Elon Musk's New Product is the Biggest Bet In Industrial History

What? Elon Musk's new product is the biggest bet in industrial history, and it depends entirely on one small town in the Netherlands. This plan is so crazy, in fact, that when I posted about it on X about how much this project hurts my head, Elon Musk himself replied to the post agreeing with me. So, in this video, I'm going to walk you through why this project is so crazy and ambitious, and if successful, how it will completely transform everything, literally everything. The way we live, the way we interact with technology, the types of products and services that will exist on the planet, and how it will transform not just our planet, but the solar system.

Let's start with the project, which is priced at $25 billion. That's how much Elon Musk is spending on Terraab, which, per Elon, is "the most epic chip building exercise in history by far."

"We have a profoundly important announcement to make, which is, uh, the most, uh, epic chip building exercise in history by far. This is really going to take things to the next level."

China has launched a number of rockets on the second day of military exercises around Taiwan.

This will be a joint venture between SpaceX, Tesla, and XAI. The goal is to build 200 billion custom AI chips per year and 1 terawatt of compute power annually, 80% of it going to space in the form of orbital AI satellites.

Now, I've been covering Elon's companies for over 14 years, and this is easily one, if not the most ambitious plan that he's ever communicated. But when I spent some time really analyzing Terraab, something didn't add up. Because the numbers Elon threw out on stage would require machines that almost nobody outside the semiconductor industry has ever heard of. It's a machine so complex, so difficult to build, that there's only one company on Earth that makes them. They produce about 50 of these super advanced machines per year, and every single one of them is already spoken for for years in advance.

But before we get into that, let's understand what Terraab actually is, because it's not really as simple as it sounds. The joint venture for Terraab between Tesla, SpaceX, and XAI is under one roof. The initial target is 100,000 wafer starts per month, and the full-scale target is 1 million. And to put that in perspective, TSMC, which is the company that makes chips for Apple, Nvidia, AMD, Qualcomm, basically everyone, their entire global output across all their fabs is about 1.4 million wafer starts per month. Elon wants one factory that's doing 70% of that alone.

Now, this $25 billion factory is going to be building two types of chips: inference chips for Tesla vehicles and Optimus robots, and what he's calling the D3 chip for orbital AI satellites launched by SpaceX. Now, these factories are meant to have everything under one roof: design, lithography, fabrication, memory, advanced packaging, testing, and mask making. Each one of those steps is super complex, especially lithography and mask making, and I'll make sure to explain more about them as we go through the video.

Now, the crazy part about this entire project is that all the current fabrication facilities on Earth produce only about 2% of the chips of what Elon's companies will need across all his projects. And you will quickly see that the Terra project alone will not be enough, because that little town in the Netherlands that I described earlier is by far the biggest bottleneck. This is the big problem.

Every single chip at two nanometers requires a machine built by one company in one small city in the Netherlands. A machine that took 25 years and over 10 billion euros to develop. A machine that weighs 180 tons and contains more than 100,000 components from 5,100 suppliers across 15 countries. That machine is called an EUV lithography system. EUV stands for extreme ultraviolet. The company that makes it is called ASML, and this company is headquartered in Veldhoven, Netherlands. The population is about 45,000 people. And from that small city, ASML controls the single most critical bottleneck in the entire technology industry.

So, what does this machine actually do? And why is it so hard to build? And why is this such an important piece of the puzzle? The YouTube channel Veritassium did an unbelievable job explaining in detail how these machines work. But I really want to go through the summary in this video as well, because that is extremely important to understand why Terraab is so crucial. I'll make sure to put a link for that very video in the description below.

Now, to understand this whole thing, you need to know what chip making actually involves. Every chip starts as quartz sand, which is silicon dioxide. You purify that sand into a single crystal ingot. You slice it into wafers thinner than a fingernail that are insanely smooth. And then you print incredibly tiny circuit patterns onto those wafers using light. The smaller you can print them, the more transistors you can fit, and the more powerful your chip becomes. And for decades, the industry kept making the light shorter in wavelength to print smaller patterns: visible light, then ultraviolet, then deep ultraviolet. And each step was very hard.

But at some point in the late 1990s, the industry hit a wall. Deep UV light, which is 193 nanometers, couldn't print small enough for the next generation of chips. They needed a massive leap to even shorter wavelengths. That leap was over two decades to figure out, which is where we are today with EUV lithography machines.

Now, these machines print the patterns on silicon wafers that become the transistors in your chips. At two nanometers, we're talking about features so small that conventional light can't print them. You need extreme ultraviolet light, 13.5 nanometer wavelength. Now, for context, visible light is around 400 to 700 nanometers. So, we're talking about 30 to 50 times shorter.

Now, the insane thing about EUV is at 13.5 nanometers, everything absorbs the light. Air absorbs it, glass absorbs it, water absorbs it. You can't use lenses at all. The entire optical system must be reflective with mirrors, and the entire machine must operate in an ultra-high vacuum, so there can't be no air or anything. And those mirrors are actually freaking crazy. They're made by a company called Zeiss in Oberkochen, Germany. I don't even know if I said that right. Each mirror is polished to picometer flatness. A picometer is 1 trillionth of a meter. If you scale the Zeiss mirror to the size of Germany, which is about 900 km across, the tallest bump on the surface would be less than 1 millimeter. These are the smoothest surfaces ever created by humans. And each mirror takes about 18 to 24 months to fabricate. So, you can't rush this. You can't build a second factory and go faster. The physics of polishing at that precision just takes time, at least with today's technology.

Now, how does ASML actually generate 13.5 nanometer light? Their solution literally sounds like science fiction. A high-powered laser fires at tiny molten tin droplets 50,000 times per second. Each droplet gets vaporized into plasma hotter than the surface of the sun. Then that plasma emits EUV light. The mirror collects it, it focuses that light, and then projects the chip pattern onto the silicon wafer. 50,000 tin droplets per second, laser vaporized into solar temperature plasma, reflected off picometer flat mirrors into a vacuum to print a transistor smaller than a virus. What the... I'm just going to repeat that again, 'cause it's so freaking crazy. You're firing a laser at tin droplets in a vacuum, creating plasma hotter than the sun, bouncing that light off of mirrors that are the smoothest objects ever made by humans to print patterns that are smaller than a virus on a thin slice of purified sand.

Every single iPhone, Nvidia GPU, AMD server chip, Qualcomm modem, Tesla FSD computer, all of it depends on this process working perfectly billions of times per day across every fab on Earth. That's what the entire world is running on. It depends on that technology working.

Now, each of those machines cost between $200 and $400 million each. Each one weighs about 180 tons, which is the weight of an empty Boeing 747 plane, at roughly the size of a city bus. When ASML ships one, it takes 20 trucks and three cargo planes just to move the parts. And the installation takes months, and the calibration takes more months. It took ASML from their first EUV prototype in 2001 to 2019 to ship a machine that could reliably produce commercial chips. That's 18 years of development after they'd already been working on the concept for a decade before that. 25 years and 10 plus billion euros to build one type of machine. That's why nobody else can make one, or even wants to make one. It's not that they don't want to. They literally can't. The institutional knowledge, the supplier relationships, the process expertise, it can't just be copied from a blueprint. Apparently, these are things that just can't be copied. And ASML has over 15,100 suppliers across more than 15 countries that are feeding everything into one machine. A single company called Zeiss provides the optics, and they're the only ones who can. If Zeiss has a bad quarter, the entire world's advanced chip production slows down, because of course that's going to happen. That's one single point of failure in the most important supply chain on the planet.

In 2024, ASML shipped 44 EUV machines globally. In '25, they shipped 48. In '26, they're forecasting 60, and in '27, maybe close to 70, including some next-generation high-NA systems. So, call it 50 to 60 machines per year, trending up very slowly. And every single one of those machines is already allocated. TSMC, Samsung, Intel, they're booked for years. You can't just call ASML and order 20 machines. There's no spare capacity, and there's no second source either. ASML is it. And with the insane demand in AI, the backlog for these machines is only going to increase.

Okay, now let me show you the math behind Terraab that literally made me stop and recalculate this freaking video 10 times. At two nanometers, each chip requires 15 to 25 EUV lithography layers. Each EUV tool can process about 45,000 wafer starts per month per layer. If you do the math for Terraab's initial target of 100,000 wafer starts per month, you'll need 30 to 75 EUV machines just to get started. In a full scale of 1 million wafer starts per month, you need 300 to 500 EUV tools. 500. There are approximately 400 EUV machines installed on the entire planet right now. And Terraab, at full scale, would need more machines than currently exist on Earth.

And then the cost: at an average of $300 million per machine, the initial phase would be $9 to $22 billion just for the EUV machines. And in full scale, it would be over a hundred billion. That's 4 to 7 times the entire announced budget for Terraab for one type of equipment. And in the timeline: even if Tesla ordered machines today, the lead time is 18 to 36 months. And ASML's entire production for the next 2 years is already accounted for. Terraab might be able to receive a handful of machines by 2028, if they're lucky, maybe 10 by '29. So, 100,000 wafer starts per month is realistic, maybe by 2030, but a million is not going to be this decade, or even next decade, maybe.

So, the obvious conclusion is: if we're talking strictly about EUV lithography machines from ASML, this can't work. The math just doesn't add up. It's literally impossible, especially from competition from everyone else like TSMC and Nvidia and Intel. They're not going to be the only ones getting these machines. Elon is promising something physically impossible right now in today's supply chain, given the supply constraints. Literally, every semiconductor analyst that's looking at these numbers, they're all saying this is the craziest thing I've ever heard. And Elon says a lot of crazy things.

Now, this is where the Tesla bears and haters will step in and say, "Remember Battery Day in September 2020? Elon promised his 4680 cell revolution with 10 gigawatt hours within a year, 3 terawatt hours by 2030." And then five and a half years later, Tesla hit maybe 2% of their original Battery Day target. And this is a perfect place for the same crowd to say this Terraab thing is complete and utter nonsense. The semiconductor supply chain has an unsolvable bottleneck.

But forget about the whole SpaceX thing, being told the same exact thing when it was first founded, and it launched, and they ended up building their own engines, their own avionics, their own everything. But forget about that. The argument here is that lithography is potentially even more complex and difficult than rocket science. Literally.

Now, I need you to pay close attention, because this is where the story completely changes. And I spent way too much time digging into this freaking Terraab since it was announced. And I really want you to understand what it actually is versus what people think it is. And I think I found some pretty insane conclusions that not only make Tesla's goals seem real, but seem quite feasible.

Now, before we move forward, about 80% of you that regularly watch my videos aren't subscribed. And if you enjoy my videos, the easiest way to support is by clicking the subscribe button on YouTube. It's completely free and it goes a very long way. Thank you so much.

Okay, so a lot of people are trying to frame this entire project as a direct competitor to TSMC, which is ASML's biggest customer for chip making. But that's not what Terraab actually is. At least from my estimation, Tesla's going to be much closer to a design iteration factory and an advanced packaging factory. This is what Elon actually pointed towards. And yes, over time it's going to do volume production of chips. And the long-term version is going to be 100,000 to a million wafer starts. But the part that gives Tesla an immediate structural edge with just a handful of machines, potentially, is the iteration and packaging, which are two radically different things from volume lithography, but they'll have just as big of an impact, if not even bigger. And I'm going to walk you through both.

First is the iteration. When Tesla, or really any company, looks to design a chip today, the process goes something like this: Tesla's engineers design the chip on their computers, and that takes days, maybe hours, maybe weeks. Then they send the design to TSMC in Taiwan, who happens to have a ton of ASML machines. They have roughly 60% of the world's EUV lithography capacity. And then, once TSMC has a design, they make a mask, basically the stencil that prints the chip pattern. That takes about 2 to 4 weeks. Then TSMC runs the wafers through fabrication. At advanced nodes, that's roughly two to three months for a full wafer run, because you're processing 50 to 100 or more layers, each taking about a day. Then they ship the test chips back to Tesla. Tesla tests them, finds bugs, and redesigns them. And then they go back to step one. That total cycle time is roughly 3 to 4 months per iteration. A complex chip needs 5 to 15 iterations before it's ready for mass production. So, you're looking at well over a year to finalize one chip design, and in many cases, two to three years.

Now, let's picture this process with Terraab. Tesla's engineers design the chip revision, which takes hours, days, whatever it might be, with AI assistance, who knows? Then, instead of sending it to TSMC, they walk the design down their hall to their own in-house mask shop. Then they make the mask right there. Then they run targeted wafer lots on their own lithography tools. That might take days to a couple of weeks for the specific layer that they're testing, not months for full production. And if you find a bug, you go back to step one. And that entire process takes 1 to 2 weeks per iteration instead of 3 to 4 months. That means Tesla could do in a few months what TSMC customers take years to accomplish. The compression from months per cycle to weeks per cycle is massive. Perhaps even conservatively, Tesla's iteration speed could be five to 10 times faster than the traditional foundry loop. That's not going to be a small difference. That's a structural competitive advantage that compounds over time.

Now, here's a key detail: for prototyping and iteration, Terraab doesn't need 500 EUV machines. It needs maybe two to five, if that. They might not even need the EUV machines. They might even get by with the cheaper, less advanced lithography machines that don't need to go as small as the EUV ones. The mask shop is what makes rapid iteration possible. Without it, you're shipping designs to an external mask maker and waiting weeks. But with it, you've collapsed the single biggest bottleneck in chip development from weeks to days.

Now, the second piece is advanced packaging. This is the part that solves the volume problem without needing hundreds of ASML EUV machines. The old way of building a powerful chip is monolithic. You have one giant piece of silicon, and you're trying to cram everything onto one die. A bigger chip equals more area equals more chances for a defect to land on that chip. And at two nanometers, a monolithic chip the size of a large GPU might have 30 to 40% yield, which means 60 to 70% of what you manufacture goes in the trash. That's insanely expensive.

The new way, and this is the direction the entire industry has been moving, is chiplets. You build multiple small chips, each one focused on a specific function. So, maybe your compute cores can be at two nanometers with the EUV machines, but your memory can be at five nanometers, and input/output can be at seven nanometers. Each chiplet is small enough that yields jump to about 80%. So, if you get a bad one, you just throw it out and simply snap another good one in place, like a Lego brick. And I'm obviously massively oversimplifying here, but it massively increases the yield of these machines, which means way more chips per run. And for this packaging method, the magic is how they connect, not in each individual piece. That's what advanced packaging does. It connects chiplets together with thousands of microscopic links. These are high-bandwidth interconnects that let the chips talk to each other at speeds that approach what you'd get if they were one monolithic piece of silicon, but with way better yield and way more flexibility.

Now, why does this solve Tesla's ASML problem? Because only the compute cores need two nanometer EUV lithography. The memory chips can be made on older five nanometer equipment. The IO chips can be made on even older seven nanometer equipment. Those older nodes use cheaper DUV lithography. And you wouldn't need ASML's EUV machine, which is extremely expensive and extremely limited. So, DUV instead of EUV. And instead of needing EUV capacity for the entire chip, you need it for maybe a quarter of the silicon or less. The rest would use readily available, much cheaper equipment.

Now, there's a real-world proof point for this that's already played out. In 2019, AMD switched to a chiplet architecture. They moved to TSMC for manufacturing while Intel was using their own fabs. But the chiplet packaging architecture was the key differentiator. It let AMD deliver competitive performance at better yields and lower costs. AMD went from under 5% server market share to north of 30% by '25. Their stock went from about $30 to $200. The packaging architecture was worth more than the transistor technology. Or look at Nvidia. They use the same packaging technology, and they are literally the most valuable company in the world.

Now, all this considered, Terraab does plan to do its own in-house fabrication as well. That's the long-term vision with a full fab. But in the medium to short term, while they're ramping up and before they have hundreds of EUV machines, hopefully, or whatever lithography machines, the architecture could work like this: You start by supplementing in-house production with TSMC and Samsung making some of the two nanometer compute chiplets. You do have to remember that Elon is very closely working with Samsung in Texas to power the next generation AI chips. Then you use cheaper external fabs for the five nanometer memory and seven nanometer input/output chips that don't need EUV. Then Terraab does what nobody else wants to do under one roof: that stacks the chiplets in 3D configurations, connects them with high-bandwidth links, and tests the integrated system, iterates designs rapidly using the in-house mask shop, and ships the finished AI systems. The ASML-intensive volume work can be shared with external foundries who already have the machines, while Terraab owns the design intelligence and the packaging integration.

Okay, so at this point, you might be thinking, "That's clever, but why does iteration speed actually matter so much?" Because of what those iterations optimize. Think about it this way: Nvidia makes general-purpose GPU chips. And these are incredible chips. They really are. Think of these GPUs as a restaurant kitchen. You have ovens, stove tops, prep stations, deep fryers, grills, warming drawers. They can cook any meal. Their chips have incredible flexibility. But when Tesla runs inference, which is the specific computation that powers Tesla's FSD for their self-driving cars, as an example, or Optimus for their human or robot project, a significant chunk of that Nvidia kitchen sits idle. There might be a video decoder Tesla doesn't use, or graphics renders that serve no purpose for running a neural network. But Tesla still pays for those transistors. They still consume power, and they still generate heat.

Now, imagine Tesla designs its own chip at Terraab. They'll do iteration after iteration, each cycle faster than competitors can even get a mask step back from their foundry. And in each generation, they identify these transistors might fire during inference, but they never contribute anything to Tesla's specific AI model. So, we need to remove them. Or these cache pathways are too long for how Tesla's architecture accesses memory. So, shorten them. This logic block handles a feature no Tesla model ever freaking uses. So, delete it. Give me more of what I actually need. And after many rapid iterations, Tesla's chip becomes a custom kitchen. It has dramatically more of what inference actually needs, which is, let's say, the ovens, but no deep fryers. And the pantry isn't down the hall anymore. Memory is freaking integrated directly into the chip via some sort of packaging. And so, it may very well have the same number of transistors as the Nvidia chip, but a far higher percentage of them doing useful work for Tesla's specific models.

This is literally Apple's entire strategy. This is why their machines are far more powerful and power-efficient versus PC competitors, because Apple's entire strategy was to marry hardware and software. And Tesla's doing exactly the same thing, but bringing everything in-house. This approach gives Tesla three massive advantages if successful.

First is the power efficiency. And this is the metric for Optimus, which is Tesla's human robot. Every watt the chip burns is a minute less the robot can operate. A 50% more efficient chip means the robot works for much longer per charge. It would have the same battery, but way more runtime. And then the iteration removes the waste. Each cycle finds more transistors that fire but don't contribute. So, you remove them, and that's less heat, less power, and longer runtime.

The next is latency. Now, for humans or robots, milliseconds matter a lot, especially if you're reaching for something that could be dangerous. Each iteration will shorten the critical data pathways on the chip, and a rapid iteration compounds into significant latency reductions over time. That's the difference between a fluid robot and one that's freaking just doesn't know what it's doing.

And lastly, the throughput. More computation per chip means fewer chips per robot means lower cost per unit. This is how you get from the current cost structure to the $2 per hour robotic labor cost that changes everything about manufacturing. That $2 number per hour sounds crazy, but the global market for human labor is north of $40 trillion per year. And so, if you can build a robot that works 20 hours a day for two bucks an hour with custom silicon that's been iterated into near-perfect efficiency for its specific task, the economics become undeniable.

Now, for the part that should make Tesla's competitors nervous if Terraab is successful, which is: the disadvantage widens over time. It doesn't narrow. In year one, Tesla's custom chip is maybe 40% more efficient than Nvidia's for inference, but by year two, that's 100 plus iterations. By then, competitors might be on their second Nvidia generation. That would be already a 60% gap. And then by year three, Tesla's chip could be co-evolving with its AI models. And the chip shapes the model, and the model shapes the chip, and the gap just keeps getting bigger. This is a very weird place for their competitors. Boston Dynamics, Figure, Unity, Xiaomi. Most of them are going to buy Nvidia general-purpose chips off the shelf. They won't be able to iterate weekly. They can't remove features their specific workloads don't need, at least nowhere near as quickly as Tesla could. They can't reshape the cache for their specific model architecture. They're stuck with the restaurant kitchen while Tesla is building a custom one and keeps remodeling it every week. That's what will inevitably happen when a company doesn't own their own fabs. They don't have the mask shops. They don't have their own packaging lines. Each of those things require billions of dollars and years of development. So, even if Figure AI wanted to do what Tesla is doing with Terraab, they would need to build the same capability from scratch. The same $2 billion bet. They have to get a relationship with ASML, packaging infrastructure. By the time they get there, Tesla will be hundreds of design iterations ahead. And that head start compounds. That's what a real competitive moat looks like. As long as Tesla is successful, this is the mother of all moats.

I talk very deeply about these types of moats in my new book, Abundance or Collapse, which helps everyone position themselves in the age of AI disruption. So, if you're someone who's trying to figure out where this AI revolution is going and want to position yourself on the winning side of it, I urge you to check out my book. I'll have a link in the description below. Thank you.

Okay, so about 40 minutes into record time and having walked through all of this, the thesis should be much more clear. Tesla's value proposition isn't competing with TSMC or ASML. It's in owning the design-to-packaging loop at a speed no competitor can match, while outsourcing the ASML-intensive work to foundries that already have the machines, at least for the time being. There's nothing stopping Tesla from partnering on a new lithography technology in the future. For example, Steve Jurvetson, one of the earliest investors in Elon's companies, literally just announced before this video a brand new company called Lace Lithography, which is developing a technology that will make transistors that are 100 times smaller than current EUV technology allows. Who knows what Tesla is working on in the background, but it's pretty clear Elon's approach here is the Apple playbook at dramatically faster iteration. Apple doesn't manufacture its own chips either. TSMC does. What Apple does is design the chips and control the packaging architecture. And the result is Apple consistently delivers 30 to 40% better power efficiency than Qualcomm, even though both use TSMC fabs. That's because Apple optimizes every square millimeter of silicon for its own workloads. Design optimization on top of the same foundry gives you a massive edge. And Tesla, with Terraab, is that model, but with the ability to iterate on chip design weekly instead of annually and build everything in-house.

And we haven't even discussed the geopolitical risk that pertains to the entire chip supply chain. A giant percentage of the world's chip capacity is with TSMC, but TSMC is in Taiwan. And what happens when, not if, China makes a move on Taiwan? What if they're successful in taking Taiwan? What does that mean for advanced chip capacity for the Western world? You don't think China is going to try and point as much of that capacity to itself and away from the US? Remember that the US currently doesn't allow China to have any access to TSMC's advanced chip capacity. That completely changes the second China takes Taiwan. And AI is by far the most important technology in the world right now.

And then here's another angle. China has its own freaking Manhattan Project for EUV right now, led by Huawei. They completed a working EUV prototype in early '25. Their prototype generates about 100 watts of EUV power. Production machines need 600 watts or more. That's a six times gap in a metric that took ASML decades to close. And China's top chip executives openly admit their domestic alternatives match ASML's 2008 era technology. So, they're 17 years behind the curve. Realistic timelines for Chinese EUV volume production is probably sometime in 2030 to 2035. And even that assumes they solve dozens of sub-problems that each took ASML years. Now, don't underestimate China. Their work ethic, their willingness to throw massive resources at problems, their manufacturing capability, all genuinely super impressive. They will eventually build an EUV machine. I have no doubt about that. The question is whether they build one fast enough to matter for the current AI race, because the AI race is happening now, or they just simply take Taiwan and then most of the world's chip production capacity with it.

By my estimation, these are the two most ambitious technology programs for Earth right now: China's national semiconductor effort with 3,000 engineers and Elon Musk's $2 billion Terraab. And both trace back to one company in a Dutch city with 45,000 people. One is trying to replicate what ASML does, and the other might have just figured out that he may not need as many machines as everyone assumes. The entire AI revolution, with every robot that will walk into a factory, with every car that will drive itself, every satellite that will process data in orbit and send it back. That whole thing traces back to tin droplets vaporized by lasers, reflected off mirrors smoother than anything in nature, built in a small town in Germany, assembled in a small city in the Netherlands, and there aren't enough of them. Maybe there won't be enough of them ever. But there's a guy in Austin that may have just figured out that if you're smart about design, iteration, and packaging, you might not need as many of them as everyone thinks. It happened with SpaceX. It happened with Tesla. It happened with Neuralink. Why not chips?

Now, look, if you're holding Tesla stock through announcements like this, the kind where every expert is saying that this is the dumbest idea of all time, I get it. The psychology of staying with the name during these moments is genuinely harder than the actual analysis. Bradford Ferguson, a rebellion, has a whole video specifically about why most people could never hold a stock like Tesla. It gets into the mental game of tolerating ambiguity when your thesis is playing out, but the timeline is uncertain. I'll link that video in the description, and these are the guys who called Elon selling his shares a month before that very famous Twitter poll. So, they deeply understand what is going on.

Thank you so much for watching. I hope this video was informative and helpful, and we'll see you on the next one. Take it easy, everybody. Ah, bye-bye.

Oh, by the way, before you go, I have a new book coming out. Another one. Master Plan: What Elon Musk is Actually Building. Pre-orders open on Amazon. It's coming out on 4/20. Totally an accident. Just, you know, just accidental, right? Bye.