Why Tesla Really Partnered With Lemonade / Elon on Digital Optimus / Cybercab Production ⚡️
发布时间 2026-01-22 01:49:36 来源
Welcome to Electrified, it's your host, Dylan Umas, quick shout out to my newest patron, Brian S. Thank you for choosing to support the channel. It's looking more likely by the hour that we were on the right track with Monday's video. Beth J. Zoes shared a post saying, I think XAI having the only one gigawatt data center in the world currently and having others in development means if you think compute scales with models, that XAI will soon be the most valuable company in the world and completely trounce rivals. J. Zoes said the only way to skip past Cloud Code is to do full computer use reinforcement learning agents, which they're apparently doing they're playing to win. I will say XAI may have the only coherent cluster over one gigawatt right now, but the way things are moving in this field that is not going to last for long.
欢迎来到Electrified频道，我是你的主持人，Dylan Umas。先向我最新的赞助者Brian S.致以特别的感谢，感谢你选择支持这个频道。随着时间的推移，星期一视频的观点似乎越来越正确。Beth J. Zoes分享了一条帖子说，她认为目前只有XAI拥有一个吉瓦级数据中心，并且还有其他数据中心在开发中。如果你认为计算能力和模型规模相关，那么XAI很快将成为世界上最有价值的公司，并彻底超过竞争对手。J. Zoes还提到，唯一绕过云代码的方法是使用完整的计算强化学习代理，而他们显然正在这样做，他们的目标是赢得胜利。我想说，XAI可能是目前唯一拥有超过一吉瓦的集群的公司，但鉴于这个领域的发展速度，这种领先不会持续很久。
Basically what this means is XAI is planning to leapfrog these AI agents altogether like Cloud Code and instead build these RL agents that are trained and to end for full computer control. It's exactly what we covered on Monday deploying millions of those human emulators. Elon replied to this saying digital optimists will superset everything except physical optimists. What Elon means is these digital optimists which are XAI's GROC based software agents that apparently will be trained on full computer use reinforcement learning for autonomous digital tasks just like we talked about coding browsing software as a service UI interaction customer service and so on will superset meaning contain or encompass and eventually go beyond basically everything in the current landscape of AI agents and tools except for the physical optimists.
这段话的意思是，XAI计划完全超越现有的AI代理工具，比如Cloud Code，而是直接创建那些已训练好的强化学习（RL）代理，用于全面的计算机控制。这正是周一我们所讨论的内容，即部署数以百万计的人类仿真器。马斯克对此回应称，数字乐观主义者将超越一切，除了实体乐观主义者。他的意思是，这些数字乐观主义者指的是XAI基于GROC的软件代理，它们显然会被训练来通过强化学习实现全计算机使用，用于自动化的数字任务。就像我们讨论的那样，这包括代码编写、浏览、软件即服务、用户界面交互、客户服务等。这些代理将会包含甚至最终超越当前AI代理和工具的几乎所有内容，除了那些“实体乐观主义者”所涉及的领域。
And remember we're not just talking about this as XAI having some crossover synergies with Tesla but there's a very real chance the deployment of these digital optimists are on Tesla vehicles. I saw someone say physical optimists will be for the world of atoms and the digital optimists will be for the world of bits. All I will say for now is if you play these out over the next 10 years things start to look pretty wild for the future. Speaking of wild metaphor said Tesla restarting dojo equals Nvidia is too expensive to which Elon said well we would literally need more dollars than actually exist to put one terawatt per year of AI in space unless we have cheap chips.
请记住，我们不仅仅是在讨论XAI与特斯拉的某些交叉协同效应，还有一个非常现实的可能性，那就是这些数字乐观主义者会被应用到特斯拉车辆上。我看到有人说，物理乐观主义者适用于原子世界，而数字乐观主义者适用于比特世界。我现在只想说，如果在未来十年中实践这些想法，未来看起来会相当疯狂。谈到疯狂，有人用比喻说特斯拉重启Dojo系统是因为英伟达的价格太贵，埃隆·马斯克回应道：其实，要想每年在太空部署一太瓦的AI计算能力，我们所需的资金会超过现有美元总量，除非我们有更便宜的芯片。
Just some rough back of the napkin math for context here let's use a black well B200 from Nvidia. One terawatt is equivalent to one trillion watts. Let's say each B200 is around 1000 watts. You would then take a trillion watts divided by 1000 watts per chip meaning you would need about 1 billion B200 GPUs. If you assume the cost of each is 40 thousand dollars you would then take a billion GPUs times 40 thousand dollars each to get a total cost around 40 trillion dollars. So this is where AI7 comes in toward the end of this decade. The scale Elon is thinking about is far beyond what the average mind will comfortably comprehend and it's precisely that scale paired with design efficiency and perhaps a Tesla tariff that will unlock these cheaper chips to eventually scale toward Elon's ambitions.
这里有一些简单的估算来提供背景信息。我们使用Nvidia的Blackwell B200芯片来做个粗略计算。1太瓦等于1万亿瓦特。假设每个B200芯片大约需要1000瓦特，那么你需要将1万亿瓦特除以每个芯片的1000瓦特，这意味着你大约需要10亿个B200芯片。如果假设每个芯片的成本是4万美元，那么你需要将10亿个芯片乘以每个4万美元，总成本大约是40万亿美元。这也是Elon设想的AI7在本十年代末出现的原因。他考虑的规模远远超出普通人能够轻松理解的范围。正是这种规模，再加上设计效率的提升，以及可能的特斯拉关税，才会使得这些更便宜的芯片得以实现，最终达到Elon的目标。
So XAI will not just be buying mega packs from Tesla but will be buying very large quantities of AI chips from Tesla furthering this flywheel for Tesla, XAI, and SpaceX. And it should go without saying but at scale like this Tesla not having to pay 60 to 70% margins for hundreds of billions if not eventually trillions of dollars worth of chips is going to be quite material and beneficial to Tesla's bottom line. And obviously numbers like this are not going to be hit in the next few years so this is more of a decade plus type of play. However the main point in the takeaway for now should be directionally this is where everything is headed and it's why I'll continually reiterate that the old Tesla that many of us grew to love is really a thing of the past.
所以，XAI 不仅会从特斯拉购买大容量电池包，还将大量采购特斯拉的人工智能芯片，从而推动特斯拉、XAI 和 SpaceX 的增长循环。显而易见的是，特斯拉在规模如此大的情况下，不需要支付60%到70%的利润，磁价值数千亿乃至最终可能达数万亿美元的芯片采购，这对特斯拉的利润有着重大且积极的影响。显然，这样的数字在未来几年内是不可能实现的，因此这是一个需要十年以上时间的长期计划。然而，目前的关键点在于，所有事物的方向正朝这个方向发展，这也是为什么我反复强调，我们曾经喜欢的那个“老特斯拉”已经成为过去。
It will now just serve as the foundation and the bridge to these next chapters. As Sawyer said XAI has now installed around 600 Tesla mega packs worth about 585 million dollars at Colossus 2. That's over 2.3 gigawatt hours of energy making it one of the single largest battery energy storage installations in the world. In case anyone has not figured it out, XAI's macro hard is a play on micro soft which I think is great. And on that wording too many people are focused on macros, proteins, carbs and fats when the real alpha is optimizing for micronutrients.
现在，这将成为接下来几个篇章的基础和桥梁。正如索耶所说，XAI目前已经在Colossus 2安装了大约600个特斯拉MegaPacks，价值约5.85亿美元。这些电池的总储能超过2.3吉瓦时，使其成为全球最大的单一电池储能装置之一。如果有人还没意识到的话，XAI的“宏硬件”其实是在玩“微软”这个词，这个双关很有趣。很多人过于关注宏量营养素，比如蛋白质、碳水化合物和脂肪，而真正的关键其实是优化微量营养素。
Energy optimization and vitality instead of just caloric intake. Both have their place but there's a reason why AG1 has been a part of my life now for years. It's the sponsor of this video but it has played a big role in supporting steady energy levels and supporting my immune health. I don't like to say it out loud but the only sickness I've had in the past few years has been the vid. And as I've been dealing with post parasitic problems, AG1 has helped to maintain my gut balance. AG1's next gen formula has lactobacillus and bifidobacterium which support microbiome diversity digestion, immune function and can help to reduce bloating.
能量优化和活力，而不仅仅是卡路里摄入。两者各有其重要性，但这就是为什么AG1已经成为我生活一部分多年的原因。虽然它是这个视频的赞助商，但它在支持我的能量平稳和免疫健康方面扮演了重要角色。我不太喜欢说出来，但过去几年我唯一感染的疾病就是新冠病毒。而当我处理寄生虫后遗症时，AG1帮助维持了我的肠道平衡。AG1的新一代配方含有乳酸菌和双歧杆菌，这些成分支持微生物群的多样性、消化和免疫功能，还能帮助减少胃胀。
And no matter how great your diet is, you're likely not getting things like ashwagandha and rodeola rusea that support stress resilience, energy and mental stamina. And in this season I need to be dialed in and fueled on a micro level to the max and I know AG1 helps me do that. If you want to support the work I do here you can use my link below to get a free AG1 flavor sampler and AGZ sampler to try all the flavors. Plus a free vitamin D3K2 and AG1 welcome kit with your first AG1 subscription. It's drinkag1.com slash electrified or you can use the QR code on the screen.
无论你的饮食多么优秀，你可能都没有摄入能增强抗压能力、精力和精神耐力的成分，比如印度人参（Ashwagandha）和北极根（Rhodiola Rosea）。在这个季节，我需要在微观层面上达到最佳状态，而我知道AG1能帮助我做到这一点。如果你想支持我的工作，可以使用下方的链接，免费获取AG1口味采样包和AGZ采样包，尝试所有口味。此外，你还可以在首次订购AG1时获得免费维生素D3K2和AG1欢迎套装。访问链接：drinkag1.com/electrified，或者使用屏幕上的二维码。
To tie a bow on Tesla's patents from the other day that really boil down to number manipulation, Elon said the distinction between float and int or integer diminishes as the number of bits decreases. In the limit of one bit there's no difference. Also worth remembering that a float is just two integers, mantisa and exponent with a lot of juggling. Once you get to four bits you have 16 possible outcomes. You can then do whatever you want with those outcomes. Boiling this down if you have a number with decimals that can be broken down into two parts.
总结特斯拉前几天的专利，主要涉及到数字处理，埃隆提到随着位数减少，浮点数（float）和整数（int或integer）之间的区别逐渐消失。在只有一位（1位）时，这两者没有区别。还要记住，浮点数其实是由两个整数组成的，即尾数和指数，需要进行大量运算。当有四位（4位）时，你就有16种可能的结果，可以随意处理这些结果。简单来说，就是如果你有一个可以分为两部分的小数。
The mantisa which is the main numbers and the exponent which determines where the decimal actually goes. And I know if you're not a fan of math all of this is going to feel like pulling teeth and all you really need to know here is what Tesla has figured out is a way to manipulate numbers in the neural nets to ultimately make Tesla's AI work faster and use less power. So in computer land everything is ultimately made up of bits. They're just tiny switches that can be a zero or a one. And adding a bunch of these bits together is how you represent bigger numbers.
尾数是主要的数字，而指数决定了小数点的位置。我知道，如果你不喜欢数学，这些概念可能会让你感觉很头疼。你真正需要了解的是，特斯拉找到了一种方法，可以调整神经网络中的数字，从而让特斯拉的人工智能运行得更快并且耗能更少。在计算机世界中，所有东西最终都是由比特构成的。比特就像微小的开关，可以是0或1。通过组合多个比特，就可以表示更大的数字。
Again, integers are whole numbers like one, two, 20, 100. They do not have decimals. Whereas floating point numbers will have decimals. As Elon said, as you use fewer bits in computer terms making the numbers smaller, the difference between integers and floats starts to disappear. The easiest way I found to understand this with one bit you only have two options. It's zero or one. With two bits then you have four possibilities. Zero zero, zero one, one zero or one one. With three bits you multiply by two again to get eight possibilities and then with four bits that's where Elon said you get 16 different possibilities.
整数是像一、二、20、100这样的完整数字，它们没有小数。而浮点数则包含小数。正如Elon所说，当你在计算机中使用更少的位数，使数字变得更小时，整数和浮点数之间的差异开始消失。最简单的理解方式是：用一位时你只有两个选项——0或1；用两位时有四种可能性：00, 01, 10或11；用三位时，你将可能性再次乘以二，就有八种可能；而用四位时，正如Elon所说，就有16种不同的可能性。
So what Elon is saying when you only have four bits in total, the old rules of how you have to deal with integers and floating point numbers really start to become irrelevant. If you only have 16 options then sticking to those strict rules doesn't make sense anymore. And oversimplifying with Tesla's patents their breakthrough really has just been coming up with their own custom rules for how to play with these numbers. And it's right in line with what we've mentioned in the past that term quantization that's been the silver lining of hope for hardware three owners which is really just taking these numbers and squishing them into fewer bits while keeping all of the important stuff accurate enough.
这段话的意思是，当你只有四个比特时，传统的处理整数和浮点数的规则变得不再重要。如果只能有16种选择，那么坚持那些严格的规则就不太合理了。简化地说，特斯拉的专利突破在于他们自己制定了一套如何处理这些数字的定制规则。这与我们之前提到的“量化”原理相符，对于硬件第三代用户来说，这是一个充满希望的亮点，其实就是在把这些数字压缩到更少的比特中，同时确保所有重要的信息足够准确。
So in short, Tesla can now use fewer bits and still maintain enough accuracy which means the models use less memory and the computers can do faster math leading to less power used. So the contrast has never been more stark. Legacy Auto is laboring over hybrids or gas cars whereas Elon and team are developing custom math. Today, Lemonade announced its autonomous car insurance. A first of its kind product designs specifically for self-driving cars starting with Tesla's FSD. The new offering cuts per mile rates for FSD engaged driving by about 50% reflecting what the data shows to be significantly reduced risk during autonomous operation.
简而言之，特斯拉现在可以使用更少的位数，同时仍然保持足够的精确度，这意味着模型使用的内存更少，计算机可以更快地进行运算，从而减少了能耗。因此，对比显得尤为明显。传统汽车制造商正在努力开发混合动力或燃油汽车，而埃隆·马斯克和他的团队则在研发定制的数学算法。今天，Lemonade宣布推出其自动驾驶汽车保险，这是一种专为自动驾驶汽车设计的产品，初步应用于特斯拉的全自动驾驶（FSD）。这种新的保险产品将FSD启用时的每英里费率降低了约50%，反映出数据所显示的自动驾驶期间风险显著降低。
Lemonade expects further reductions as Tesla releases FSD software updates. The launch is the result of a technical collaboration with Tesla giving Lemonade access to vehicle data that was previously unavailable. Data captured then feeds into Lemonade's usage based risk prediction models already among the most advanced in the industry to uniquely distinguish between autonomous and human driving as well as predict risk based on the autonomous software version installed in the car, the precision of the sensors and more.
Lemonade 预计随着特斯拉发布 FSD 软件更新，成本将进一步降低。这次发布是与特斯拉的一次技术合作的结果，使得 Lemonade 可以获得之前无法获取的车辆数据。这些数据被收集后，将输入 Lemonade 的风险预测模型中。这个基于使用情况的模型已经在行业中处于最领先地位，能够独特地区分自动驾驶和人工驾驶，并基于车上安装的自动驾驶软件版本、传感器的精度等信息来预测风险。
They said our existing pay-per-mile product has given us something no traditional insurer has, a unique tech stack designed to collect massive amounts of real driving data for precise dynamic pricing. Lemonade does support intermittent use of FSD as well as households with a mix of Tesla's and standard non-FSD vehicles from other manufacturers all under one policy. Lemonade has stated its ambition to become the lowest cost best experience insurance in the world. By connecting to the Tesla on-board computer our models are able to ingest incredibly nuanced sensor data that lets us price our insurance with higher precision than ever before.
他们表示，我们现有的按英里计费产品提供了传统保险公司没有的优势，即一个独特的技术框架，能够收集大量真实行驶数据以实现精确的动态定价。Lemonade支持偶尔使用全自动驾驶（FSD）功能，而且在一个保单下可以涵盖同时拥有特斯拉和其他非FSD车型的家庭。Lemonade已经表明，他们的目标是成为全球成本最低、体验最好的保险公司。通过连接特斯拉的车载计算机，我们的模型能够接收到非常细致的传感器数据，从而使保险定价比以往更加精准。
And they said beyond the product announcement today we're also announcing our commitment to the Tesla community. The safer FSD software becomes the more our prices will drop. The product will begin rolling out in Arizona on January 26th and in Oregon later. Tesla owners will be able to get a quote for the new product in seconds through the lemonade app. To be clear, lemonade is still offering its existing car insurance in 10 states as of now, but clearly this autonomous first insurance product is only going to be Arizona in Oregon at least to start.
他们表示，除了今天的产品发布之外，我们还宣布了对特斯拉社区的承诺。随着FSD（全自动驾驶）软件变得更加安全，我们的价格将越降越低。该产品将于1月26日在亚利桑那州开始推出，然后在俄勒冈州推出。特斯拉车主可以通过Lemonade应用在几秒钟内获得新产品的报价。需要说明的是，Lemonade目前仍在10个州提供现有的汽车保险，但显然，这款以自动驾驶为首的保险产品至少一开始只会在亚利桑那州和俄勒冈州推出。
Shy Winninger, the president of Lemonade said Tesla's FSD is two-time safer than humans so why are Tesla owners still charged so much to insure their cars? This first version brings native FSD support allowing owners to instantly connect their Tesla to the lemonade app. Once connected, we base our prices on actual usage rather than countrywide averages, making car insurance cheaper, better and more fair. He said the Tesla team was amazing to work with and Lemonade's autonomous car insurance is based on the AI and telematics infrastructure we've been developing for years.
Lemonade总裁Shy Winninger表示，特斯拉的全自动驾驶（FSD）比人类驾驶安全两倍，那么为什么特斯拉车主仍要支付那么高的保费呢？这个首个版本引入了本地FSD支持，让车主可以立即将特斯拉连接到Lemonade应用程序。一旦连接成功，我们会根据实际使用情况而不是全国平均水平来确定价格，使车险变得更加便宜、更好、更公平。他还说，与特斯拉团队合作非常愉快，Lemonade的自动驾驶汽车保险基于我们多年来开发的人工智能和远程信息处理基础设施。
It enables us to track adjust in price policies based on actual usage and driving patterns. Since FSD 14.2 is at least two-time safer than human drivers, we adjust to it and price miles driven with FSD at 50% of human driven ones. And it's only the beginning. I'll pause here and say yes, Lemonade is a smaller company that will be needing more exposure and piggybacking off of Tesla's success and viability is a smart thing to do. So, is this more of a marketing ploy or is this what the actuarial assessment really dictates? We're going to have to let this play out to really have that answer.
这让我们可以根据实际使用情况和驾驶模式来调整价格政策。由于FSD 14.2的安全性至少是人类司机的两倍，我们据此调整价格，将FSD行驶的里程价格设定为人类驾驶的一半。这只是一个开始。我在这里稍作停顿，必须承认，是的，Lemonade是一家较小的公司，需要更多的曝光，借助特斯拉的成功和可行性是个明智的做法。那么，这到底是一个市场营销策略，还是精算评估真正的结果呢？我们需要让这个过程自然发展才能真正找到答案。
But clearly this is a partnership where Lemonade has access to unique custom data that we do not have. And they're saying the official data shows FSD 14.2 is at least two-time safer than human drivers. So, this is a tangible data point that none of the community trackers have been able to give us. Shy said through our direct Tesla integration, our system captures detailed telemetry on trips and car usage. We collect the software version, hardware version and sensor pack installed, all updated in real time.
显然，这是一个合作伙伴关系，Lemonade能够访问我们没有的独特定制数据。他们表示，官方数据显示FSD 14.2的安全性至少是人类驾驶员的两倍。对于这个具体的数据点，目前社区中的追踪者都无法提供给我们。Shy说，通过我们与特斯拉的直接集成，我们的系统能够详细捕获有关行程和汽车使用情况的遥测数据。我们会收集软件版本、硬件版本和传感器包的信息，并且实时更新。
This allows us to adjust prices automatically for every mile driven and quickly adjust prices as new safer versions are released. We're headed toward a world where your car and insurance update overnight while you sleep becoming safer and cheaper without any effort on your part. As we evolve the product, an FSD becomes fully unsupervised, driver age and driving history will no longer matter as pricing factors. A young driver will pay the same low FSD rate as their mom and dad.
这使我们能够根据每行驶一英里自动调整价格，并在更安全的新版本发布时迅速更新价格。我们正走向一个这样的世界：您的汽车和保险在您睡觉时自动更新，变得更安全、更便宜，而您无需做出任何努力。随着我们产品的发展，当全自动驾驶（FSD）成为完全无人监管时，司机的年龄和驾驶历史将不再影响定价。年轻司机将享受与他们的父母相同的低FSD价格。
Naturally, you have some people pushing back on this saying that they don't want their cars driving behavior to be tracked and that's fine, no one's forcing you into a plan like this. But over the long term, if you want your insurance rates to trend toward zero, you're likely going to have to opt in to a plan just like this. I'm guessing the biggest question right now will be, well, what about Tesla insurance? Elon did chime in on this saying, insurance is half price when Tesla self-driving is activated because it increases safety so much.
自然，有些人对此表示反对，他们不希望自己的汽车驾驶行为被监控，这很正常，没人会强迫你加入这样的计划。然而，从长远来看，如果你希望保险费率趋近于零，你很可能需要选择加入类似的计划。我猜现在最大的疑问是，特斯拉保险又如何呢？埃隆对此的看法是，当特斯拉自动驾驶功能激活时，保险费用会减半，因为这种功能大大提高了安全性。
At the end of the day, insurance really boils down to actuarial risk. And every different company has its own algorithm and its own model, meaning ultimately it's willing to take on different levels of risk. A company like Lemonade that does one thing offers insurance, that's a smaller company needs to do whatever it can to earn market share to stay alive. Meaning Lemonade is likely to be more willing to offer very low rates to earn that market share and to get some marketing buzz.
归根结底，保险其实就是关于精算风险。每个公司都有自己的算法和模型，这意味着它们在愿意承担的风险水平上会有所不同。像Lemonade这样专注于提供保险的小公司，需要尽一切努力来获得市场份额以维持生存。这意味着Lemonade可能更愿意通过提供非常低的费率来赢得市场份额，同时也能获得一些市场关注。
Tesla, on the other hand, does not have the same incentive to earn new insurance market share. Tesla insurance really is an afterthought when thought of in the larger picture of Tesla's business model. So Tesla insurance is clearly not a primary business for Tesla. And further, the more third-party companies that Tesla can partner with and give this data to, the more Tesla customers win and ultimately the faster FSD adoption spreads. And right now, Tesla's insurance is only available in less than 30% of states and Lemonade's autonomous products will be available in two to start. So they can both coexist for quite a while. But these first two states for Lemonade will be a great test because Tesla's insurance is available in both Arizona and Oregon.
特斯拉在保险市场上并没有太大的动力去争取新的市场份额。在特斯拉的整体商业模式中，保险业务实际上只是一个次要考虑。因此，特斯拉保险显然不是特斯拉的主要业务。此外，特斯拉与更多的第三方公司合作并分享数据，有利于特斯拉的客户，也有助于加速全自动驾驶（FSD）的普及。目前，特斯拉的保险业务仅在不到30%的州提供，而Lemonade的自动驾驶产品将率先在两个州推出。因此，特斯拉保险和Lemonade的产品可以共存相当长一段时间。不过，对于Lemonade来说，这两个首发州将是很好的试验机会，因为特斯拉的保险在这两个州（亚利桑那州和俄勒冈州）都已经提供。
So if most of Tesla's insurance customers switch to Lemonade and things go well and Lemonade continues to scale, there is a world where Tesla would no longer have to deal with the regulatory headache of insurance if we finally have a company willing to offer proper risk-adjusted rates for Tesla owners. Tesla was never doing this to make money. It was doing it to help customers. But if someone else can offer it and work closely with Tesla and offer great customer service, which I hear Lemonade does, Tesla could eventually wind down its insurance offerings and focus on the core business.
因此，如果大部分特斯拉的保险客户转向选择Lemonade，并且一切进展顺利，Lemonade继续扩大业务规模，在这种情况下，如果我们最终有公司愿意为特斯拉车主提供合适的风险调整费率，特斯拉将不再需要应对保险方面的监管麻烦。特斯拉从来不是为了赚钱而提供保险，而是为了帮助客户。但如果有其他公司能够提供这一服务，并与特斯拉紧密合作，同时提供出色的客户服务，而我听说Lemonade正是如此，那么特斯拉最终可以缩减其保险业务，并专注于核心业务。
I do, of course, think this will take a few years to play out and things will likely change again when unsupervised launches as Tesla will then be taking liability and that will certainly be a fundamental shift in risk mitigation. Right now, the average US driver pays $2,500 in insurance annually. Simple math with Lemonade, that number could drop to around $1,250 if they use FSD all the time. That would be a savings of $104 a month, which is basically exactly the amount of Tesla's FSD with tax. So people will be further incentivized to buy a Tesla, cut insurance rates nearly in half and then use that savings to effectively get FSD for free.
当然，我认为这需要几年时间才能完全实现，并且当无人监管的自动驾驶推出时，情况可能会再次发生变化，因为那时特斯拉将承担责任，这肯定会带来风险应对的根本性转变。目前，美国普通司机每年支付2,500美元的保险费。简单计算，如果使用Lemonade，他们在全程使用全自动驾驶（FSD）的情况下，这个数字可能会降到约1,250美元。这样每个月可以节省104美元，而这刚好是特斯拉FSD的费用（含税）。因此，人们将更有动力去购买特斯拉，几乎将保险费用减半，然后用这些节省的费用来实现“免费”使用FSD。
To clarify, this new product from Lemonade does not change anything when it comes to the plan's personal liability and property damage. So this 50% reduction is only for the miles traveled portion, meaning to start in the real world, the actual overall savings are likely to be lower than 50%. However, as we discussed, over time, the rates are likely to come down as FSD improves. So if someone uses FSD all the time, then over time they are likely to save in the neighborhood of 50% overall. Now, Lemonade certainly has a lot to prove here, but the fact that a public company is willing to tie the success or failure of its business directly to the safety of Tesla's FSD is a huge sign that they're quite confident in the safety data leading to reduced claims.
为了澄清，这款由Lemonade推出的新产品在个人责任和财产损失计划上并没有任何变化。因此，这个50%的折扣仅适用于行驶里程的部分，这意味着在实际情况中，总体的节省可能会低于50%。然而，正如我们讨论的，随着全自动驾驶（FSD）的改进，费率可能会逐步下降。因此，如果有人始终使用FSD，那么最终他们可能会实现接近50%的整体节省。目前，Lemonade确实还有很多要证明的地方，但一家上市公司愿意直接将其业务的成败与特斯拉FSD的安全性挂钩，这表明他们对基于安全数据的减少索赔充满信心。
And further, already committing to continually lowering prices as Tesla's FSD safety increases, which in my opinion is exactly how it should be. The lower the likelihood your car is to get in an accident, the lower your insurance premium should be, and to date, for most people, this was just not an option. Tesla started this trend on its own, but now it's getting validation from a third party, so hopefully Lemonade can scale quickly, and I will say it'll be fascinating to watch how the rest of the industry responds as customers start calling to cancel their policies.
进一步而言，随着特斯拉的FSD（全自动驾驶）安全性不断提高，公司已经承诺持续降低价格，我认为这正是应有的做法。你的汽车发生事故的可能性越低，你的保险费就应该越低，而对于大多数人来说，这在以前是无法实现的。特斯拉起初自主推动了这一趋势，但现在它得到了第三方的验证，因此希望Lemonade能够快速扩展。我想说，看到当客户开始打电话取消他们的保险时，整个行业会如何反应将会非常有趣。
And don't forget about all of the headaches that Tesla insurance will and has caused for Tesla. Last year, Tesla was hit with an enforcement action by California's Department of Insurance. Tesla was accused of egregious delays in responding to policyholder claims, unreasonable denials, and engaging in unfair claims settlement practices. So again, it's a lot of headaches, it's a lot of regulatory red tape for Tesla that really doesn't need to be in this business. So I'm not ready to make this prediction, but I will say I will not be surprised if in five years Tesla insurance is no longer a thing.
别忘了，特斯拉保险给特斯拉带来了许多麻烦。去年，加州保险局对特斯拉采取了执法行动。特斯拉被指控在回应保单持有人理赔时严重拖延、不合理拒赔，以及参与不公平的理赔结算行为。因此，这对特斯拉来说是许多麻烦和繁杂的监管手续，特斯拉其实不需要涉足这个业务。我虽然还不准备做出这样的预测，但如果五年后特斯拉保险不复存在，我也不会感到惊讶。
But that lack of surprise would be dependent on a company like Lemonade continuing to grow and scale across the country and doing so successfully. Because if Lemonade can do that, then why would Tesla put up with all of the challenges in what's traditionally a very low margin business, and the better autonomy gets the lower those margins are likely to go. Cena Finance reported that Samsung's factory in Taylor, Texas is planning to begin trial operations of its EUV equipment in March this year, aiming to prepare for mass production of AI5 and AI6 in the second half of this year.
如果像Lemonade这样的公司能够在全国范围内持续增长和扩大规模，并且成功做到这一点，那就不足为奇了。因为如果Lemonade能做到这一点，那么为什么特斯拉还要忍受在传统上利润率非常低的业务中的各种挑战呢？而且随着自动化水平的提高，这些利润率可能还会进一步降低。Cena Finance报道，三星位于德克萨斯州泰勒市的工厂计划于今年3月开始进行EUV设备的试运营，目标是在今年下半年为AI5和AI6的量产做好准备。
The plant is planned to transition from a four nanometer process to a monthly production capacity of 50,000 to nanometer wafers. Although Samsung's two nanometer yield has improved to about 50%, it still lags behind TSMC, and its technological maturity will directly impact the supply schedule to customers like Tesla. Phase two of the factory may begin production ahead of schedule due to increased orders. Samsung plans to install etching and deposition equipment in phases with full production expected in the second half of this year. Samsung is currently preparing to apply for a temporary occupancy permit from the relevant authorities in Texas to begin production before the factory is officially done. And to remember, deposition is just adding material to the wafer, and then etching on the other hand is removing or carving out material from the wafer.
该工厂计划将工艺从四纳米过渡到月产能5万片二纳米晶圆。尽管三星的二纳米工艺良率已经提高到约50%，但仍然落后于台积电，其技术成熟度将直接影响向特斯拉等客户的供应进度。由于订单增加，工厂的第二阶段可能会提前开始生产。三星计划分阶段安装蚀刻和沉积设备，预计在今年下半年实现全面生产。三星目前正在准备向德州相关部门申请临时占用许可，以便在工厂正式完工前开始生产。需要注意的是，沉积是向晶圆添加材料，而蚀刻则是从晶圆上移除或雕刻材料。
Construction at the Taylor Fab is nearing completion. There's about 7,000 workers currently on site every day, with about 1,000 already working in the factory's office building. Overall progress has accelerated significantly compared to a year ago. Samsung has reserved land at the site to accommodate 10 more wafer fabs, demonstrating its long-term expansion strategy in the US. And if Samsung can secure more customer orders, the Taylor 2 factory currently in early stages of construction may begin operations ahead of schedule. And the timelines check out; we know Tesla's AI5 is nearing the completion of the design phase.
泰勒工厂的建设即将完成。目前每天约有7,000名工人在现场工作，其中约1,000人在工厂的办公大楼内。与去年相比，总体进展显著加快。三星在该地点预留了土地，以容纳另外10家晶圆厂，这展示了其在美国的长期扩展战略。如果三星能够获得更多客户订单，现处于早期建设阶段的泰勒2号工厂可能会提前投入运营。时间安排也都符合；我们知道特斯拉的AI5设计阶段即将完成。
And from there, it could be 6-9 months before it enters trial production. And given the report said Samsung is aiming to prepare for mass production, that's very likely to mean starting trial production of the AI5 chip in the second half of this year. Then roughly one year of getting the yield rates improved, and we should see AI5 out in the wild by the end of 2027. Jotetmeyer shared a video saying the Cortex 2, Chiller systems in Austin are making quick progress. Part 1 has 4 of the 6 fan enclosures, pipe systems, and some exterior wall segments being installed. The overall height of this chiller is now significantly higher than the Cortex facility itself. Part 2 has almost 3 of the 4 walls poured with concrete.
从那时起，可能需要6到9个月才能进入试生产阶段。根据报道显示，三星正努力准备量产，这很可能意味着将在今年下半年开始AI5芯片的试生产。然后，大约一年时间来提高良品率，我们预计到2027年底AI5芯片就会上市。Jotetmeyer分享了一段视频，显示位于奥斯汀的Cortex 2冷却系统进展迅速。第一部分有6个风扇外壳中的4个、管道系统和一些外墙部分正在安装中。这个冷却系统的整体高度现在明显高于Cortex设施本身。第二部分有近3面墙已经用混凝土浇筑完成。
And overall, it looks like parts 1 and 2 will be mirror images of each other with a total of 12 fan units, which would be double the current cooling capabilities of Cortex 1. So the brain of Optimus being constructed in real time, this is set to be a 500 megawatt site, ready to go online sometime this summer. Tesla cleared up some fun today saying they have no plans to curb production or cut staff at Gigaburlin. This denying a media report that the workforce there had shrunk significantly over the past few years. In an email, Tesla said compared to 2024, there's been no significant reduction in the number of permanent staff, nor are there any such plans.
整体来看，部分1和部分2将是相互镜像的，总共有12个风扇单元，这将是Cortex 1目前冷却能力的两倍。因此，Optimus的大脑正在实时构建中，这将是一个500兆瓦的场地，预计将在今年夏天上线。特斯拉今天澄清了一些有趣的谣言，表示他们没有计划在Gigaburlin限制生产或裁员。这是否认了一则关于过去几年Gigaburlin员工数量大幅缩减的媒体报道。特斯拉在一封邮件中表示，与2024年相比，没有明显减少固定员工的数量，也没有这样的计划。
The message from factory management remains, the situation and outlook at the Gigabactory are stable, especially with regard to employee jobs. Tesla did say it was completely normal to see some fluctuation at the nearly 4-year-old factory, adding the need for temporary workers had decreased since the initial ramp of production. I am so glad Elon shared this post responding to Sawyer talking about CyberCab production. Elon said with the important caveat that initial production is always very slow and follows in S-curve. The speed of the production ramp is inversely proportionate to how many new parts and steps there are.
工厂管理层的信息显示，超级工厂的情况和前景是稳定的，特别是在员工工作方面。 特斯拉表示，在这家运营了近四年的工厂中，看到一些波动是完全正常的，并补充说，自从最初的生产加速以来，临时工的需求已经减少。我很高兴Elon回应了Sawyer关于CyberCab生产的讨论，并分享了这条消息。Elon提到一个重要的前提，即初始生产总是非常缓慢的，并且遵循S形曲线。生产加速的速度与所涉及的新零件和步骤的数量成反比。
For CyberCab and Optimus, almost everything is new, so the early production rate will be agonizingly slow, but everything ends up being insanely fast. To date, it feels like it's mostly just been me trying to temper production expectations with CyberCab and Optimus, but there's a reason why this year I'm only expecting a few thousand CyberCab and maybe a few hundred possibly a few thousand Optimus produced. This is exactly why, and it's right in line with everything at Tesla right now. It's slowly at first, sometimes painfully slow, then eventually at some point it will be all at once.
对于CyberCab和Optimus来说，几乎所有东西都是全新的，因此早期的生产速度会非常缓慢，但最终会变得非常快。到目前为止，我感觉主要是我在努力降低对CyberCab和Optimus生产的期望。这也是为什么今年我只预计生产几千辆CyberCab，或许会生产几百到几千个Optimus。这正是原因所在，也与特斯拉目前的一切相符。开始时很慢，有时候甚至让人感到痛苦地慢，但最终会在某一时刻突然加快进度。
So just prepare for the possible outcome that even if the 2500 vehicle cap for those FMVSS regulations is not lifted this year, that Tesla doesn't even run into that limit this year with the CyberCab. I do actually think it's a reasonable outcome that Tesla ends up producing right around 2500 CyberCab for this whole year, which certainly won't win me any fans, but it really is what I'm expecting. ARK Invest put out an article about Tesla's FSD and changing the pricing model, and they said ARK does not include FSD subscriptions in our public Tesla valuation model, as we believe this business line will be dwarfed by its fleet-based Robotaxi service, which could generate thousands of dollars in cash flow per car per year.
所以，做好准备，即便今年那些FMVSS法规的2500辆限制没有取消，特斯拉今年也可能在CyberCab上不会碰到这个限制。实际上，我认为特斯拉全年生产大约2500辆CyberCab是个合理的结果，尽管这说法可能不会被很多人接受，但这确实是我的预期。ARK Invest发布了一篇关于特斯拉FSD（全自动驾驶）的文章，并提出了定价模型的改变。他们表示ARK没有将FSD订阅服务纳入其公开的特斯拉估值模型中，因为他们认为，相较于车队基础的自动出租车服务，这项业务的收入会显得微不足道，而自动出租车服务每辆车每年可能会带来数千美元的现金流。
According to ARK, Tesla's Robotaxi business could account for around 90% of the company's enterprise value in 2029, and don't forget what Elon said last year, 80% of Tesla's value will be optimists. Which ARK Invest really hasn't even started factoring into its Tesla model. And of course, Elon's comment is looking more 10 plus years into the future, but I have a feeling when Tesla starts to ramp optimists materially. I'm talking somewhere around 5 to 10,000 per month, I think Wall Street will start to price in a lot of those future cash flows much faster than one would expect. Because at that point, the business line will be validated Wall Street can create their precious models, and it's really that confirmation that we're still waiting for when it comes to unsupervised, and that scaling to at least a few cities.
根据ARK的说法，到2029年，特斯拉的自动驾驶出租车业务可能占到公司企业价值的约90%。别忘了去年埃隆提到过，特斯拉80%的价值将来自乐观主义者。而ARK Invest实际上还没有在其特斯拉模型中真正加入这一因素。当然，埃隆的评论着眼于未来超过10年的发展，但我有一种感觉，一旦特斯拉的乐观主义者项目开始显著扩展，大约每月生产5000至10000台，我认为华尔街将比预期更快地开始将这些未来现金流计入其中。因为在那时，这条业务线将得到验证，华尔街可以创建他们珍视的模型，而我们仍在等待的是无人监管的验证以及在至少几个城市的扩展。
I'm quite confident right now Wall Street is waiting for that confirmation. Tesla's say questions are live for Q4 for the call that will be held next Wednesday the 28th. Number one question will long-term Tesla shareholders be prioritized if SpaceX IPOs? Elon will say he wants to, but he won't share any detail. Question two, what's the current bottleneck to increase Robotaxi deployment and personal use on supervised FSD? Which I think is a great question. Question three with Robotaxi live, can you share early unit economics, approximate revenue and cost per mile, whether rides are contribution margin positive, and the main drivers in timeline to reach high margin scaled cash flows as the fleet grows and supervision drops.
我现在非常有信心，华尔街正在等待确认。特斯拉的问题于第四季度现场开放，电话会议将在下周三（28日）举行。第一个问题是，如果SpaceX上市，特斯拉的长期股东是否会被优先考虑？埃隆会说他愿意优先考虑，但他不会透露任何细节。第二个问题是，目前增加无人出租车的部署和个人使用的最大瓶颈是什么？我认为这是一个很好的问题。第三个问题是，在无人出租车上线后，您能否分享初期的单位经济情况、每英里的大致收入和成本、车费是否在贡献边际上是正数，以及随着车队增长和监督减少，实现高利润规模化现金流的主要驱动因素和时间表。
Regarding Optimus, could you share the current number of units deployed in Tesla factories and actively performing production tasks? What specific roles or operations are they handling, and how has their integration impacted factory efficiency or output? And one more when is FSD going to be 100% unsupervised? Elon said AI is a supersonic tsunami, so I think it's fair to say he's seeing something out there right now. On these data centers in space that Tesla will likely play a big role in, John Carmack said, I wonder how the thinking on Kessler syndrome has evolved where cascading collisions with debris makes orbit unusable. Kessler syndrome is just a scenario where orbital debris in Leo becomes high enough that collisions between them ultimately creates more debris, which then causes more collisions in this self-sustaining cycle.
关于Optimus，能否分享一下目前部署在特斯拉工厂中正在执行生产任务的单位数量？他们具体承担了哪些角色或操作？他们的整合对工厂效率或产量产生了怎样的影响？另外，FSD（全自动驾驶）什么时候能够实现100%无人监督？埃隆曾说过AI是“超音速海啸”，所以我想他现在应该看到了些什么。在特斯拉可能会扮演重要角色的太空数据中心上，约翰·卡马克提到，我想知道对“凯斯勒综合症”（Kessler Syndrome，指低地轨道上的碎片足够多，导致它们之间的碰撞生成更多碎片，进而引发更多碰撞的循环）的看法有何变化。这种综合症会使轨道无法继续使用。
Another way to think of it, space waste. Elon said it's certainly an argument for staying low enough at first to allow atmospheric drag to clean things up. At least for the first several years, AI hardware will advance fast enough that an orbital lifetime of around five years is fine. Once you get far enough into space, it's just so absurdly vast that Kessler syndrome cannot happen due to extreme sparsity. Hunter said two years ago he invested in XAI at a $20 billion valuation, and now his investment has nearly 10x. Last week, he invested in SpaceX at an $800 billion valuation, saying 100x seems possible.
可以换个角度来理解，就是太空垃圾的问题。埃隆提到，确实有理由一开始保持足够低的轨道，这样大气阻力可以帮忙清理垃圾。至少在最初的几年里，AI硬件发展的速度快到可以接受五年左右的轨道寿命。一旦足够深入太空，空间广阔到凯斯勒综合症（指太空垃圾引发的连锁碰撞）几乎不可能发生，因为物体的密度极低。亨特在两年前以200亿美元的估值投资了XAI，现在他的投资几乎增长了十倍。上周，他以8000亿美元的估值投资了SpaceX，并表示增长100倍也是有可能的。
To which Elon said if we harness even a billionth of the sun's power for AI and robotics, it'll be over a 1000x return. Money won't mean much at that point. So if Elon is referring to the SpaceX part, which I think he is, and over 1000x return from SpaceX's market cap today would take it to an $800 trillion company. I honestly feel weird even sharing this. Farzad said Tesla, XAI, and SpaceX is the ultimate industrial scale synergistic cash flow generating flywheel behemoth, aka Elon can't lose. To which Elon said complacency is a sure path to defeat.
埃隆表示，如果我们能利用到哪怕是太阳能的十亿分之一用于人工智能和机器人技术，那么回报将会超过1000倍。那时候，金钱可能已经不再那么重要。所以，如果埃隆是在指SpaceX这一部分，我觉得他是，那么从SpaceX目前的市值来看，超过1000倍的回报将使其成为一个80万亿美元的公司。说实话，我分享这个感觉有点奇怪。Farzad说，特斯拉、XAI和SpaceX是一个终极的工业规模协同产生现金流的巨型飞轮，也就是说，埃隆不可能失败。而埃隆回应说，自满是失败的必经之路。
Peter Diamanda said the marginal cost of intelligence is trending to zero, to which Elon said it will drop to a low number, but never zero. The cost will be measured in jewels, which is why more and more people are saying it's all about the energy. Tesla has officially changed its mission to building a world of amazing abundance. Elon shared that saying universal high income. And one day, we will talk about this a bit more in depth, but I would love to hear how you guys are thinking about this right now.
Peter Diamandis说，智能的边际成本正在趋向于零，而Elon Musk认为它会降到一个很低的数字，但不会是零。成本将用“珠宝”来衡量，这就是为什么越来越多的人说一切都与能源有关。特斯拉已经正式将其使命改为打造一个充满奇迹的富裕世界。Elon还提到要实现普遍的高收入。有一天，我们会更加深入地讨论这个话题，但我很想知道你们现在对此是怎么想的。
Tesla's thought closed the day at $431.44 up to 0.91% while the NDX was up 1.36%. The volume was 3% below the average. Don't forget, check out AG1 linked below, grab those 3Vs for those of you interested, and thank you in advance if you do. Hope you have a wonderful day, and a huge thank you to all of my Patreon supporters.
特斯拉的股价收盘时为431.44美元，上涨了0.91%，而纳斯达克100指数上涨了1.36%。成交量比平均水平低了3%。别忘了查看下面链接的AG1，对于那些感兴趣的人，可以领取3V。在此先感谢大家，希望你们度过美好的一天，特别感谢我所有的Patreon支持者。