to site, and then we crane them into position. This method saves us 15% on our deployment costs, and we can install a site in a matter of days.
Once we install a site, we also operate it really efficiently. Over the last few years, we've been able to cut our per kWh cost by 40%. A lot of reasons for this, but one at the top of the list is we've focused on increasing our site utilization. Site utilization is just how many sessions or kWh can we push through a site or a post, basically doing that allows us to spread our cost over more sessions, thus lower cost per kWh. Of course, that's easier said than done because when we push up site efficiency, of course, the risk is that we have a poor customer experience and we have wait time at our sites. This is where Trip Planner comes in. Trip Planner is our in-vehicle routing or navigation system.
Other electric vehicle manufacturers, well, some of them have vehicle-side data. Other infrastructure providers have site data. At Tesla, we have both. What that allows us to do is to use Trip Planner to route vehicles towards available sites and away from congested sites, so we can balance our utilization without risking wait time. The results speak for themselves. Over the last few years, we've been able to drive up site utilization by 30%. That means lower per kWh cost, while also cutting our wait time in half. We think this can get even better. Today, we're feeding Trip Planner with real-time site information and information about vehicles currently at sites. Going forward, we'll be moving that to projecting site occupancy based on the understanding of what vehicles are currently routing to those sites.
Ultimately, our vision for Trip Planner is that it's the air traffic controller for electric vehicle charging across all infrastructure on a global basis. The last thing we focused on to get here is quicker charge times. We're really proud of what we've been able to accomplish here. We've shaved off 30% of our charge time, average Supercharging site time, visit time over the last few years. This has taken improvements on the hardware side, with our software, on our vehicles, and on our infrastructure, and we're really excited to continue pushing this trend down. Looking forward, the job does not get any easier. As you heard earlier, we aspire to a fully electrified global fleet. That fleet, from an industry standpoint, needs 9 PWh of charging on an annual basis.
While Tesla charging certainly doesn't have to supply all of that, it does require that for our part, we have a few new focuses in order to scale. First, if you want a fully electrified global fleet, all of those vehicles have to have a great and reliable charging experience. As many of you know, recently, we have started opening up our networks on a global basis. Over 50% of our Superchargers in Europe are currently open to other electric vehicles. We've also opened up in Asia-Pacific with our first sites in Australia. Just yesterday, we opened our first 10 Supercharger sites here in North America, in the U.S., to other electric vehicles. We've also invested a lot to be able to make this a really easy experience for our new customers.
All you have to do is sign into our Tesla app to unlock a post and start charging. We've also on physical site changes, we've also added hardware to our sites. They're called Magic Docks, and basically, that is installed in areas where we have different charging standards, and they allow for other electric vehicles to come to our site without having to bring their own adapters or hardware. We've also just started installing our fourth-generation Supercharger posts. Those are being installed in Europe first. While it's not a big mention, not a big thing to mention, they do have longer charging cables so that we can more easily reach the charge ports of different vehicles.
The second thing we need to make sure we do when we talk about scaling all of this charging infrastructure is we need to make sure that it's powered from renewable sources. We're very proud that over the last two years, we've procured enough renewable energy to offset the amount of charging we've provided to customers. As we look forward and we talk about this fully electrified fleet on a global basis, we really wanna make sure that the demand for charging more closely follows when renewable sources are available. This chart is on an aggregate basis for the U.S., and it varies based on what your renewable generation sources are locally. Basically, what this shows is we want to kind of move that charging curve. For solar and wind in this example, it means more daytime charging.
We think the best way to go towards daytime charging is to install AC charging that is convenient and low cost everywhere vehicles are typically parked throughout the day. That's really the plan. While supercharging and home charging definitely stay a big part of the puzzle, our teams are currently scaling to install AC charging everywhere vehicles charge during the day, so we can power them from renewable sources. Summing it up, while we've spent 10 years installing infrastructure that is low cost, it's efficient and it provides a great customer experience, we're really just getting started when we talk about supporting Master Plan Part 3. We've got to scale capacity on an industry-wide basis to 9 PWh on an annual basis.
We've got to open up to non-Teslas and make the charging experience for everybody really great, and we've got to power all of this via renewable sources. One more thing, yes, we've got to scale, we've got to scale our infrastructure, and yes, we wanna power it via renewable sources, but we're Tesla, so we also wanna make sure that we're continuing to focus on providing really incredible charging experiences. With that, I wanna welcome up our leaders of supply chain, Roshan and Karn, to talk a little bit about that topic.
Thanks, Rebecca. Hi, my name is Karn, and my team leads, supply chain for, electronics, powertrain, and battery at Tesla. We've also got responsibility for indirect purchasing, construction procurement, and warehousing and distribution.
Hi. My name is Roshan. I've been with Tesla for over 14 years. My team manages vehicle commodities, solar, logistics, planning, and capital equipment.
10 years. Forgot that. Today we wanted to provide you a quick overview of Tesla's supply chain and what really separates us from the typical automotive supply chain. As you've seen in presentations prior from Pete and Drew and some others, Tesla designs a lot of the subcomponents that go into our vehicles. We're not buying things that are off-the-shelf from suppliers, so supply chain is not purely a commercial relationship as it is in most other companies. We actually have an arm within supply chain called Supplier Industrialization Engineering, and the burden of taking a drawing, a design from concept and turning it into thousands of products produced at the right cost and at the right yield falls with this group.
Of course, they work very closely with design engineering, really, the level of detail that our engineers get involved in terms of building the capability of the suppliers is pretty detailed. You know, once we get a new part, naturally with electric vehicles, a lot of the supply chain doesn't exist. Our engineers will basically take the drawing, turn it into a manufacturing concept, do the equipment selection, and physically go to the supplier and stay there for weeks or months, however long it takes, to basically do the line bring-up. Once the line is brought up and we're hitting the right rate, the right quality at the right yield, then we work on things like automation and yield improvement and those types of activities.
There's a big group of mechanical, electrical, industrial, you name it, engineers within supply chain, and it's their responsibility to take care of this. This is a huge strategic advantage because we manage every detail of our supply chain, because we're so integrated with our partners that we know about issues before they happen. It's almost live. We've got dashboards, we've got connections, and across the 10, you know, 10,000 or so factories across the world that are making our components, we have a pretty good view of the health of those suppliers, and this has been a key asset in how we've been able to kinda manage things through a lot of different issues over the past couple of years. You know, supply chain is a game of perfection. Perfection is a passing grade.
It's like, you know, if one part doesn't show up, that line goes down, and our CEO finds out about it in less than 20 minutes. Tesla is very different than Apple. They outsource manufacturing, we do it in-house, so supply chain's kind of right in the middle. And perfection is increasingly elusive and very expensive, if you kinda look at all the issues we've had to encounter over the past year. Today we're just gonna talk about how we got through the past, and then what are we gonna do going forward, and how are we gonna get there. Let me draw your attention to this slide. On the left, you'll see a portfolio, our existing portfolio of products, right?
At the top you've got the S and X platform, in the middle you've got the 3 and Y. I think everyone's more than familiar with those. Then the Tesla Energy platform, which is a very high rate of growth for us.
Don't forget the solar and Superchargers as well.
And solar and Supercharger. Thank you, Roshan. You'll see a reduction. We've basically got the tier 1 and tier 2 parts. This might be boring, but a tier 1 supplier is any supplier that builds and supplies parts directly for consumption in our five factories. Our tier 2 supplier basically does the same thing for a tier 1 supplier, so they are the supplier to our supplier. Depending on the supply chain of any one of these products, this can actually go down to the tier 6 levels. For example, if you're talking about the battery cell, we really get involved with the mines of, like, where we're procuring the lithium, the cobalt, and all that sort of stuff because we're procuring such high volumes of it.
You'll see a reduction in the number of parts on the S and X platform, to the 3 and Y platform, and this is because of the great work our design engineers have done of making the car simpler and easier to manufacture. In parallel, we actually had a similar strategy in supply chain. We had cast a wide net to a lot of suppliers for the S and X platform. There was a lot of suppliers that we invited to partner with us. We used that platform as sort of a filter to see which suppliers have the technical, financial, and cultural capabilities to match us, right? Who, like who do we invite to the party when we really scale things up?
We, you know, we've really dramatically reduced the number of suppliers that we were dealing with for the Model Y platform. This is an approach we'll continue to do for future platforms as they come along that are even more high volume. This has really been our approach. It's made things a lot easier and it, you know, we could figure out the suppliers that were capable of moving at our pace and then the ones that didn't have the desire to or the capability to. You need both desire and capability. Secondly, the complexity on the tier two level is very high, right? These are the suppliers to our suppliers. There are a lot of components. Think about silicon, think about resistors, capacitors, diodes, all the little assemblies-
All the raw materials.
Raw materials, all the... That's even further down.
Yeah.
All these little sub-assemblies that have to come into Tesla, the complexity is immense. Even though at the tier one level we're talking about 8,000 parts total, it seems like a lot, but it's not a lot. The management of the tier two is really where we excel, and I'd like to illustrate that with an example. Meet the car computer, very innocent-sounding name, and it's anything but. This is an absolute monster to manage. I think you've all seen the picture on the left of the Autopilot board. That's the top of the Autopilot board. The Autopilot board also has a bottom, and the bottom is populated heavily with components. On the other side of the heat sink, you've got the MCU, the multimedia cluster board. This board is equally complex.
It's also double-sided, 8-layered, and these boards, these computers run so hot at peak operation that they have to be liquid cooled through a heat sink. This assembly requires taking those two boards and then bonding them to a heat sink that's hermetically sealed as an assembly. Of course, flashing the software and all that sort of stuff, and then that's 1 part number that comes to Tesla. This is an example of 1 tier 1 part number that's a very complex assembly to manage at the tier 2 level. There's more than 7,000 components here. There's a, you know, as we stand here, a component's being assembled onto a car computer every 1.4 milliseconds. The line that builds this computer is the length of a football field, so it's quite complicated.
Initially, when we first started building this board, due to the complexity of it, we had to rely heavily on labor. Once we dialed in the quality, the rates, and the yield, we started focusing on making this more efficient. The only way to control cost is by removing labor. The first step is removing labor. The second step is fully automating. The third step is turning off the lights and letting the factory run, ideally. That's gonna be the goal for us here. 95%, but our work is not done. We're gonna be going a lot further than this. This illustrates the point that a part is not a part. A part has a lot of complexity underneath it. Supply chain is a game of multiple tiers.
What's made us successful is our involvement in all the details with our supply engineers that are Tesla employees. They're Tesla supply engineers, so they're like on our payroll that go and ramp up these capabilities at our suppliers, and then just managing each and every attribute of it. You know, 7,000 a day through the chip shortage, through the pandemic, through all the other stuff that we had to deal with, was very difficult to manage. Because we had all the details, we were able to pull this off. Of course, supply chain is not just about parts. It's also about logistics and a lot of other things, I'll hand it off to Roshan for sharing some stats.
Talking about inbound complexity, just over the cost-course of just last year, we moved about 16 million pallets from our suppliers to our factories. Just to visualize that, if you put those pallets side by side, it could cover the half the circumference of the Earth. Talking about, again, the tier two complexity, over 1 billion electronic components get moved every week to support the production that Karn and team manages.
This is right now, 1 billion components a week.
Right.
$52 billion a year. This is gonna go up, by the way.
On top of that, we also have the responsibility to make sure the right component is supplied to the right service location to support the growing fleet of our vehicles, and we have about 685 service centers that these components need to show up just in time. We've also been as a result of pandemic and even before that, we're working on dual sourcing and triple sourcing some of our critical components. As a result of it, we have suppliers around 45 countries in the world.
Yeah. Our approach really has been to bring the manufacturing of sub-components closer to the point of consumption. It makes the supply chain more green 'cause you're not burning diesel moving stuff around, and it also gives us the security of supply of having multiple sources in case one factory gets taken offline.
That's how we got through pandemic also.
Yep.
Last 3 years were incredibly hard. Karn and I used to look a lot younger pre-COVID. On a serious note, pandemic really changed the game of supply chain management. We were just not being a regular supply chain person. We're rolling up our sleeves and negotiating with state governments, city managers, mayors to help kickstart our suppliers to restart their operations. You know, there are some really inspiring stories where our supply chain teams really fought hard to make sure that the factories don't go down. It's also a testament to the resilience of our supplier partners. Having said that. You know, it was not all about resilience and never give up attitude.
We had also laid the foundation of having dual sources and also, you know, in many instances, we had many, you know, things that we had simplified, the parts that we had simplified, the supply chain design that we had simplified, due to which we were able to get through the pandemic in much better shape than rest of the automotive.
It forced us to adapt.
Right.
You know, we encountered issues one at a time. It was really a team effort. We heavily relied on the government relations team. The COVID pandemic, as you're aware, you know, happened in different countries at different times. Unfortunately, you need all the parts to make the car. At any given point, there was one factory that was lying down.
Right
There was one jurisdiction that was being, you know. They wanted to shut down a factory because you had an outbreak. Learning the rules, understanding, coming up with argument points, leveraging the team, and getting those. It was a difficult but good experience.
Exactly.
We learned a lot.
Yeah. As we were just figuring out to work with COVID, then the port congestion hit in, on the West Coast. logistics lead time went up by, you know, twice the amount, and we were operating with less than 35% accuracy in the system. That means at any given point, the ETAs of the vessels were 75% off. We figured out how to work with that, and then the chip shortage hits.
The chip shortage. That was a nightmare. We got through it. I mean, you know, I think it's a testament to the tenacity of the team and the resilience of our supplier partners. It's mostly behind us, and I think the results speak for themselves. Tesla was able to grow production volumes at a time where the rest of the industry had a hard time staying flat. I think they shrunk, actually, cumulatively. You know, we're often asked the question, how do we manage through it? And the answer is very much kind of baked through the presentations today. First, we design our own electronics. In order to control cost, we really had to go and build relationships with all the tier two suppliers to negotiate them directly.
These relationships were formed, you know, over 10 years ago, and we've really gotten to know them well. When a shortage hit, we knew, you know, who to escalate to, who to call, who to ask for help, and really kind of swarming the issue at the point of constraint. The best team, you know, strong relationships with vendors. We also have a pretty sophisticated capacity planning function. Tier 2 capacity needs to come on earlier than Tier 1 capacity. Tier 3 capacity has to be even earlier. Our team does a very good job of having that strategic alignment with all our partners to make sure that the capacity is ready before we need it. We're a mission-based company. Our goal is to put as many EVs out there. We did not let the pandemic phase us.
We were full throttle, it was pedal to the metal in terms of our demand signal. That also helped. The silicon industry has very long lead times. They invest billions of dollars in sophisticated fabs that take years to build. So that stability and demand is something they appreciated, and it's something that is very much a Tesla thing. This was looking back, right? This is how we got here. This is how we got from 0 to 40,000 cars a week. Let's talk about how we get to 20 million. The question gets asked, you know, is Tesla gonna basically break the chip industry if we grow to 20 million vehicles? The answer is no. Let me walk you through our logic on that.
As it stands right now, a Tesla enabled with a Full Self-Driving computer has four times the silicon content of a regular vehicle. At about the 2 million run rate, we're consuming about 700,000 12-inch wafer equivalents. That's the industry term that's thrown around for capacity in silicon. That's how much we're using, so a little below 1 million. The global wafer capacity, based on the estimates we've seen, is about 135 million. We're about 0.5% of the industry. Not a lot. If you get to 20 million, when we get to 20 million, and if we don't make any simplifications in terms of part count, which is also a separate work stream. I think Colin talked about it, Pete talked about it. We're not gonna have all these different components.
We're really gonna simplify the architecture because simpler is better. Even if we don't do that, and we take the most pessimistic case, we're gonna need 8 million wafers, okay. The industry is gonna scale to 200 million by that point. Still, we're less than 5% of the market share. If we run the same offense we ran all along and continue to work with our partners and, you know, invest in fabs and invest in capacity, we'll have the capacity ready by the time we need it. I really don't see this as an impediment from a supply chain perspective to get to 20 million.
I would like to talk about a page that we took out of the electronic playbook and applied it to the mechanical side. Our vehicles use currently the heat pump-based technology for the thermal system. Before heat pump, our vehicles used to have a dispersed thermal system connected by hoses. Engineering was inspired from circuits of circuit boards created a system which was extremely integrated, where 60% of the system was the size of a basketball, where it contained about 100 components, 50 sealing interfaces , and several different manufacturing processes all were packaged into this tight size. Why are we talking about this? Well, we launched the system right during the outbreak of the pandemic in Model Y. The team was forced to manage 100-plus components from different suppliers around the world remotely.
There are four videos playing at the same time. One is the video of the manual line that was created when we started the production. This manual line produced a lot of quality issues, let alone throughput. We quickly, you know, pivoted to a semi-automated line. In spite of that, although the throughput was great, the quality issues sustained. Our team decided that fully automated is the only way to go. Before we automate the line, we created a 3D simulation.
Not many companies or teams have this capability, where at this point, now when a part gets spec'd, we have the ability to create a 3D simulation of all the complex sub-subassemblies that would be involved in making the part and tell the suppliers exactly what equipment to purchase, what to test, and create an end-to-end line in simulation. That's what we did here. With the help of the simulation, we had a very detailed subassembly cell, which was juxtaposed with the suppliers' factory layout. We after a back and forth of iteration and optimizing the processes, we implemented this fully automated line in less than 8 months. The results were remarkable. 99% reduction in labor force.
In a line that would need 1,000 associates to build the Supermanifold in full volume now just takes 10 associates. The quality also went up by an order of magnitude. We have a defect rate of less than 0.05%, and a Supermanifold rolls out every 7 seconds because we have created many automated lines, and these factories are the size of 2 football fields, by the way.
Roshan had to one-up me on that one. 99%, 2 football fields.
Why are we talking about this? In a world where labor costs are increasing, labor participation is decreasing, there's a huge turnover, and the cost of implementing automation is going down by 3x in the next 10 years. This is how we are thinking about complex assemblies, and we are going to push this type of thinking into our supply base more and more, who will be building our complex assemblies. This is the only way we think we can confidently scale to the 20 million vehicles target. To summarize, we have laid the foundation in getting ready to execute Master Plan 3. Our strategy is supply chain design simplification. We are going to make sure that we have more control into the tiers of the supply chain. We are going to grow responsibly and sustainably with our long-term partners, and we'll automate.
Yeah. It's like, you know, going from 0 to 40,000 cars a week was tough. I mean, it took a lot of trial and error. It took a lot of learning. There was a lot of pain. There's a lot of mental scar tissue, as Elon calls it, through that process. But now that we're at that level and the foundation is set, going from 40 to 400. It doesn't really phase us. We have a capable team. We've got capable external partners that have, you know, gone through hell with us. Our leadership team is galvanized behind the mission, and we're gonna get it done. I'd like to thank you for your time and hand it off to Drew and Tom for the next presentation. Thank you.
Hi, everyone. My name is Tom Zhu. I'm now responsible for global production and sales and delivery service. I joined the company back in 2014, and I've been running the company's business in Greater China and APAC. Today I'm representing all the Gigafactories and talk about how we can make more cars and faster. Right now, we have four vehicle factories on three continents to serve our markets. The total number of manufacturing employee is over 65,000. This is also including our amazing team in Reno. They are making powertrains, battery packs and drive unit for our vehicle factories. The total installed capacity in all these vehicle factories, about 2 million cars a year. That said, we're always looking for opportunity to grow more capacity from the existing footprint.
You expect this number will grow over time. To build more factories is the start to building more cars, and we're certainly the expert of it. This is the before and after image showing you the progress we made back in 2019 when we built a Gigafactory in Shanghai. In 9 and a half months' time, we built a greenfield project to establish the vehicle factory. 3 months later, we start deliver our first customer vehicle. People always ask me, you know, how Tesla can build a factory that fast? Really, we learned from our Fremont factory a lot. We talked to the survivor from the production hell and try to avoid all the mistakes we made.
We decided to design a straight line with minimum number of up and downs and turns for easier manufacturability and easier construction. We also challenged all the assumptions that people ever know to build a vehicle factory, when we delete and simplified all the redundancies and the buffers, that's helped us to save a lot of time. Also, we have a very strong in-house construction team in Tesla. If there ever a job cannot be done by others better and we bring this in-house. We have this in-house construction team have a full control over all the activities on site from design engineering to construction site management. This team not didn't just build Giga Shanghai, they also built Giga Berlin, Giga Texas, and Giga Nevada.
They're really the hero behind the scenes. Like Lars shared earlier, going forward with the new platform, the more modular design and densified Gigafactory, we're be able to make more cars from even reduce the footprint. That also means we can build more factories at the same time. With the joint effort of all manufacturing teams around the globe, I'm happy to share early this morning we hit the 4 million mark for total Tesla ever built. The 4 millionth vehicle actually built in this factory, when you took a tour on our shop floor, you probably walked past it. Took us 12 years to build the 1 million vehicle, and about 18 months to build the 2 million.
The third million took us 11 months, and just shortly, less than seven months, we built the four million cars. We're getting better and faster.
Exponential growth.
Really kudos to the team. What it takes to ramp a Gigafactory? Well, if you have 600 robots, 10,000 trained employees, or 5,000 human and 5,000 Optimus, and hundreds of processes, you can do it. Sounds simple, but it's extremely hard. There's two key metrics that we predominantly focus on is the overall equipment effectiveness and the cycle time. In Tesla, we're setting the passing grade for our vehicle factories with 90% OEE and 45 second cycle time. What that means, the OEE really evaluate the equipment uptime, the machine performance, and the quality. Simply put, this is the actual production time on a good quality product versus the planned productive time. The higher, the better.
The 45 second cycle time, that means, you know, like you expect every 45 seconds there's a car rolling off the final assembly line in the factory. The faster we ramp, the faster we can get to economies of scale. If you look at the chart on the right, on the left, Shanghai be able to significantly drop our labor hours per car during the ramp. The little dip there happened in the last Q2 2022 because of the COVID shutdown. On the right is the Fremont Model Y shop. Even this is a 6-year-old facility, the team there still be able to optimize the material flow, eliminate all the single point of failure, and drive higher outputs, hence, reduce the labor hours.
Actually, this factory keep setting new record. Yesterday, they just had a new factory daily record.
Congrats, Fremont team.
Absolutely. How can we reduce the cycle time, keep improving efficiency? We follow the philosophy Elon shared about building rockets, which is questioning, basically find the right problem to solve. We start with delete, simplify, then we try to accelerate, pressurize the line, find if the solution actually work. Last, we think about automation. There's one example in Gigafactory Shanghai, in the paint shop, we find that there is a overlap baking range between the PVC sealer and the topcoat. It was done by two different ovens. We decided to combine the two processes. Eventually that helped us, didn't just help us to reduce the cycle time, but also save the 9% of energy consumption and 9% of the CO2 emission.
Also, we commonized the S and X design in Fremont. Right now we have a common body and common headlamp that we eliminated about 40 parts and reduced about 10% of the cycle time. All these improvements really help us to get to a faster ramp trajectory. We didn't just learn from these high-volume existing factories, we also learn from the new factories. Most recently, our Giga Berlin factory team implemented a 5G private network on the shop floor, helped us to overcome about 90% of the overcycle issues for a particular process in GA shop. We're gonna soon implement this globally.
Right now we have an integrated global organization from production all the way to sales and delivery service. This will help us to strengthen feedback loop between manufacturing and the service. We obviously wanna bring a delight for customer ownership experience to our car owners. With this direct feedback loop, we be able to turn customer escalations and feedback into quick actions improvement on the shop floor immediately. With that effort in the past 6 months, we've been able to reduce the time in service, and early ownership service, also service appointment wait time significantly. 4 factories, not enough.
If we wanna hit the 20 million car a year target, we're gonna keep building new factories, new lines, and then we'll come up with new product as well. With all Gigafactories now in the one organization, we'll be able to replicate unified ideas across factories. We will help the new factories to ramp faster, and also produce better quality at a lower cost. I'm going to hand over to Drew to talk about cell manufacturing.
Thanks, Tom. Yeah, we're not just talking about new vehicle factories, but also new cell factories. we're gonna talk about how we follow the same model when we do so. But first, I thought I'd just provide a little bit of an update on cell production. Remember Battery Day, we showed this video with the spoon, and how we went from dry powder to film. Let's just say there's no spoon now. Many of you in this room saw this on your tour today, but, you know, here is our dry electrode machine here in Texas, one of the lines we have installed here. Fully automated, no spoon, from powder and foil in to coated electrode out.
From a peak productivity perspective, per tool perspective, this is over 20 times the productivity of the tools that we showed folks on the tour in Kato back in Battery Day. We've made a lot of progress on one of the key parts of the cell manufacturing process. We've also continued to focus on refining the way we make cells and the factories that the cells go inside. As you can see on this chart, from typical 2170 cylindrical cells to 4680, we made a huge leap, which is basically a 5x reduction in the factory footprint, and volume and footprint. From going from what we did in Fremont in our pilot line to Texas, we improved further, and we're improving again when we go into Nevada.
What this actually represents is a series of actions taken by a very integrated, holistic design team across, you know, the product design, the manufacturing design, the process design, the equipment design, and the facility design. They all need to work together to make this happen. You can see on the bottom, just as an example of simplicity, investment down and scaling up, you know, parts, we've reduced the number of parts in the cell, the number of processes, significant reduction in the number of processes.
Collaborating with those five design groups, we've been able to result, you know, with a factory that's 10 times smaller volumetrically, which means it's faster to get built, it's much lower CapEx per gigawatt-hour output, and we can go and scale, you know, to our objectives of, you know, 240 total TWh, 1 terawatt-hour a year of stationary storage and 20 million vehicles a year, with the scalability that is required to achieve our goals. We're not just looking at the cell factory itself, but also upstream materials where necessary. What you see here is the rendering of the 50 gigawatt-hour a year Corpus Christi lithium refinery that we've already broken ground on here in Texas.
The facility will start commissioning by the end of 2023 this year. This is a good example of something where we're basically talking about breaking ground and starting commissioning within 10 months and with actual production within 12 months. That's the target. Similar to what we did in Shanghai. Again, the result of collaboration and, you know, internal execution of construction in partnership with local communities. This site is 30 minutes from the Corpus Christi ports, located directly on rail. The process route we're taking is a direct soda ash leach of the input material, which means there's no acid roasting. We don't have any of the sort of waste products associated with an acid roasting step.
We're designed to consume lithium spodumene, which is a very commonly traded lithium rock, but it is flexible to other feeds from primary and secondary sources. Similarly, we're working on our cathode facility here in Texas, maybe you've seen it as you're driving around, which is a 60 gigawatt-hour a year cathode facility behind the main building here. 10-month build time there as well. The equipment's being installed for the first line as we speak, and we'll be commissioning starting next quarter. We are doing this where we need to. Our plan is not to do it always.
There are lots of competent companies out there. We are also trying to sort of accelerate the pace of the industry by trying some new things that are a little bit more scalable, and de-risking certain innovations that improve productivity per, in terms of CapEx per gigawatt-hour and things like that. With that, I think that sums up our efforts on the factory side. Mike, you wanna come up and we'll talk about energy?
All right.
All right. Take it away.
All right.
Here, you.
Hello, everyone. My name is Mike Snyder. I lead our Megapack organization. I've been with Tesla for almost nine years. I'm excited to talk to you today about the Megapack product and the business and some of the exciting things we have ahead of us. From the beginning with this business, we have always focused on building successful projects for our customers, and not just the batteries in the box. We've built a hardware and a software platform that is able to adapt to environments all over the world and that is able to scale from small island projects to large gigawatt-hour scale batteries.
We've invested our time and our talent to best understand every aspect, every step in the process, and every risk of a project to ultimately provide a solution that is thoughtful, intuitive, provides great value to our customers, and is as plug-and-play as possible. Yeah, just looking at these projects, it's incredible to see them all come together in one montage.
Seeing the future.
Yeah. Yeah.
Compressed.
It's 10 years of building these projects, and it's incredible to see the impact and to see what we have in front of us.
Absolutely.
We're on our sixth generation of our industrial product, which is the Megapack XL that we're building out of Lathrop, California. We've deployed over 16 gigawatt-hours of industrial and residential products across 50 countries. Really, Megapack is a market leader. It's best in class in efficiency, reliability, energy density, and easiest to install, lowest cost to install.
We call it XL because it is actually the like largest, heaviest object that you can transfer around the roads of the world without, you know, having to shut them down and get crazy permits and things like this. It is extra large and there is not gonna be an XXL, let's just say it that way.
There's been incredible demand for the project or for the product. 2023 is gonna be a great year. We have gigawatt-hour scale projects being built. The Lathrop factory continues to scale and ramp up. There's new products on the near term roadmap. A lot to look forward to. How did we get here? Number one, a maniacal focus on all aspects of delivering stationary storage. What do we mean by that? What are some examples? As you see here, the Megapack enclosure, for one example, we designed the wire ways directly into the base of the enclosure. Typically, you need to install a bunch of conduits and cables under the ground. Now you can lift them up and install them directly into the base, increasing the speed of deployment and lowering the cost.
We see the batteries and the power electronics. The power electronics is really the beauty and the elegance of the Megapack. This is what converts the DC power to AC power on the grid. The brain is built directly into the product. It is able to connect to any grid in the world right out of the box. It allows it to all the Megapacks on a given site to work as one unit. Really, it's what makes it what it is. I think it's the most incredible feature of the Megapack. As you build out the full site, what we end up with is the most energy dense solution on the market, upwards of 300 MWh per acre.
Just as a frame of reference, this solution is two times more power dense than a typical gas peaker plant. This is the future. This is where we're going.
This is the product that retires the fossil fuels.
Yep.
One power plant at a time.
We talk about power electronics a little bit more. Tesla is a leader in power electronics. We've deployed over 1.4 terawatts across energy storage and vehicle, and we deliver more power electronics than the solar and the wind industry combined on a per annum basis.
Power electronics, it's hard to overstate how impactful they are. They are really the glue in the sustainable energy economy between generation, storage, and the end use. You know, those power electronics devices are switching, you know, thousands of times per second, hundreds of thousands of times per second, to efficiently react to whatever is needed, either in the car or in the grid. Because they are so like, sort of software driven at their core, they can provide functionality that hasn't been available to the grid in the past. It's one of the reasons why renewables and storage together are such a great solution.
Yep. One reason that we focus on power electronics and control so much is because of the impact that it has on the projects directly. Just as an example here on the left, we have a firmware feature that we call Virtual Machine Mode. What Virtual Machine Mode is, it contributes to grid stability like a car shock absorber that dampens oscillations or vibrations and keeps the ride smooth. You can imagine if you don't dampen those vibrations, the vehicle could lose control or someone could get hurt. On the grid, you could have a blackout.
We have one grid operator that's utilizing Virtual Machine Mode and said they will not operate their grid at 100% renewables unless they have this feature, unless they have Virtual Machine Mode working.
Yeah.
They tried and it didn't work very well.
Why is it Virtual Machine Mode? It's like synthetic inertia, where you turn the battery power plant into, like, through software behaving like it's a giant spinning machine, literally. And that like inertia stabilizes the grid.
Right.
You don't need a giant spinning machine. You don't need like a huge fossil fuel power plant or a giant hydro turbine. You can just have the battery do it. You can program it to whatever you need at that part of the grid, or even have it be dynamically changing as the grid conditions change.
Right. That's why we focus so much on it. We believe it's the future of how batteries are gonna add value. It's gonna be more about adding power stability just as it is about energy.
Yep.
Yep. Autobidder is another software feature I wanted to mention. Autobidder is an autonomous energy trading platform that in its most basic sense, is buying energy low and selling it high, and the owner can net the difference of that. Operating battery storage is actually very complex. Needing upwards of hundreds or even thousands of decisions every 5 minutes. It's much more complex than a PV, a solar plant or a wind plant or a thermal generation plant. It's mostly because of the versatility and the ability of the battery to provide so much value in different ways that these decisions need to be made in real time in order to optimize and get the most value out of the battery as it can.
There wasn't a solution that existed in the market, so we built one ourselves. In the markets we've deployed them in Australia, in Texas, in the U.K., it has proven to be a market leader. We're continuing to invest in Autobidder as we expand to new markets. Second point of how we got here, a relentless focus on speed of execution. There's two points to be made here. First one is in building factories. What you're seeing here is a time lapse of the mega factory that was built in Lathrop. We took it from a JCPenney distribution center to a world-class manufacturing facility in less than 12 months, which is incredible. Really, the way that we did this is similar to what my colleagues have described. It's leveraging the vertical integration of Tesla.
It's getting the vehicle manufacturing team in the same room as construction and engineering, and making decisions quickly with all the decision with all the stakeholders and decision makers. We're using the lessons learned from Lathrop as we speak, as we plan for our next factories. Second, it's about installing projects faster. I've mentioned this already, but plug and play is kind of at the core of what we're going for. Over the past four years, we've increased the installation speed by 4x, and we've reduced the total labor involved across both construction and manufacturing by 3x. We think that this is key to unlocking our ability to scale and our customers' ability to scale with us. We need to be laser-focused on reducing that time from when the Megapack leaves the factory to when it is operational on the grid.
It's not just about centralized storage, it's also about distributed storage, the future roadmap of, well, at least here at Tesla. Maybe some of you know about this sort of retail plan we have here in Texas, and I just wanted to talk a little bit about how that all fits together. Tesla Electric really unlocks the full value of distributed energy and storage products. It enables our customers to become their own utility. The data on the page is from our South Australia virtual power plant, another Tesla Electric setup that we have. Over 2022, it's 5,000 customers that we have the data for. You can get a sense for what's happening.
Just if they were being provided default utility service and you looked at the cost of serving them that electricity, on average, $140 a month. If they all had solar and Powerwall, but the solar and Powerwall wasn't interacting with the grid, wasn't participating in the energy market, the cost of serving that customer would halve, go from $140 a month to $70 a month. If Tesla Electric is operating those assets in an intelligent way to benefit the grid, Tesla Electric being basically a software that we developed out of Autobidder for the purposes of these distributed energy resources, we can actually pay the customers to bring their energy services to the grid. That's what happened in Australia last year.
Australia's a little bit of a special case. They're they represent the future, South Australia in particular. Solar and wind supplied 70% of South Australia's energy in 2022. That compares to 30% in Texas and 35% in California. This is an indication of where this is all headed, of both centralized and distributed storage resources, providing the, the key to unlocking fully renewable grids, and Tesla Electric is the retail plan that we're using to bring that to bear for our, our customers that have our products. You know, this is how we're rolling it out. First, as it exists, it's available for in Texas for people who have Powerwalls in their homes. We need to extend beyond that.
Over 1 billion people live in markets with competitive retail electricity, there's over $2 trillion in annual energy spend in these markets. This is a huge opportunity. Our intention is to bring this to, you know, market by market in the same way that we've approached Tesla Insurance, where we can bring value to our customers to reduce their total cost of ownership of our products. Actually, what we're going to do next, and this is pretty exciting here in Texas, by this summer, we're going to offer a retail electricity plan to people who have our cars where they can have unlimited overnight home charging for $30 a month.
This is part of reducing the total cost of ownership of our vehicles. The reason why we can do this is because Texas has a ton of wind. In Texas, the wind blows at night. Actually serving these customers with electricity at night for their cars is the best thing to do for everyone. This is a way to incentivize people to charge at home at night directly from renewable power. It's part of the grand master plan we talked about at the beginning. We're very excited about this, and we do see this similar to Tesla Insurance, as I said, as further reducing the total cost of ownership of electric vehicles. Putting this all together, we're really at the beginning of this massive ramp in energy storage deployment.
Yeah. Yeah, reflecting back on the Master Plan, we talk about tens of TWh needing tens of TWh of stationary storage. We have our sights set on annual production rates of one terawatt-hour, which is 25x our capacity at Lathrop. In the near term, we see really strong demand for the Megapack product, over 100 gigawatt-hours per year, in 2023, and growing by over 100 gigawatt-hours per year over the next few years. The demand is there, and naturally, as we continue to focus on cost and speed and value, the things that we've mentioned, it's clear we need to build more capacity, and we need to ramp it quickly. While the challenge is big, it's also a huge opportunity for Tesla.
It's a huge opportunity for the Megapack business. I'm excited to see the impact that it's gonna make on our grid transition.
As am I.
Yeah.
All right.
All right. Thank you.
Thank you all. Yes. All right, Laurie, Brandon, come on up.
Hi, I'm Brandon with the legal team. Thank you, not just for coming, but being investors. We have an active and engaged board and management team. We've met with you, and today is the culmination of that. We've heard you, which is why we're excited to share this investor day with you. You will hear more from our board at the appropriate time, but for now, I'm excited to introduce Laurie Shelby, who's gonna discuss sustainability and our employees. Laurie?
Thank you, Brandon. Hi, everyone. My name is Laurie Shelby, and I lead the Environmental Health, Safety, Security, and Sustainability team. I've been with Tesla for 5 and a half years. I'm going to talk about Tesla's impact. Of course, that starts with our awesome team members. You've heard a lot today about our amazing products, our process, our software. You've seen lots of numbers and lots of graphs. You've also heard a common theme, and that common theme is that Tesla's team is what drives our success and makes a difference. Whether you're in our factories, in our sales service and delivery centers, our warehouses, or in the field, you can see and you can feel the collaboration that's happening, and you also can see the pride that everyone has in working for Tesla.
Before I go on, I want to give a big shout out and a thank you for our Tesla team globally. That was great. That was great. Thank you. We are creating a lot of meaningful work through our jobs. We're a huge employer, and we're growing every day. We're at 129,000. More than half of our team works in vehicle manufacturing. Almost 60% of our team is based in the U.S. People want to join our mission, and we need to grow, and we need to hire the best people. Engineers want to work for Tesla. Look at number one and number two, SpaceX and Tesla. In 2021, we got 3 million applicants. 3 million. That's incredible.
Our culture, our culture at Tesla is anyone at any time can raise an idea or make an improvement suggestion across safety, people, cost, production, quality. Because we have a very simple process called Take Charge. Take Charge is really employee engagement. As Take Charges go up, as you can see in this graph, injuries go down. You know why? People who do the work, the people who do the work know how to improve the work. Sustainability, sustainability is everything. It's our mission. You've heard about it all day, part of our Master Plan. At Tesla, we don't just tick boxes to meet a requirement. We really focus on the impact that our products and our software will deliver. In 2021, our customers avoided 8.4 million metric tons of greenhouse gas.
That's equivalent to taking 1.7 million ICE vehicles off the road for a full year. Yes. Yes. We're also focused on doing really good work and driving sustainability into our operations. We have year-over-year reduction in our water use, in our waste, we're growing and building a renewable energy program within our operations, just like supercharging, as you heard. It's not just about our products, it's the way we make our products, both phases. Tesla, our company, is solving both sides of the clean energy equation. We've got the clean generation, the clean storage. We also have zero emission transport. As this slide shows, from 2012 to 2021, our products generated more energy that was consumed by all Tesla vehicles and our factories. It's amazing. Yeah, I like the claps. Yeah. I'll take claps. Our vehicles use less emissions than gas-powered vehicles.
Even in the worst-case scenario, 100% coal grid, the Model 3 has less emissions than the average ICE car. We're only going to get better. You heard about that today. Every product we make is more efficient. Every factory we build is more sustainable. It's only going to get better and better. As we build our zero emission future, we've been talking with our stakeholders, with our shareholders, about how do we improve our reporting framework. We are aligning with the TCFD. We're aligning with the TCFD, which I know many of you have been asking us about. I also want you to stay tuned because there's going to be much more coming in our impact report. Thank you all. Up next is Zach Kirkhorn. Thank you.
Thank you, Laurie. All right. This is our last section of our prepared remarks today. We'll kind of take what we've discussed today and summarize it at the corporate level for our financials. Maybe before I jump into the details here, we've never done an investor day as a company. We've never brought our leadership team out and asked them to talk about the things that they're working on. You know, I feel very fortunate to be able to work alongside this group of people and support their teams, and I also just want to thank them, they're in a room in the back listening to this, for all of the work that they put into today and getting the company to where it is today. Thank you to the Tesla team.
When we were preparing for today, folks were asking, "Well, Zach Kirkhorn, what should we talk about?" Really the only guidance we gave folks is, "Well, talk about the things that you're working on and talk about it in the context of the Master Plan." What did almost everybody talk about today in great detail? All of the work that they're doing to take cost out. Because in this industry, in this business, you survive or you die based upon the ability to manage your costs. I'd like to talk about cost as well. If we look at our longest-running scale production product, this is the Model 3. We reached 5,000 cars per week, which was our design capacity, in mid-2018. Since then, we've taken 30% of cost out of this product. There's two points I wanna make about this.
The first is that cost reduction, as you have heard throughout the course of the day, is deeply ingrained in our culture. I think one of the most important reasons why we are here today as a company. The second point that I wanna make is that when we're working on cost reduction, it's easy just to take cost out and make our products worse. We have to take cost out and improve our products at the same time. This is the hard thing to do, but it's the necessary thing to do to continue to move forward. If you look at a version of a Model 3 that we've built today and you compare it back to a Model 3 that we built in 2018, there's a long list of improvements that we put into the car while also taking cost out of the product.
Cost reduction doesn't just come from one place. There's no silver bullet here. We pushed the boundaries on volume with the Model 3 program, increasing volume 3x over this period of time. We've improved productivity. In our Fremont factory, we are twice as productive now as we were in mid-2018. We've made a lot of progress on overhead efficiencies and product improvements, as we've discussed, and a long list of other things, including localization, our factory in Shanghai. As we ramp up volume and as we find efficiencies, we work with our suppliers to do the same, and that leads to material cost reductions that improve the affordability of our products. As we look forward to our next gen vehicle, our target is to reduce 50% of cost, and we've talked about that a bit earlier today.
Going from the Model S and X platform to the 3 and Y platform, we took out 50% of cost, so the task here is to do it again. This is very important because as we improve affordability, the number of customers who have access to our products dramatically increases. As we link this back to our Master Plan, it enables an exponential growth in our volume with linear reductions in the cost of our products. The second point I'll make here is that, and again, cost reductions don't come from any single one place. You can see the buckets here on the vehicle side, battery and powertrain, manufacturing cost reductions, and others. These buckets are relatively equal in size. In order to take 50% of cost out of the product, we have to go through everything.
More importantly than just the cost of the car up front, when we're transitioning to a sustainable economy, particularly with vehicle ownership, it's really important to think about the lifetime cost of the products. This chart here is showing what the total cost of ownership per mile is over the course of 5 years, and we have to think about financing costs, insurance costs, the cost of power. Drew talked a little bit about that with the plants that we're doing in Texas, wear and tear and maintenance on the cars, et cetera. We're already at a place today in the U.S. where a base Model 3 on a cost per mile basis is less than a Toyota Corolla, which is the highest selling car in the world. As we move towards our next gen platform, we will continue to reduce this.
As we work on robotaxi variants of this platform, this cost will come down even further. This is a product that we expect to have substantially lower cost per mile than the highest volume products in the world. The next thing on cost reduction I want to talk about is operating expenses. We spend a lot of time on our earnings calls talking about gross margins and product cost. We don't spend a lot of time talking about OpEx, but I think this is one of the really important parts of our story. Since 2018, we've reduced our non-GAAP OpEx 60% as a percent of revenue and 75% on a per delivered car basis. This is quite staggering, and I want to go into the details about how we've done this.
The first component of OpEx, which is R&D, the single most important thing here is that we constrain the number of programs that we work on, minimizing the number of programs that we're running in parallel. The key here is to maximize the revenue and cash generated from every program that we work on and also maximize the technology sharing between the programs. We talked a little bit about how power electronics in our energy business is shared with power electronics in our vehicle business. If we look at the SG&A side of this, I think the story here is pretty dramatic. We're showing here a comparison of SG&A per car delivered compared to the traditional auto industry, which is both an OEM and the dealership network. Our estimates are that we're 60%-70% lower on a per car basis.
If you turn this into dollars, this is many thousands of dollars per car. This is part of the reason why on the last earnings call, I made the comment that we think about operating margin more so than gross margin as a company because the integration that we have here on the SG&A side, particularly with vertical integration into our dealership network, this provides efficiencies that give us a competitive advantage from a cost structure standpoint relative to other companies. The other thing that I'll note here is that this wasn't always the case. It wasn't that long ago where these bars would be inverted.
If we go back a little bit in time, and we talk about when we launched Model 3 in the Fremont factory, we went from 2,000 cars per week, which was S&X, adding 5,000 a week to that for Model 3 got us to 7,000 cars a week. You've heard some folks here today talk about production hell. What I don't think we've ever talked about as a company is back office operations hell, which also happened at the same time. 7,000 cars a week, 350,000 cars a year, we want to go to 20 million cars a year, and we're struggling at 7,000. We're saying to ourselves, we have to completely rethink the way that we're managing our back office operations if we have any hope of efficiently scaling this company.
This began a process that we're still continuing to work on today, but it's something that we refer to as the Tesla operating system. In the same way that we're vertically integrated into the software that manages our cars, we've also vertically integrated into the software that runs our company. All of our major departments in the company are using, not all of them yet, but almost all of them are using our in-house software. Most recently, we removed third-party recruiting software, and we're now running our own recruiting software as a company.
This matters a lot because rather than having a complex web of third-party systems that are generic solution that take a lot of effort to customize for our particular needs, we've been able to put in place with an in-house applications engineering team dedicated lightweight software that does only what we need it to do and nothing more. That team works very closely with process improvement teams that sit within the business who are looking through all of the processes that we're managing as a company and following the exact same process that Tom mentioned in the manufacturing space to look at our processes, delete process steps that we don't need, simplify process steps that are existing, and automate what's remaining. The results of this have been quite staggering, to be frank. I think it's exceeded our expectations as well.
On the left side of this chart are some examples of the efficiency improvements that we've seen within our SG&A areas as a result of this strategy. The North American sales team is 4 times more efficient now than they were in 2018. Order operations team, financial services, I mean, these levels of improvement are large. There's not enough pixels on the slide here to list out everything, it's just a couple of examples. In the same way that when we look to take cost out of our products, we want to feature up the product or improve the product while taking cost out, the same has been true with the use of Tesla operating system and that integration with our process improvement teams within the company. We've increased performance and added capabilities while also dramatically reducing cost on a per-car basis.
There's a list of self-service functionality that we've put in place, which customers like. It also simplifies our operations behind the scenes. We've used the Tesla operating system to expand into our captive insurance space, which we are fully vertically integrated into. We run our own in-house software. We have our own agents, our own claims software, and claims folks who run our insurance business. This is also very important because as a technology company, data is very important. We use these systems and the access to the data in those systems to pull out extremely granular and targeted reporting that enables us to see every aspect of our business in real time so we can make adjustments to our operations as needed. There are sometimes examples where we say, man, I wish we were tracking this piece of data.
Well, you know, it's a teams meeting away from the applications engineering team to make a request, and then we can start tracking that data and make those changes. The feedback loop associated with process improvement inside the company is pretty astounding. The last thing I'll just mention here, which is something I'm personally quite proud of, we've made a lot of progress in our closed processes within the finance team, getting our 10Q and 10K filed. It's amongst the fastest of the 20 largest market cap companies. Great work to the finance team here. All of this work on cost reduction is extremely important because we have a lot of money to spend ahead of us to achieve our goals within the master plan.
We've mentioned over the course of the day today 20 million annual vehicle production as our target, 1 terawatt-hour of annual energy storage production, and then expanding cell production service and charging in line with the growth of those other businesses. We've estimated what we think the total cost to get there will be, and there's certainly error bars around these numbers as we continue to progress and innovate. Of this, we've spent about $28 billion of that so far in the history of the company. Maybe this total investment looks large. I actually think it's quite small relative to our ambitions. If you look at our 2022 operating cash flows and you just say, well, let's assume some modest growth to that, maybe not all that much if you're being conservative.
The ability to pay for this level of investment from the forecast that we have is very achievable for us. I'm often asked about capital allocation. I think our strategy here is very straightforward. I touched upon it a little bit in our last slide. Just to be explicit about it. Obviously, the priority here is to ensure that we're using capital to run the day-to-day operations of the business. This is a working capital intensive business, and this has to be managed very carefully. You know, over the course of a quarter, there can be $multi-billion swings in our cash balance up and down just depending upon the timing of when we build and deliver cars. We obviously have to have money set aside for that. Downside protection is another really important area where we have money set aside for that.
Over the last couple of years throughout the pandemic, I think this just highlights why being ready for downside protection is important. Preserving daily operations is the engine that then generates cash that allows us to reinvest into the growth of the business. I'm very proud that over the last couple of years, despite the ups and downs in the macro environment and interruptions to supply chains, we have never once pulled back on our investments in growth. That remains a huge priority for us. We do then after investing in growth, we look at opportunistic ways to spend our cash. These are more financially oriented where if a return exceeds a hurdle, then we can place cash in that. This is part of our story of cost reduction in the Fremont factory.
On the excess side, you know, this is where, you know, the board meets on a regular basis and thinking about what our cash flow projections are looking forward and whether it makes sense to do buybacks or dividends. To wrap up the finance portion of the presentation today, a couple of key messages I just want to leave you all with here is that we use innovation at an intense level to drive cost down and improve the efficiency of the business. The reason this is so important is it allows us to improve product affordability. This is particularly the case when we move to our next generation platform. Improving affordability allows us to comfortably make investments that grow volume, that volume generates cash that then allows us to make more investments.
As we integrate that going forward, you know, it's our belief as a leadership team here at Tesla that we're going to achieve unprecedented scale in the manufacturing space. This is what's ultimately required to accelerate the world's transition to sustainable energy. With that, this completes the prepared portion of today. We're going to take a short break and then we'll resume with a live Q&A. Thank you all very much.
Oh, I can talk.
All right. Well, let's see. This is probably the most significant announcement of the day is that we're excited to announce that we're going to be building a Gigafactory in Mexico. Yeah. We'll have a, you know, obviously a grand opening and, you know, groundbreaking and whatnot. We're excited to announce that the next Tesla Gigafactory will be in Mexico near Monterrey. Super excited about it. Now I do want to emphasize, we will continue to expand production at all of our existing factories, so including California, Nevada, here in Texas, obviously, and Berlin, Shanghai. We intend to increase production at all factories.
The Giga Mexico would be supplemental to the output of all the other factories. This is not, to be clear, moving output from anywhere to anywhere. It is simply about expanding total global output. Yeah. It's gonna be good.
Peanut gallery.
Anyway, we're, you know, we're quite excited to announce that. In fact, I believe the governor is here. It's hard to see. Welcome. Secretary of State, I believe. Anyway, it's, we're looking forward to it and having a big event for the great groundbreaking and opening. Let's see. With that, we can move to I guess Q&A. We've obviously got significant bench strength here. Should probably have like a bench. Anyway, We maybe have too many people on stage, but we'll try to answer questions within reason.
You know, this is meant to be kind of more of a long-term sort of discussion as opposed to, you know, what will be the production for the rest of the quarter type of thing. Let's try to orient our questions towards, long-term value creation. With that, fire away.
Rob?
Hi, it's Rod Lache with Wolfe Research. Very exciting plans about the next generation vehicle and powertrains and batteries. I was hoping you can maybe talk to us a little bit about the timeline for deploying this. It sounds like it's more than just a vehicle. This is a kind of a paradigm change on how vehicles are assembled, how batteries are put together and everything. Does that also just get reconfigured into everything that you do? The Model Ys that are being built here will be built very differently in the future. Maybe just give us some feel for what happens from here and what's the timeline for implementation.
I'll talk a little bit about that, but broadly speaking, the most profound architectural changes will be in future vehicles. Retooling a factory means bringing the factory down for an extended period of time. We prefer not to do that, I think. There are variants in how Model Y is produced. We've got variants where there's a rear casting, where there's a front and rear casting, and we have the structural battery pack. Then there are a number of smaller improvements that occur. I think for really, really big changes, those would be future vehicles. Yeah, I don't know. You guys want to add? Lars, maybe?
Me? Yeah. I mean, as far as I agree with you 100%, Elon. It's really easy to put innovations in new vehicles, but long term, we'll obviously bring them back. We've always talked about that. We don't want to take our factories down. As far as the timeline goes, you know, we're going to go as fast as we can left to right. As always, you know, Elon alluded to the fact that Mexico will build our next gen vehicle. We will also be doing that in our other plants. It's really about getting them all up and running. We expect that to be a huge volume product and, yeah, we're gonna move that quickly over the next couple of years.
Let's go to Adam.
Thanks. First, enhorabuena, Mexico, vamos. That's great. Congratulations. Elon, a question on applying first principles thinking and innovation to an area that up to now has seemingly been outside of your control, and that is on mining and extraction of some of the key materials. I believe two years ago, you had a patent on sodium chloride to extract lithium from some clays and spodumene clay and things of that nature. How does that fit into the plan of maybe bringing real innovation into a mining sector that could use a little, you know, maybe waking up and getting those costs down? 'Cause that could be a real gating factor, it seems.
Well, we're gonna address whatever we think the limiting factor is at any point in time. We would like to do the least amount possible. We don't want to get into the mining or refining sector. We will do that if we have to. I do think the focus really should be on refining capacity. You know, we need to make just a very giant amount of anode, cathode, lithium hydroxide, lithium carbonate. It's really the refining capacity that is the biggest choke point. Yeah. That's why we're building a lithium refinery in Corpus Christi.
In terms of the mining companies that are out there, looking at that part of the value chain, we do have large suppliers of lithium right now, and they are aware of, you know, how we're approaching the Corpus refinery and the technologies we're trying there. The reason we're making them aware of it is because we think they're fundamentally more scalable. As we prove them out, we plan to share that with them, 'cause as Elon says, like, it's not really like we want to do these things. We're doing them because it's not happening fast enough. If we can prove that it can be done faster, the intention is to transfer that knowledge to our large, our current suppliers. The same is actually true...
It was actually a clay process that we were playing with and we continue to work on. Same is true on that. We've worked with our suppliers as well on trials and we're sharing knowledge there. The intention is just to help the whole world do this better. Ultimately, this here is getting the lithium out, or whatever it is, out of the ore.
We're obviously building a cathode processing facility just adjacent to this building. A little further down the road, you'll see another large construction. That's for cathode refining. Like I said, we'd really prefer if others did that. We're doing it 'cause we have to, not 'cause we want to.
Yeah. In that case, there just isn't really any large scale cathode production in the United States, and it needed to be done. Again, if we're gonna do it, we're gonna try to do it from a first principles perspective. We have tried a bunch of new things there. We're confident that they will work. As they prove out, again, we want to bring them back to our suppliers, so that they can build new facilities more quickly with less in-investment.
All right. Let's go to Ben.
Hi. Ben Kallo from Baird. Similar, on the renewable side, is there more that you can do in Tesla to nudge the rest of the renewable industry to speed up since it's such a big pillar of the Master Plan 3?
I mean, I don't know what we can do. I can say that if there's like entrepreneurs out there that want to have like a guaranteed chance of success, it would be refining lithium or anode and cathodes or any materials whatsoever for lithium ion cells. It is a no-brainer. I think, I mean, we're doing everything we can.
Yeah. On the renewable energy thing, the thing that we can do at Tesla, is the more we reduce the cost of storage, the more we reduce the cost of stationary storage, and the more we bring like flexible load to the grid in the form of like cars charging at the right times, the more valuable renewables are. Because in like Texas right now, I wasn't joking, like there's a lot of wind, and at night wind is there's like almost too much wind 'cause they don't wanna turn the nukes off or whatever. Bringing really low cost storage onto the grid makes renewables more valuable, which ultimately will accelerate their deployment. That's how we focus on it, I guess.
Yeah. Philippe?
Thank you. It's Philippe Houchois at Jefferies. I've got two questions. The first one, when I think about in this industry, everybody wants to be Tesla. Every carmaker is trying to emulate what you're doing. Well done. The one thing they don't do is focus on having as much growth as possible with as few models as possible. I'm just trying to understand, as you aim for 20 million units in 2030, how many models do you think you need to get there? How does it fit into your drive to hyperscale? How do you manage this? Also the fact that probably consumers at some point don't wanna see a Tesla or the same Tesla at every street corner. I'm just trying to get your sense of that. My other question is on bi-directional.
I mean, you talked a lot about making, you know, the world more renewables and better usage of cars. I think bi-directional charging is one way of better using cars. You seem to be reluctant doing that in the past. I'm just wondering what your latest views are on the topic.
Sure. On bidirectional, it wasn't like a conscious decision to not do it. It just wasn't a priority at the time, I think is maybe the way to think about it. As we continue to improve the power electronics in our vehicle, we've found ways to bring bidirectionality while actually reducing cost of power electronics in the vehicle. As at all things Tesla, the goal is usually to get more for less. We are in the middle of kind of like a power electronics retool, I would say, that will bring that functionality to all of our vehicles over the next, you know, 2 years, let's say. But it's... Yeah, I guess that's how I'll say there.
I don't think very many people are gonna use bidirectional charging unless you have a Powerwall, because if you unplug your car, your house goes dark. This is extremely inconvenient.
Yeah, most of the value comes in charging the car at the right time. It's not really about sending energy the other way.
I mean, if you have a Powerwall that can take the house load, then you can use the your car as a supplementary energy source to the Powerwall, and then, you know, you're not gonna drive everyone crazy by unplugging your car and having the house go off. I think there's some value there as a supplemental energy source down the road where if you have a Powerwall, you've not diminished the convenience of the people in the house.
The question of number of models?
Sorry, the what? Oh, how many models?
Vehicle models.
Not that many. Really 10? I don't know. Not that many. I mean, what's happened with conventional cars is people have run out of things to do. When you run out of things to do, they just end up reshuffling the deck, and you have pretty much the same... I mean, how many variants of car are there on the road? There's, like, hundreds. Are they good variants? No, mostly not. They're just variants for the sake of variants. Look at how have things converged with the phone? I mean, there used to be hundreds of flip phones. Now what do we have? It'll be like that.
George?
George Gianarikas from Canaccord Genuity. Just had a question about your plans to grow market share in China. Also whether or not the political tensions between United States and China impact your long-term ambitions there. Thank you.
I don't know. Hey, Tom, don't sweat too hard.
Well, we're still growing our market share in China, quite strongly actually. Especially, you know, early this year, we had this price adjustment. After that, we actually generated a huge demand, more than we can produce really. As Elon said, as long as you're offer a product with value at affordable price, you know, you don't have to worry about demand. That's basically the philosophy that we follow. We try everything to cut cost from supply chain, from improve the efficiency in the factory, and pass down that value to our customers. I think, yes, as long as we're continuing to do that, I'm not too concerned about the market share in China.
The second part, geopolitic, no one knows. We do our best. We actually create a lot of jobs to the local community, and we Also, with at our suppliers' factories, we create a lot of jobs as well, and we contribute a lot to the local economy. I think, you know, as long as we're needed in this country, I don't see there much of a risk of that.
Colin?
Colin Rusch from Oppenheimer. You know, one of the things I was struck by in the presentation was the operational efficiency metrics that you guys are talking about. Can you talk a little bit about the velocity of learning cycles and how you guys track that and think about that as an organization as you work into a variety of other areas from an innovation perspective?
Anyone wanna try that?
I mean, I'll say one very high-level thing, which is you can't improve something you don't measure, and we're, like, ruthless measurers at Tesla. Once you start measuring things that contribute to operational efficiency, you actually have a path to doing something about it. I guess the key is a good measuring stick.
Okay. Something I should say with respect to demand, which I've said a few times over the years, but it sometimes needs to be said again, which is, overwhelmingly, the desire for people to own a Tesla is extremely high. The limiting factor is their ability to pay for a Tesla, not do they want a Tesla. It's easy for people in this room to lose sight of that. If your income is far in excess of what a car costs, then you look at value for money, but you do not consider affordability. For the vast majority of people, it is affordability-driven. This is why we cannot simply double the price of the car.
Or you could say, and think about things in the limit, where if you had an, you know, a infinitely desirable car, but it costs $10 million, it wouldn't matter because people do not, cannot, most people do not have that. Demand is very much a function of affordability, not desire. Very important. One of the things we weren't sure about was the price elasticity of demand for Teslas. Like, as we lower the price, how much does demand increase? We found that even small changes in the price have a big effect on demand, very big. That was a good thing to learn. Yeah.
As the sort of autonomy question is very, very big, 'cause you could potentially have, I don't know, 5 times the utility of the asset that you currently have. So if passenger car is 10 or 12 hours a week of usage, plus a lot of parking expenses, an autonomous car could be 50 to 60 hours a week or something like that. And you could get rid of a lot of parking expenses. So, you know, if this is true, then as autonomy is effectively turned on for the fleet, it may be the, probably will be the biggest, asset value increase in history overnight.
Emmanuel?
Thank you so much. Emmanuel Rosner from Deutsche Bank. As you start launching this next generation vehicle and ramping up volume, what will be your nearest term priority in terms of segment or vehicles focus? The slide that you showed with two vehicles on the wrap seemed, based on form factor, one of them maybe looks like a van, another one looks like maybe like a smaller vehicle, like potentially a Model 2, I guess. What is the nearest term focus for you in terms of ramping up the next gen vehicle? How do you make sure that by lowering the price point so much because the cost is going down 50%, you're not cannibalizing demand, you know, for your existing vehicles?
I mean, demand for our vehicles in terms of desire to own them may as well be infinite. It's indistinguishable from infinite at this point. Affordability is what matters. As you make the car more affordable, we will have demand go crazy, basically. The issue is how do we build the cars? The hard part is building the cars. I cannot emphasize that enough. The hard part is building the cars and the entire supply chain that goes with the cars. This is a logistics challenge of extraordinary difficulty. All the things that have to go into the car have to scale with the car while everything is doing an exponential ramp. If you miss even one of those things, doesn't matter why, earthquake, flood, fire, revolution, I thought I've heard them all.
I mean, it's, any part of that supply chain gets interrupted, you're now, then you have a seizure. The hard part is building the cars by far and the supply chain that goes with it. I mean, do you guys wanna talk about supply chain stuff?
Yeah, they might.
I think, you know, in the presentation I'd mentioned, perfection is a passing grade. We really need everything to happen perfectly. The strategy for mitigating the different risks, some of which were anticipated and some weren't, is really bespoke to the situation. Really, you need subject matter expertise and a deep knowledge of the particular supply chain you're managing to come up with those strategies. We've seen everything from, as Elon mentioned, you know, first the tariffs that flew back and forth, then the ocean logistics issue, a chip shortage, COVID, floods. There was a fire in a fab in Japan that knocked down. There was a massive COVID spike in Malaysia, where a lot of the chips, the back end of a lot of chips was done there and more.
This was one that Elon was involved with as well. Yeah, it's nerve-wracking, somehow it works.
Yeah, although it's difficult, I think we have been laying the foundation to be as intimate to our, all the tiers of the supply chain, building that, control and you know, directing the different tiers of supply chain. That's the way we are trying to mitigate the risks that come with it. You know, again, dual sourcing, triple sourcing, having redundancy, is the way we've been trying to mitigate for it.
Which segments...
Yeah, on that point, when we think about vehicles, when you think about 3 Y as an architecture, SX as an architecture, our next generation platform is more than one segment. Really, we're thinking about all the segments that are available that we haven't captured and where the market would be, and designing it with our supply chain partners so that we can go quickly through those segments for where we need. To Elon's point, if you make a car desirable and affordable, you know, oftentimes it doesn't necessarily matter what segment it's in because it's one that you want. We've seen that with Model 3 when a lot of people thought the sedan was not gonna be a great hit, but we sell tons of them.
The next generation platform is not 1 vehicle, it is multiple, and it's on the segment that we will, you know, really try and focus on the affordability and desirability point moreover than where we started.
e's like, I think, you know, there's an old saying, like, battles are won with tactics, wars are won with logistics. The logistics challenges here are enormous. And when you start, like, being a very significant percentage of an industry, you can't over-capacitize. It's not realistic. And some of these things, like you say, like, well, you dual source or triple source, you can do that maybe for small things, but you can't do it for big things 'cause if you're triple sourced and one of... It's like having a plane with three engines, where if any of the three engines fails, you crash. It's like, you know, so.
You have to either over-capacitize, which drives your CapEx up, and has idle suppliers somewhere and big warehouses, or you design to some overage that's, you know, reasonable, and then you have expedite costs because inevitably there's something goes wrong somewhere, and you gotta fly things around. It's really just the rate of progress is the rate at which we are able to scale a 10,000 logistics problems. Most significant of those is the cell production. We actually deliberately try to overdo cell production or cell supply to have that exceed what is needed in vehicles. Because if it goes below what's needed in vehicles, then the factories stall. What do you do with all these extra cells?
It's like, well, okay, the much easier thing to scale up and down is Powerwall and Megapack output, stationary storage. We can then over-capacitize in cells and packs and scale production of stationary storage, which is much easier to scale than vehicle production. That's strategically, I think, a good thing. Yeah, I mean, the CapEx for Megapack is tiny compared to CapEx for vehicles. And also Megapack demand is quasi-infinite. It basically, as long as we are competitive with utilities, we can sell as many. Like, it's, yeah, right. Quasi-infinite demand for that really, for multi-tera, many TWh. We got a long way to go to get to many TWh per year.
Dan. Let's go to Dan.
Thank you. Dan Levy, Barclays. I think we know competitive dynamics differ significantly by region. We know cost dynamics differ by region. We saw that with your China Gigafactory, far superior cost to what you had at Fremont. I realize it was a new factory, but there are clearly different dynamics. To what extent are the cost strategies that you've laid out today, do they differ by region or is there more of a global one-size-fits-all approach on reducing cost?
Who wants to talk? Or who's next?
We're pretty consistent under the strategy here. We'll try to as much localize it as possible. In China, 95% of, over 95% of our supply chain are localized, all the way from the first tier to second tier and then to the third tier. That give us generate a tremendous savings on costs. We also have very localized labor force. We have access to skilled labor force in the region and Yangtze Delta region in Shanghai, I'm particularly referring to. Among over 30,000 of employees in China, we have probably less than 20 expats.
It's a, you know, very deep localization we have done there, to be able to have such cost structure. We try to replicate this approach elsewhere in different Gigafactories. Of course, you know, supplier base are different. The, you know, labor market are different, region by region and country by country, but we try our best to localize. Yeah, and I think that's absolutely give us an advantage to compete with the auto OEMs around the globe. Another thing is, you know, we have this direct selling mode, like Zach said, you know, we save a tremendous money on OpEx as well. We have full control over all expenditures across the company.
You know, we don't spend money on marketing or advertising. You know, everything we saved will become, you know, the value we can offer to the consumers. On that part, we also follow pretty consistent strategy over the years.
I mean, certainly the manufacturing stuff that Colin, Pete, and myself talked about, whether you make it here or you make it in Europe or you make it in Asia, it applies everywhere. We think about that from the ground up, not specific to any region.
Yeah, just one thing I would add, to Tom's point about localization and to Lars' point here. I would actually say that we're moving in a direction of more standardization in terms of our factories and our processes and our cost reduction approaches. As we've been thinking about the next generation platform, we've been thinking about the volume that we aspire to build against that, how many individual factories do we need to build, and what is the fastest possible way to expand that footprint around the world? What we've done with Model Y is each factory is an incremental improvement of the previous factory, which requires engineering hours, engineering spend for each factory design, and then you end up with factories that are slightly different from each other.
As we move towards the next generation platform, I think the term you used, Lars, is copy-paste. To get it right from the first time, you know, certainly there will be things that we learn and we make adjustments to, but try to have as much standardization and commoditization as possible, which allows us to go as quickly as possible with expansion.
Overaka, on this side.
Thanks. Two questions, one for Zach and one for Ashok. For Zach, how do you think about the long-term growth and operating margins for the business? When we look at $150 billion-$175 billion in CapEx, is the remaining balance of that, is that essentially the CapEx guidance through 2030? For Ashok, the multi-trip reconstruction that you highlighted, how is that different from what Mobileye does with the REM map that they have? Can you give us any color on the AP4 hardware?
With respect to operating margins, you know, we intend to continue to improve operating expenses as a percentage of revenue over time, and we're continuing to take cost out of our products, and try to keep gross margins at a place that's healthy as well. You know, from the hardware perspective of the business, you know, it's our expectation that we'll continue to stay in a healthy place over time. Then there's a software portion that's added on top of that. Elon has commented on this many times about the impact that Full Self-Driving software can have on the economics of the business. In that space, you know, profitability, operating margins would be impacted dramatically.
Specifically to your point about CapEx guidance, you know, the intent of that slide was not to provide specific guidance, but to just be transparent about internally what our rough estimates are and to provide context that we think getting to 20 million vehicles per year and 1 terawatt-hour of energy storage is very feasible relative to our expected cash generation for the business. You know, that's a number that will move as we learn. We may choose to vertically integrate more into things. We may find efficiencies elsewhere. To the whole conversation we were having earlier about do we do mining, do we not do mining? The entire point is to say that this is something that's entirely possible based upon our forecast.
Regarding the multi-trip reconstruction, my point I was trying to make was that we want to auto label most of the data. We can collect data from the fleet and then reconstruct local areas that can act as supervision for those clips. Since we want to build a scalable self-driving system, we don't want to really rely on HD maps or anything, even though we could, like, trivially build something using the same technology. This is mostly for auto labeling, and this provides, like, precise 3D labels for all the video sequences involved in the reconstruction.
Thank you.
Something that's, I don't know if we touched on this much, but in terms of training, we have one of the biggest neural net training systems in the world, and that, we expect to increase that capability by an order of magnitude by the end of this year, and probably another order of magnitude by the end of next year, with some combination of NVIDIA and Dojo. That much of the scale of that is quite daunt on people. It's enormous.
Maybe the final two questions, one from Alex and one from Chris.
Alex Potter with Piper. I definitely wanna ask Drew or anybody else up there, for an update on dry battery electrode, right? If you're trying to overcapacitize towards cells, scale a lot of this. Clearly there's a lot of moving pieces, and it's a complex sort of orchestra with the supply chain, but a lot of it comes down to dry battery electrode, at least to me. How are you trending? That was the most fascinating part of the factory tour for me, and looking through that window and seeing that this is clearly not a science project. Anything that you're willing to disclose, yields, progress, where you are today versus where you thought you were going to be. Thanks.
Sure.
Yeah, I mean, as you saw, right? Like, it's this is a real factory making a lot of dry electrode in an automated fashion. We've made a lot of progress. It's a spectrum. Like, we're perfectionists, and we have clear end goals. We are, every week that goes by, making progress towards those end goals, whether it's speed of the tool, yield of that process or the downstream process. We haven't stalled out yet on the rate of progress either, and that's both on the anode and the cathode side. I think the great thing about where we are with the overcapacity that Elon mentioned is it's giving us the opportunity to
Experiment as we go, rather than just being, like, stuck to something that we happened to kick off a year and a half ago. It is here. Something that Elon has said to the team many times is, "It's okay to scrap equipment or money. It's not okay to scrap time." The way we've been approaching it is probabilistically, what do we think is the most likely thing to succeed? Even actually in the factory here, you saw more than one anode line that they are actually operating two slightly different versions of the final process step of the powder entering the tool, and it's a competition. You know, which is gonna be higher yield, which is gonna perform better.
We have the luxury to be able to do that, and we're taking full advantage of it to advance the technology as quickly as possible.
The dry electrode problem is really quite a hard problem. I mean, we acquired Maxwell really just for the dry electrode technology, but this just illustrates what a gigantic gap there is between something working at small scale and at large scale. We've had an extremely talented team of engineers working on sort of scaling the dry electrode process and having it be reliable and consistent. We're, we've been grinding hard, literally and figuratively, on this for quite a while. It seems likely that we will be able to scale it to volume this year. Yeah.
Yeah. I mean, we're basically increasing the output week over week, like, roughly 1K a week per quarter is our internal target. We're, you know, we're tracking to that. I think what you saw is effectively, like, a ton of material per hour per tool. It's kind of, like, hard to, like, rationalize, like, what that really means. It's different than, like, oh, something works in a lab. Well, when it's tons of material per hour, there's just different kinds of problems. Even if you have, like, 0.1% escape of, like, fines into the enclosure that your equipment is in, now you have a dust problem, and you, like, that could short out some electronics.
It's, like, just things like this, where when you're on a lab scale, you don't even notice it, but when you're doing thousands of tons over the course of months, it's like, oh, a new failure mode we found, and that's where we're at. We're knocking those out, though. The team is grinding through it, but progress every week.
The last question from Chris.
Thanks for taking the question. I have a few follow-ups on the next gen vehicle. First, when do you think we'll get a look at it, maybe a prototype? Second, are there any details that you think you can share in terms of the size, the content, the performance? And then third, I think you mentioned that you would produce it in other plants in addition to Mexico. Should we take that to mean that you can launch it at an existing plant before you're finished constructing the new plant in Mexico?
I think we'll actually have to probably decline that answer. We will have a proper sort of product event. That would be jumping the gun if we were to answer your questions. Maybe another question if there's. Yeah, yeah. I don't know. Anyone?
Will Danoff.
Hey, I have two questions. This has really been a very impressive afternoon. Thank you so much. Elon, I'm curious, as you've doubled, and you'll double again, how, what you've learned about sort of managing a larger enterprise and what you might have to, you know, do to manage a bigger enterprise. Secondly, I'm curious on your thoughts on how, you know, generative AI and, you know, these rapid breakthroughs in AI in the last months could help you make cars sort of, you know, less hard to make. Thank you.
I don't see AI helping us make cars anytime soon. At that point, at no point any of us working, I think.
Big, big problems.
We should all just chill out. I mean, I'm a little worried about the AI stuff. I think it's something, I don't know, which we should be concerned about. I don't know. I think we should need some kinda, like, regulatory authority or something that's overseeing AI development and just making sure that it's operating within the public interest and, you know, it's quite a dangerous technology. I fear I may have done some things to accelerate it, which is, I don't know. I mean, some of the AI stuff I think is just obviously useful, like what we're doing with self-driving, which is, you know, some people think is an AGI-type problem.
I don't think it's quite an AGI problem, but it's certainly requires very sophisticated neural nets, because the road system is designed for eyes and biological neural nets. Naturally, the analog to that is cameras with digital neural nets. Yeah, the thing we've we found is it's just there's a If you actually look at our neural net architecture in the car, it's kind of insane, frankly. Nets upon nets upon nets upon nets.
Yeah. The visualization tool crashes when you open it.
Yeah.
It's, like, so complicated that no one can visualize this right now.
Yeah. Just, like, literally, it looks insane. I guess something like that's happening in our brains while we drive around, which is pretty wild. You know, I don't know. Like, Tesla's doing good things in AI. I don't know. This one stresses me out. I don't know what to think or say about it. You know?
Okay. Given it's 7:00 P.M., I think that's all the time we have. Thank you very much for coming.
Thank you.
I'll see you later.