It's Time: The State of the (Tesla) Union ⚡️
发布时间 2026-03-15 00:22:10 来源
以下是内容的中文翻译：
演讲者承认特斯拉正在经历前所未有的、根本性的变革，这导致一些长期追随者感到困惑和沮丧，因为盈利能力停滞不前、项目延期以及被认为是分散注意力的举动。然而，他旨在阐明这些变化，特别是公司向代理式AI（Agentic AI）的转变，并解释为何他认为这些是特斯拉长期发展的正确决策。
以下是主要新闻和更新的详细分解：
**特斯拉的战略转变和新项目：**
1. **TeraFab项目：** 埃隆·马斯克在推特上证实，TeraFab项目将于七天后启动。演讲者认为这是特斯拉最大的项目，估计资本支出（CAPEX）为100-300亿美元，旨在2.5-4年内生产芯片，以避免未来的瓶颈。尽管其成功并非板上钉钉，但特斯拉很可能会与三星和台积电等公司合作。
2. **代理式AI（数字擎天柱）与XAI：**
* 特斯拉和XAI正在探索一条“代理式路径”来开发数字擎天柱（Digital Optimus）。
* **XAI重组：** XAI正在进行大规模重组。埃隆·马斯克正在重组公司，承认自己最初“招聘失误”，现在正从零开始重建XAI，并联系那些之前被拒绝但很有前途的候选人。
* **Ashish Lswamy的角色：** 特斯拉人工智能软件负责人Ashish (Lswamy) 正在临时监督MacroHard（XAI）的工作，使其步入正轨并与特斯拉的工具和团队整合。
* **进展声明：** 埃隆声称XAI将在今年迎头赶上，并在三年内远远超越竞争对手。
* **部署：** 代理式AI将适用于所有AI4汽车，在车辆非驾驶时支持“办公工作”。
* **专用单元：** 特斯拉正在超级充电站部署数百万个“专用数字擎天柱单元”，利用7千兆瓦的可用电力，并由超级充电团队协助供电。
* **时间表：** 埃隆表示这项功能可能在六个月内推出，数字擎天柱比实体擎天柱的市场化路径更快。
* **AI4效率：** 阐明了AI4的成本（每个片上系统SOC 650美元，两个可能1300美元），突出了特斯拉在推理计算方面的强大地位。
* **成本优势：** 特斯拉和XAI的模式涉及在AI4芯片上本地完成大部分计算，与基于云的代理式AI（如Perplexity，用户使用成本可能很高）相比，显著降低了成本。
* **独特方法：** 数字擎天柱的设计理念是像FSD（全自动驾驶）一样“观察屏幕”（像素输入，控制输出），而不是依赖自定义API或工具，理论上可以与任何软件配合使用。它不仅能处理静态图像，还能处理屏幕视频，以实现实时适应性。
* **风险：** 承认的风险包括提示注入攻击。
* **愿景：** 长期愿景是利用特斯拉App和Grok提示特斯拉车辆来供电和执行各种任务。
* **招聘信息：** 特斯拉在帕洛阿尔托发布了一个招聘职位，为Robotaxi（自动驾驶出租车）客户运营招聘高级项目经理。
3. **CyberCab和Semi生产：** 两者都在为生产做准备。
* **CyberCab FMVSS认证：** 奥斯汀快闪店的照片显示，CyberCab贴纸上写着它“符合所有适用的美国联邦机动车辆安全标准（FMVSS）”，这意味着特斯拉的自我认证表明它不需要555部分的豁免，也不会受到2,500辆的产量限制。
* **无人监督的自动驾驶出租车：** 尽管奥斯汀的法规不限制无人监督的自动驾驶出租车，但特斯拉目前似乎只运营一辆（演讲者质疑Whole Mars声称有22%的车队是无人监督的说法）。
* **用户界面/硬件细节：** CyberCab的用户界面目前显示“踏板和方向盘”选项，并且车载软件是标准的特斯拉软件（例如，尽管没有前备箱，但仍有“打开前备箱”的选项），这表明它尚未专门为CyberCab定制。照片中显示的小型方向盘并非生产就绪版本。
4. **擎天柱（Optimus）生产：** 埃隆在接受彼得·戴曼迪斯采访时表示，Optimus 3将于今年夏天开始生产，这可能是一个提前的计划。
**其他值得关注的新闻：**
* **特斯拉-SpaceX投资：** 特斯拉此前对XAI的20亿美元投资被转换为SpaceX（或一个新的SpaceX AI实体）不到1%的股份。这被定性为一项行政行为，并非特斯拉与SpaceX即将合并的迹象。
* **英国电力供应商：** 特斯拉获得在英国向家庭供电的许可，正式成为英国的获准电力供应商。
* **标普500指数规则：** 标普500指数正在考虑修改其纳入规则，这可能会加速SpaceX在IPO后进入该指数。
* **本田电动汽车重组：** 本田作为上市公司正面临近70年来的首次年度亏损，这归因于与其电动汽车业务相关的157亿美元重组费用。
* **赞助商：** 该视频由Surfshark VPN赞助，重点介绍了其新功能，如搜索结果中的安全徽章、替代身份、电子邮件诈骗检测、后量子加密以及一项零关键漏洞的独立安全审计。
演讲者最后重申了他对特斯拉的长期看涨观点，尽管短期内存在挫折和风险/不确定性增加。他认为，特斯拉对这些巨大潜在市场的追求，加上其文化、效率和集成方法，使其在未来10年内拥有巨大的上行空间，并设想在未来12个月内实现CyberCab的规模化、擎天柱的生产、Semi的生产、能源业务的增长以及车辆中的代理式AI。
The speaker acknowledges that Tesla is undergoing unprecedented, fundamental changes, leading to confusion and frustration among some long-time followers due to stalled profitability, missed timelines, and perceived distractions. However, he aims to clarify these changes, particularly the company's shift towards Agentic AI, and explain why he believes these are the right decisions for Tesla's long-term future.
Here's a breakdown of the key news and updates:
**Tesla's Strategic Shifts & New Ventures:**
1. **TeraFab Project:** Elon Musk's tweet confirmed the TeraFab project launches in seven days. The speaker believes this is Tesla's biggest undertaking, with an estimated CAPEX of $10-30 billion, aiming to produce chips in 2.5-4 years to avoid future bottlenecks. Its success is not guaranteed, but Tesla will likely collaborate with companies like Samsung and TSMC.
2. **Agentic AI (Digital Optimus) & XAI:**
* Tesla and XAI are pursuing an "agentic path" to develop Digital Optimus.
* **XAI Reorganization:** There's a massive shake-up at XAI. Elon Musk is reorganizing the company, admitting he "hired wrong" initially and is now rebuilding XAI from the foundations up, reaching out to previously declined promising candidates.
* **Ashish Lswamy's Role:** Ashish (Lswamy), head of AI software at Tesla, is temporarily overseeing the MacroHard (XAI) effort to get it on track and integrate with Tesla's tools and teams.
* **Progress Claims:** Elon claims XAI will catch up this year and exceed competitors by a long distance in three years.
* **Deployment:** Agentic AI will work on all AI4 cars, enabling "office work" when the car is not driving.
* **Dedicated Units:** Tesla is deploying millions of "dedicated digital optimy units" at Supercharger locations, leveraging 7 gigawatts of available power, with Supercharging teams helping to energize them.
* **Timeline:** Elon suggests this feature could be available in six months, with a faster path to market for digital optimists than physical ones.
* **AI4 Efficiency:** Clarification on AI4 cost ($650 per SOC, potentially $1300 for two), highlighting Tesla's strong inference compute position.
* **Cost Advantage:** Tesla and XAI's model involves doing most computing locally on the AI4 chip, significantly reducing costs compared to cloud-based agentic AIs (like Perplexity, which can be expensive for users).
* **Unique Approach:** Digital Optimus is designed to "watch the screen" (pixels in, controls out) like FSD, rather than relying on custom APIs or tools, theoretically working with any software. It also works with video of the screen, not just still images, for real-time adaptability.
* **Risks:** Acknowledged risks include prompt injection hacks.
* **Vision:** The long-term vision is to use the Tesla app and Grok to prompt Tesla vehicles to power and perform various tasks.
* **Job Listing:** Tesla has a job listing in Palo Alto for a Senior Program Manager for Robotaxi customer operations.
3. **CyberCab & Semi Production:** Both are getting ready for production.
* **CyberCab FMVSS Certification:** Images from the Austin pop-up show CyberCab stickers stating it "conforms to all applicable US FMVSS," implying Tesla's self-certification means it does not need a Part 555 exemption and will not be subject to the 2,500-unit cap.
* **Unsupervised Robotaxis:** While regulations don't restrict unsupervised robotaxis in Austin, Tesla currently appears to be operating only one (the speaker questions Whole Mars' claim of 22% of the fleet being unsupervised).
* **UI/Hardware Details:** The CyberCab UI currently shows "pedals and steering" options, and the in-car software is standard Tesla software (e.g., "open frunk" option despite no frunk), indicating it's not yet specialized for CyberCab. A small steering wheel shown in a photo is not production-ready.
4. **Optimus Production:** Elon stated in an interview with Peter Diamandis that Optimus 3 production would begin this summer, a potential pull-forward in the timeline.
**Other Notable News:**
* **Tesla-SpaceX Investment:** Tesla's previous $2 billion investment in XAI was converted into an under 1% stake in SpaceX (or a new SpaceX AI entity). This was framed as an administrative action, not an indication of an impending Tesla-SpaceX merger.
* **UK Electricity Supplier:** Tesla received the green light to supply electricity to British homes, officially becoming an approved electricity supplier in the UK.
* **S&P 500 Rules:** The S&P 500 is considering changes to its index inclusion rules, which could potentially fast-track SpaceX's entry after an IPO.
* **Honda's EV Restructuring:** Honda is facing its first annual loss in almost 70 years as a listed company, attributed to a $15.7 billion restructuring charge tied to its EV business.
* **Sponsor:** The video was sponsored by Surfshark VPN, highlighting new features like safety badges in search results, alternative ID, email scam checker, post-quantum encryption, and an independent security audit with zero critical vulnerabilities.
The speaker concludes by reiterating his long-term bullish view on Tesla, despite the short-term frustrations and increased risks/uncertainty. He believes Tesla's pursuit of these massive addressable markets, coupled with its culture, efficiency, and integrated approach, positions it uniquely for immense upside over the next 10 years, envisioning CyberCab scaling, Optimus production, Semi production, energy growth, and agentic AI in vehicles within the next 12 months.
摘要
Go to https://surfshark.com/electrified or use code ELECTRIFIED at checkout to get 4 extra months of Surfshark VPN! A massive ...
GPT-4正在为你翻译摘要中......
中英文字稿 
I followed Tesla for a long time. I think right now, maybe even this week, we've had more fundamental changes with the company than I can remember ever. I mean, the TerraFab is now happening. Tesla and XAI in this agentic path is now happening. There's massive shake-up at XAI right now. Elon is reorganizing that company. A show seems to be overseeing that. CyberCab and semi are getting ready for production. I could go on and on. So on one hand, we have more paradigm-changing company altering plans from Tesla than we've ever had before. But on the other hand, I think we have more people following Tesla that are confused, maybe disillusioned, upset with progress. They're sick and tired of the promise after promise of the future. The future next year, a few years it's coming. And a lot of people just want something to be delivered now. And I get it. It's been a long five years. Profitability has stalled and retreated.
我关注特斯拉已经很长时间了。我认为就在最近，甚至可能就是这周，公司发生的根本性变化比我以往记忆中的都要多。我指的是，TerraFab现在已经启动。特斯拉和XAI在这条代理路径上也正在进行着巨大的变革。XAI内部目前正在发生重大动荡，埃隆正在对这家公司进行重组，似乎是由一个名为"A show"的团队在监督。CyberCab和半挂车也都在准备投产。可以列举的事情还有很多。一方面，我们看到特斯拉比以往任何时候都拥有更多改变游戏规则的计划。但另一方面，我觉得有更多人关注特斯拉时感到困惑、也许是失望，对发展的速度不满。他们厌倦了一次又一次的承诺"未来"，不论是明年还是几年后兑现的承诺。很多人只是希望现在就能看到实际的成果。我理解这一点，毕竟过去五年确实时间漫长，盈利能力停滞甚至有所倒退。
And it's a fact. Tesla stock was higher back in November 2021 than it is today. So I think we have to at least acknowledge the folks that are upset a bit and maybe are frustrated and feel like Tesla the company they knew and loved is now completely different. Going a different direction, maybe there's some distraction taking place. I just want to acknowledge these concerns and frustrations are not unfounded. However, I want to do everything I can to try to make all of this change as clear as possible so we can all understand why Tesla is doing this and why now. And most importantly, why I personally do believe that for the next three, five, seven, ten years, these decisions Tesla is making are the right ones. But one of them, I have to admit, I am not positive.
事实上，特斯拉的股票在2021年11月时比现在更高。因此，我认为我们至少应该承认那些感到有些不满和可能沮丧的人们，他们觉得曾经熟悉和喜爱的特斯拉公司现在完全不同了。公司方向发生了变化，可能还存在一些干扰。我只是想表达，这些担忧和挫折并非毫无根据。然而，我想尽我所能使所有这些变化尽可能清晰明确，以便我们能理解特斯拉为何在此时做出这些决定。最重要的是，我个人确实相信，在接下来的三年、五年、七年甚至十年里，特斯拉正在做的这些决策是正确的。不过，有一点我要承认，我并不十分确定。
Today's video, the state of the union, is going to be a bit different. Not as news focused will still cover everything. But thanks to all of your comments, I now understand that some of you are still not grasping this whole agentic AI. And so, primarily, I want to make sure that everybody here understands this next phase and why it's so important and why Tesla's approach is going to be different and better than the competition, at least as planned. And I just want to go over some things and share some thoughts because truly, I'm assuming some of you are also overwhelmed because I am having moments of overwhelmed trying to keep track of everything going on. The AI space alone is moving so fast every week there's a new breakthrough, a new company with a new feature or a new tool. And as we know, Tesla is an AI company so those worlds are colliding more than they ever have.
今天的视频，关于联盟现状的内容，会有所不同。虽然仍然会涵盖所有内容，但重点不再是新闻报道。感谢所有的评论，我了解到你们中仍有一些人对自主AI的概念还不是很明白。因此，我主要想确保大家都理解这个新阶段的重要性，以及为什么特斯拉的方法会和竞争对手不同，并且至少理论上会更好。我也想讨论一些事情并分享一些想法，因为我猜测你们中有些人可能和我一样感到有些不知所措，因为要跟踪所有事情实在是很难。单是AI领域，其变化速度之快令人震惊，每周都有新的突破、新公司推出新功能或新工具。而我们知道，特斯拉是一家AI公司，因此这两个世界的交融比以往任何时候都要密切。
And last thing real quick, it was killing me not to upload Thursday and Friday, but I did tell you guys, I'm not out of the woods yet for behind the scenes stuff with electrified in this career reset going on. With each passing day, we're getting a bit closer, but there's still a lot going on behind the scenes, so thank you for your understanding. One thing I wanted to clarify, last episode I talked about the AI4 and the cost being $650. There were some comments that the actual cost for AI4 would be $1300 because there are two AI4 system on ships for each module. My understanding was that when Elon said AI4, he was talking about the entire chip model that would include two AI4 SOCs. So that was my read. I could certainly be wrong on it. It's either 650 or 1300. Either way, Tesla is incredibly well positioned from an inference compute standpoint.
快速说一下，周四和周五没能上传内容真是让我感到煎熬，不过我之前也告诉过大家，由于与电气化和这个职业重启相关的幕后工作，我还没有完全走出困境。随着时间的推移，我们一点点接近目标，但幕后仍有很多事情在进行中，所以感谢大家的理解。我还想澄清一件事，上集我提到AI4的成本是650美元。有评论说实际成本是1300美元，因为每个模块上有两个AI4系统芯片。我的理解是，当埃隆提到AI4时，他指的是整个芯片模型，包括两个AI4系统芯片。这是我的看法，我可能会错。无论是650还是1300美元，不论怎样，从推理计算的角度看，特斯拉都处于非常有利的地位。
I'm sure most of you have seen it by now, but Elon's post on this agentic AI that Tesla is now pursuing. It will work on all AI4 cars, so you can do office work when the car is not driving. And Tesla is also deploying millions of dedicated digital optimy units in the field at superchargers where they have around seven gigawatts of available power. And Max from Tesla said, the supercharging teams will be helping to get digital optimist units energized, especially at the long distance superchargers where they already have the infrastructure built. And a lot of the times outside of holidays, it sits idle. Also, no worries, I will fix my lighting in future episodes. I'm working on a whole new setup, getting ready for some interviews, and just again, shaking things up.
我相信大多数人现在已经看过埃隆的那篇关于特斯拉正在研究的代理型人工智能的帖子了。这个AI系统将应用于所有AI4汽车，这样当车辆不在行驶时，你就可以在车内处理办公事务。而且，特斯拉正在为超级充电站部署数百万个专用的数字优化单元，这些充电站大约有七吉瓦的可用电力。特斯拉的马克斯表示，超级充电团队将帮助为数字优化单元供能，特别是在那些已经建有基础设施的长途超级充电站，那里通常在非假期时间处于闲置状态。另外，不用担心，我会在未来的节目中调整好照明。我正为一些采访准备全新的设备，并努力带来一些改变。
And Elon said this very feature could be available in six months, which is exactly what we talked about. The path to market for digital optimists should be much faster and easier than physical. Some of you may have seen that bright speed data breach that just exposed over a million customers, personal info, names, emails, addresses, all of it. Well, Surfshark, the sponsor of this video, actually just put out a new feature that could help you avoid sites like that before you even click them in the first place. Surfshark's browser extension now shows safety badges right inside your Google search results. So before you even click on a link, you can see if that website has had a data breach or if it has malware. Surfshark flagged over 141,000 online threats last year alone. So this is a layer of protection. I didn't really know I needed, but now I'd rather not browse without.
埃隆表示，这个功能可能在六个月内就能推出，这正是我们谈到的。对于数字乐观主义者来说，进入市场的路径应该比实体产品快得多、容易得多。你们中的一些人可能听说过Bright Speed的数据泄露事件，它曝光了超过一百万客户的个人信息，包括姓名、邮箱、地址等。Surfshark，是这个视频的赞助商，刚刚推出了一项新功能，可以帮助你在点击之前避免访问那些有风险的网站。Surfshark的浏览器扩展现在在你的谷歌搜索结果中直接显示安全标识。在你点击链接之前，你就可以看到该网站是否曾发生过数据泄露，或者是否有恶意软件。仅去年一年，Surfshark就标记了超过141,000个在线威胁。这是我之前不知道需要的保护层，但现在，我更愿意不浏览没有它的网页。
And if you're signing up for new sites this year, which most of us are, Surfshark's alternative ID lets you use a completely separate name, email, and number so your real info never touches those databases in the first place. Pair that with Surfshark's email scam checker in Chrome and Gmail that uses AI to flag phishing attempts before you fall for them. And I think that makes for a pretty comprehensive setup. Surfshark also just completed a brand new independent security audit with zero critical vulnerabilities found, and they've rolled out post-quantum encryption, which will future proof your data against any next-gen threats. And one account still covers unlimited devices for your whole family. And this list is not even close to an exhaustive feature set when it comes to all of what Surfshark offers.
如果你今年打算注册一些新网站（大多数人都会这样做），Surfshark的替代身份功能允许你使用完全不同的姓名、电子邮件和电话号码，这样你的真实信息就不会被这些数据库获取。再加上Surfshark的电子邮件诈骗检查器，可以在Chrome和Gmail中使用AI技术在你上当之前标记钓鱼尝试。这套工具组合非常完善。此外，Surfshark刚刚完成了一次全新的独立安全审计，未发现任何关键漏洞，并推出了后量子加密技术，可以保护你的数据，抵御未来的高级威胁。而且，一个账号仍然可以覆盖你全家所有设备。Surfshark提供的功能远远不止这些。
So if you'd like to support the work I do here, you can go to surfshark.com slash electrified or use code electrified at checkout to get four extra months of Surfshark VPN. And now if you already understand a Genetic AI, then feel free to skip ahead a little bit. I just want to do my best to not leave anybody behind. When you hear a Genetic AI, just think of like a sports agent and what they do. They're the ones coming up with the contracts, they're setting the meetings, negotiating the contracts, trying to get their players the best deal that they can. They're the ones actually executing all of the work, and the player is giving the instructions on what they want the agent to actually do.
如果你想支持我在这里的工作，可以访问 surfshark.com/electrified，或者在结账时使用代码 electrified 来获得额外四个月的 Surfshark VPN 服务。如果你已经了解遗传AI，可以略过这一段。我只是想尽力不让任何人掉队。听到遗传AI时，可以想象成一个体育经纪人以及他们的工作。他们负责制定合同、安排会议、谈判合同，努力为他们的球员争取最好的交易。他们是在实际执行所有工作的那个人，而球员则是给出他们希望经纪人执行什么的指示。
So all this Genetic AI really is, is you give a command, you tell the computer what you wanted to do, maybe build me a Tesla financial model, and then you sit back and do nothing and the computer builds it itself. And now Tesla's position getting into all of this, where there is admittedly a lot of competition. But we know that one of the main bottlenecks right now is access to power. The interconnection cues can be months long at this point. So these companies getting access to new power is not easy. Tesla meanwhile has seven gigawatts of available power sitting ready to use. So rather than all of that power at Superchargers just sitting idle, Tesla is now going to plug in some AI for racks to tap into that already existing power and then to have those AI for chips actually become agents.
所以，所谓的“遗传人工智能”其实就是，你给电脑一个指令，告诉它你想要它做什么，比如构建一个特斯拉的财务模型，然后你什么都不用做，电脑会自行完成。而特斯拉在这其中的地位则进入了一个竞争激烈的领域。我们知道，目前的主要瓶颈之一是电力的获取，电力的接入时间可能长达数月。因此，这些企业要获得新的电力并不容易。然而，特斯拉拥有7吉瓦的可用电力，随时可以利用。因此，与其让这些在超级充电站的电力闲置，特斯拉现在打算接入一些AI设备，利用这些现有的电力，让这些AI芯片实质上成为“代理”。
And we also know what Tesla has at a lot of supercharging locations, which is Wi-Fi. One simple example of how it could work, we get a software update, we talk to Girok just like we do now. So let's say I need to find a local contractor in my city to build a custom deck and they have to be highly rated on Yelp. So I would explain this to Girok and tell it to research and find the best three options and then send an email to all three requesting an appointment for a quote. And then from there the team of AI for and Girok and XAI servers would work together and either tap into the user interface of your car or even your computer at home or a laptop with you.
我们也知道，特斯拉在很多超级充电站都提供Wi-Fi。这是一个简单的例子解释它是如何运作的：假设我们收到一个软件更新，就像我们现在跟Girok交流一样。比如说，我需要在我所在的城市里找一个本地承包商来建造一个定制的露台，而且他们在Yelp上的评价必须很高。我会把这些要求告诉Girok，让它去研究并找出三个最佳选项，然后给这三家公司发邮件，预约一个报价的时间。接下来，AI团队、Girok和XAI服务器会协作，可能会通过你的车载用户界面、家里的电脑或者与你在一起的笔记本电脑来完成这些任务。
And that becomes the workspace where digital optimists as we call it goes to work. It navigates the screen and does the task while you sit there. And a lot of the agentic AI right now as you're seeing on the screen you can actually see what it's thinking and what it's doing. So you can watch it do the research and browse the web and write the emails and send the emails in real time and so on. And even if you don't have a laptop in the car with you, Grok in time should be able to connect to Google Drive or your file system and then work on your spreadsheets and your documents and your emails with the car's AI for chip and Grok as the project manager.
这就成为了一个工作空间，我们称之为“数字乐观主义者”的人可以在这里工作。它浏览屏幕并执行任务，而你只需坐在那里观看。目前你所看到的许多智能代理AI，它会在屏幕上显示自己的思考过程和操作步骤。因此，你可以实时观看它进行研究、浏览网页、撰写和发送邮件等。即使你在车上没有笔记本电脑，Grok未来也应该能够连接到Google Drive或你的文件系统，然后与车载AI一起处理你的电子表格、文档和邮件，Grok作为项目经理来协调这些工作。
But the cost structure here is going to be critical as you'll see shortly. Other companies like Anthropic and Proplexity are building AI data centers paying billions of dollars for land, construction, facilities, power interconnects, cooling, and so on. Tesla and XAI already have all of that. They have data centers and supercharger locations and that seven gigawatts of available and the best inference hardware in the world with AI for meaning Tesla's cost per inference unit should be a fraction of the traditional cloud-based structure.
但是，这里的成本结构将是关键，您很快就会明白这一点。其他像 Anthropic 和 Proplexity 这样的公司正在建立 AI 数据中心，为土地、建筑、设施、电力互连、冷却等方面支付数十亿美元。而特斯拉和 XAI 已经具备了这些条件。他们有数据中心和超级充电点，并拥有7吉瓦的可用电力和世界上最好的 AI 推理硬件，这意味着特斯拉每个推理单元的成本应该只是传统云计算结构的一小部分。
And just to be clear, Tesla's AI for is way more powerful than the Mac mini. It's designed to handle video data from eight different camera feeds in real time. So taking in video data from one screen is going to be a joke for AI for. And this should work out to be a big differentiator because most of the agentic AI right now, the competition is cloud-based, which at this point is still quite expensive as I'll show you. But Tesla and XAI are planning to do a lot of the computing tasks locally on the car with AI for only having to reach out to the cloud for further instruction every now and then.
为了更清楚地说明，特斯拉的AI远比Mac mini强大得多。它被设计用来实时处理来自八个不同摄像头的视频数据。所以处理来自一个屏幕的视频数据对AI来说可以说是一件轻而易举的事情。这将成为一个重要的差异化因素，因为目前大部分的智能AI竞争都基于云端，而云计算在现阶段仍然相当昂贵，如我将要展示的。而特斯拉和XAI计划在车上本地处理大量计算任务，只有在需要进一步指令时才会偶尔连接云端。
And perhaps the biggest differentiator for Tesla. Most software automation right now requires custom integrations for every single app you want to control. Excel, Google Sheets, Google Drive, browsing the web, Gmail, the list goes on and it's usually through APIs or scripts. But digital optimists sounds like it's going to skip all of that. It will literally just watch the screen like a human operating on that pixels in and controls out just like FSD, which would mean in theory it would work with any software.
也许对特斯拉来说，最大的区别在于他们的技术。大多数的软件自动化目前都需要针对你想控制的每一个应用进行定制整合，比如 Excel、Google Sheets、Google Drive、浏览网页、Gmail 等等，通常通过 API 或脚本来实现。但是，“数字乐观主义者”的方法似乎会跳过这一切。它实际上就像一个人在屏幕上观察和操作，输入像素、输出控制，就像全自动驾驶（FSD）一样，这意味着理论上它可以适用于任何软件。
It does not need permission from app developers, it handles any problems on the screen because it sees them directly and can jump from one app to another seamlessly. And so just like with FSD, it outputs control of the wheel and of the pedal and so on. And in a similar fashion, controls can also be mapped to moving a cursor and assimilating keystrokes instead. Another differentiator, most of the agentic AI right now works on still images or screenshots.
无需获得应用开发者的许可，它可以直接处理屏幕上的任何问题，因为它能够直接看到这些问题，并且可以无缝切换不同的应用程序。因此，就像全自动驾驶(FSD)一样，它可以输出对方向盘、油门等的控制。类似地，还可以将控制映射为移动光标并模拟按键操作。另一个区别在于，目前大多数主动型人工智能主要处理静态图像或屏幕截图。
And don't get me wrong, it's still very useful, but it's limited. Digital Optimists on the other hand will work with video of the screen so any errors or pop-ups or menu navigations that are totally up to date not relying on an outdated screenshot means it's going to handle whatever comes up directly in the screen. And as far as we know, digital optimists would be the only one able to do that.
虽然它仍然非常有用，但也有局限。另一方面，数字乐观者会采用屏幕视频的方式进行工作，这样一来，无论是错误信息还是弹出窗口，或者是最新的菜单导航，都能即时处理，而不是依赖过时的截屏信息。据我们所知，目前只有数字乐观者能够做到这一点。
So when the mouse moves to a certain point and a menu pops up, you get more options. When you hit submit, the button changes color and a new page load. So over time, the AI is going to learn all of this just like we have and again, it'll react in real time. So eventually, digital optimists will have a complete UI world model from pixels alone and a neural network to look at those raw pixels and instantly understand all of it.
当鼠标移动到某个点并弹出菜单时，你会获得更多选项。当你点击提交按钮时，按钮会改变颜色并加载新页面。随着时间的推移，人工智能会像我们一样逐渐学会所有这些，并能够实时作出反应。因此，最终，数字乐观主义者将能够仅通过像素建立一个完整的用户界面世界模型，并使用神经网络来观察这些原始像素，瞬间理解其中的一切。
And this is very different than existing options that use certain tools like APIs, browser doms, and different hooks. And doms are just document object models, which is basically a map of a web page structure in code form. But digital optimists will just see the screen like we do. No tools or hacks or workarounds needed at all.
这与使用某些工具如API、浏览DOM和不同钩子的现有选项非常不同。DOM实际上就是文档对象模型，本质上是一个网页结构的代码形式的映射。然而，数字乐观主义者将会像我们一样直接看到屏幕。完全不需要使用工具、黑客手段或其他变通方法。
And right now, you basically need a different tool or a workaround for each different app or interface or platform that you want to use, which is laborious, slow, expensive, and unreliable. But with digital optimists, all of that is removed. So anything with a screen, Windows, Mac, Linux, web apps, legacy tools, video games, it will just know how to navigate it and operate it.
现在，你基本上需要为每个不同的应用程序、界面或平台使用不同的工具或变通方法，这既费力、又慢、成本高，而且不太可靠。但对于数字乐观主义者来说，这些问题都能解决。无论是Windows、Mac、Linux，还是网页应用、老旧工具或视频游戏，任何带屏幕的东西，它都能轻松操作和导航。
All while being hyper efficient with the ability to scale to massive numbers, thanks to Tesla's vehicle fleet and AI series production already in place. And so Tesla's solution, the way they're describing it, should be a lot faster, a lot more reliable, much more efficient than other options available on the market now.
得益于特斯拉已经建立的车辆队列和AI系列生产能力，特斯拉能够在极高效率的同时，实现大规模扩展。因此，根据特斯拉的描述，他们的解决方案应该比目前市场上的其他选项要快得多、更可靠且更高效。
But this is what I want to show you. This is my perplexity computer back in. So every month for $200 a month, you get 10,000 credits. And you'll see right here, my usage for this month so far is 5,079 credits. And here, I just plugged in some random tasks that would make sense for this video of things that I could do for my electrified channel.
但这就是我想展示给你的内容。这是我重新使用的困惑计算机。每月花费200美元，你可以获得10,000个积分。你可以看到，到目前为止，我本月已经使用了5,079个积分。在这里，我只是输入了一些随机任务，用于这个视频中可以用于我的电气化频道的事情。
Really just testing it so I can learn what's going on, but you'll see basically in one night, I burnt through half of my available credits for the month, which effectively means I spent $100 in one night using this agentic AI. And mine was all text based. You'll see I didn't do anything with video or audio. So my guess would be that this feature is not going to be free and unlimited for Tesla owners with AI for over time, costs will certainly come down, but right now they're still pretty high.
实际上，我只是想测试一下，以便了解情况。但你会看到，几乎在一个晚上，我就用掉了这个月一半的可用额度，这实际上意味着我一晚花掉了100美元，使用的是这个有自主能力的AI。而我使用的全部都是文字功能。你会注意到，我没有用到视频或音频。 所以我猜，这个功能对特斯拉车主来说不会是免费和无限制的。随着时间的推移，成本肯定会下降，但现在仍然相当高。
Now could Tesla cut these costs by 70% because so much of the compute is done locally on Tesla's inference compute? I think absolutely that's possible. But I think we at least have to consider there might be another subscription to actually use this feature or it'll be capped. But really who knows, maybe Tesla decides to eat all of those costs just to offer the feature and get it out there for the marketing? Because of course in the early days, this is not going to be a core part of Tesla's business. But by far, one of the main takeaways here is that Tesla's inference efficiency and intelligence density per gigabyte should come in to play in a major way here. As it is right now, it's very easy to burn through credits with the current agentic AI on the market that is primarily cloud based.
现在，特斯拉是否可以通过在本地计算上进行大量推理计算，将成本降低70%？我认为这绝对有可能。但是，我认为我们至少要考虑可能会需要另一个订阅来实际使用这个功能，或者说功能使用会有限制。但谁也说不准，也许特斯拉决定承担所有这些成本，仅仅为了提供这个功能并进行市场推广。因为当然在初期，这不会是特斯拉业务的核心部分。但目前最大的一个关注点是，特斯拉的推理效率和每GB的智能密度在这方面应该会发挥重要作用。因为现在，使用市场上主要基于云的自主AI，很容易烧掉大量积分。
If Tesla and XAI can primarily pull this off locally, not having to constantly reach out to the cloud because it has this neural network and understanding of the world model in pixels in these screens without having to use APIs and all of these other tools, that is a major breakthrough. But it's also true that we know XAI is currently the laggard when it comes to agentic AI encoding. And as I've been defending them online, they started late. And Elon just admitted that he hired wrong at XAI, but he's fixing all of that now. And listen, I know some of you are sick are hearing things like this, but Elon just said today XAI will catch up this year and then exceed them all by such a long distance in three years that you'll need the James Webb telescope to see who's in second place.
如果特斯拉和XAI能够主要在本地实现这一点，而不需要不断连接云端，因为它们在这些屏幕上用神经网络掌握了世界模型，而不必使用API和其他工具，那将是一个重大突破。不过，目前XAI在具备自主智能编码方面确实落后。我曾在网上为他们辩护，他们起步较晚。Elon刚刚承认他在XAI招聘方面出了问题，但他正在修复这一切。我知道有些人已经厌倦了听这样的消息，但Elon今天刚刚说XAI将在今年赶上，并将在三年内遥遥领先到需要用詹姆斯·韦伯望远镜才能看到第二名的位置。
So the vision for Elon and XAI is finally abundantly clear. If you missed it, Elon has been poaching some really solid top talent from the competition and the speed run to catch up and eventually leapfrog the AI competition is fully on. But in this case, Elon and XAI are actually playing catch up and the leaders are certainly not standing still. And of course, XAI still needs to prove that it can do agentic AI well. To put some context on that seven gigawatts of power that Tesla has, that's roughly enough power for five to six million average American homes at one time. And again, that power is already in place, connected, largely sitting idle with cooling and everything required.
所以，Elon和XAI的愿景终于非常清晰了。如果你错过了，Elon一直在从竞争对手那里挖走一些非常优秀的顶尖人才，他们正在快速追赶，甚至准备超越AI领域的竞争对手。但在这个情况下，Elon和XAI实际上是在追赶，而领导者们当然不会停滞不前。当然，XAI仍然需要证明它能做好自主AI。为了让你理解，特斯拉拥有的七吉瓦电力，大约可以同时为五到六百万个普通美国家庭提供电力。这些电力已经到位并连接好了，冷却和其他所需设施也基本就绪，但大部分还闲置着。
So these new AI4 racks are just going to be plugged into that infrastructure, avoiding all of the current connection cues and costs associated. And let's assume AI4 uses about 150 watts, which is very efficient. Seven gigawatts would be enough to run around 40 million AI4 chips at full speed. And yes, long term theoretically, Tesla could offer subscriptions to third parties if companies wanted to rent out some of these AI workers that will be installed around the world at supercharger locations. However, Tesla is going to have to develop a competitive advantage to bring people to actually do that because there will be plenty of options available.
这些新的AI4机架将直接接入现有基础设施，从而避免当前各种连接延迟和相关成本。假设AI4每个芯片的耗电量大约是150瓦，这已经非常高效了。七吉瓦的电力足以让大约4千万个AI4芯片全速运行。是的，从长远来看，如果有公司希望租用这些将安装在全球超级充电站的AI工作者，特斯拉可以向第三方提供订阅服务。不过，特斯拉需要开发出竞争优势，才能吸引公司选择他们的服务，因为市场上将会有很多选择。
But at all, at the end of the day, it boils down to the cost advantage that Tesla and XAI should have. Digital Optimus just working natively locally on the screens with AI4 not having to access the cloud, where the compute costs are much higher. That is the breakthrough that no other company has figured out. And what do we always say? Tesla has the very best inference compute in the world. And that paired with all the power that Tesla already has and the fleet of AI4 vehicles Tesla already has, and the data centers that XAI and Tesla already have. I hope the light bulb is at least flickering a little bit to understand why this could be so special.
总而言之，归根结底，特斯拉和XAI的成本优势是关键。数字Optimist能够在本地主机上直接运行AI4，而无需访问云端，这样计算成本就降低了许多。这是其他公司尚未实现的突破。我们总是说，特斯拉拥有世界上最好的推理计算能力。再加上特斯拉现有的强大实力、大量的AI4车辆以及XAI和特斯拉已有的数据中心，希望大家能开始理解这为何如此特别。
And I'll be the first person to tell you, there are plenty of risks with this agentic AI, one of which is prompt injection hacks. So let's say you're at a supercharger, you tell Grock to read your emails, give you the five most important ones, and then respond to three of them. What if somebody sends you an email with instructions to give over all of your credit card and banking information? It could be laid out in plain text, or it could be hidden in an image. That is prompt injection risk. And sure, a lot of these newer models are getting a lot better at defending against that, but it's certainly not bulletproof.
我会是第一个告诉你，这种自主AI存在很多风险，其中一个就是提示注入攻击。假设你在一个超级充电站，你让Grock读你的邮件，挑选出五封最重要的，然后回复其中的三封。如果有人给你发了一封邮件，要求你提供所有的信用卡和银行信息，那该怎么办？这些信息可能直接写在邮件里，也可能隐藏在一张图片中。这就是提示注入的风险。当然，很多最新的模型在防范这方面变得更擅长了，但肯定还没有达到万无一失的程度。
And that's why personally, I have not yet given any of these agents access to any of my most important documents without backing them up in the cloud separately, just in case something goes sideways and everything gets deleted or destroyed or whatever. But to me, it could not be clear where all of this is headed. There's going to be a world where we will open our Tesla app from anywhere on earth. We will prompt Grock whether with text or our voice. And then our Tesla vehicles will be powering all of the work that we need done and it will do it for us.
这就是为什么我个人还没有允许这些助手访问我最重要的文件，而是先在云端备份了一份，以防万一出现意外导致文件被删除或损坏。但在我看来，未来的发展方向已经很明显。我们将进入一个这样的世界：无论身处地球上的哪个角落，我们都可以打开特斯拉的应用程序，通过文本或语音指令给Grock下达任务，然后我们的特斯拉汽车将完成我们需要的所有工作。
And I know it sounds absolutely insane if you're not familiar with this world, but I can assure you this is where it's headed, whether it's two years or five years I cannot tell you, and to some degree is here already. And so eventually we'll just be able to say Grock, build me a Tesla website with the history of the company and the current Wall Street price targets in a community forum section to connect with other Tesla fans and a membership section where I can charge $10 a month for exclusive content and interviews. And then as you sit in your house or in your car, digital optimists will just do the work for you. It will build all of it and host it and then you can tweak it.
我知道如果你对这个世界不熟悉，这听起来简直不可思议，但我可以向你保证，这就是未来的发展方向，不论是两年还是五年，我无法告诉你，但在某种程度上，这种趋势已经出现了。因此，到时候我们只需说：“Grock，帮我建一个特斯拉的网站，包含公司的历史、当前华尔街的价格目标，还要有一个社区论坛部分，以便与其他特斯拉粉丝交流，并设一个会员专区，我可以收取每月$10的费用，提供独家内容和采访。”然后，无论你是在家里还是车里，数字化的助手都会帮你完成所有工作，它会建立并托管网站，然后你可以进行微调。
And as I've seen with my own eyes with some of these front tier models, that capability already exists today, maybe not one shot, but with a few tweaks and if you know what you're doing, it's already happening. And again, if you want an easy way to familiarize yourself with this world, I would recommend perplexity computer. It's now available to pro users for just $20 a month, but you have the ability to give it whatever crazy task you want to try to give it and see what it can do. And I really don't think I need to explain the upside in the addressable market of all of this. Any job or task that is done on a screen, any game that is played on a screen in time is likely to be done better by digital optimists or the competition's agentic AI.
根据我亲眼所见的这些前沿模型，这种能力已经存在于今天的世界中。可能不是一次就能完成，但经过一些调整，如果你懂得如何运作，这已经在发生。如果你想轻松了解这个领域，我推荐使用Perplexity电脑。现在专业用户每月只需$20即可使用。你可以给它任何你想尝试的任务，看看它能做什么。我真的觉得不需要解释这一切在可达市场中的潜在好处。任何在屏幕上完成的工作或任务，任何在屏幕上进行的游戏，随着时间的推移，很可能由数字乐观主义者或竞争对手的人工智能代理来更好地完成。
And for Tesla and XAI, a lot I would argue almost all of the infrastructure and hardware and software is already in place or in the works. So given that this is where the world is headed and the addressable market is so incredibly massive, I think it would be a mistake for Tesla and XAI not to pursue this. So I understand it may feel like a distraction and we want things like summon and 14.3. So we don't want to hear that a show's attention is taken away from FSD at all. And just real quick, on that point, people out there are saying that a show is now overseeing MacroHard and like he's leaving the autopilot team, there's mixed reporting out there on it right now.
对于特斯拉和XAI来说，我认为几乎所有的基础设施、硬件和软件都已经到位或正在开发中。鉴于世界的发展方向以及市场的巨大潜力，我认为特斯拉和XAI不去追求这个机会将是一个错误。我明白这可能看起来像是分心，因为我们希望能有像召唤功能和版本14.3这样的东西。所以我们不希望听到任何关于注意力从自动驾驶功能上被转移的消息。在这个问题上，还有一些人在说某位负责人现在正在监管MacroHard，好像他要离开自动驾驶团队，目前关于此事的报道有些不同。
For me, I'm guessing Elon just temporarily wants a show to oversee the XAI team and all of the work that they've done to get them on the right track as quickly as possible. And further, to get their team integrated with Tesla's team and structure and tools. Since again, they have that partnership now. And of course, for somebody as brilliant as a show, Elon's gonna want his hand in as many cookie jars as possible. But I would imagine once XAI hires a new talent and gets them integrated and gets all of the team settled and heading the right direction, then a show can come back and focus on FSD if he needs to.
对我来说，我猜埃隆可能暂时想让Ashok负责监督XAI团队及其工作，以便尽快将他们引导到正确的轨道上。此外，还希望将他们的团队与特斯拉的团队、结构和工具整合在一起。毕竟，他们现在有合作关系。当然，对于像Ashok这样出色的人才，埃隆会希望尽可能多地参与各方面的工作。但我想，一旦XAI招聘了新人才并将他们整合到团队中，使团队稳定并朝着正确方向前进，Ashok就可以返回专注于FSD（全自动驾驶）项目，如果有需要的话。
Shifting gears a bit, you may have seen SIRN shared some images of the cyber cap at the pop-up currently underway in Austin. But you'll see the stickers on the cyber cap right here. This vehicle conforms to all applicable US FMVSS. Which means Tesla has officially said, we believe this meets every rule and we do not need an exemption. Which is the massive step of clarification that I personally was waiting for. Now this is just a self certification from Tesla. So NITSA could absolutely still challenge Tesla's interpretation.
转换一下话题，你可能已经看到SIRN分享了一些有关在奥斯汀正在进行中的快闪活动中展示的Cyber卡车的图片。你可以在这里看到Cyber卡车上的贴纸。这个车辆符合所有适用的美国联邦机动车辆安全标准（FMVSS）。这意味着特斯拉官方认为这辆车满足了所有规定，不需要任何豁免。这对于我个人来说是一个重要的澄清步骤。不过，这只是特斯拉的自我认证，因此美国国家公路交通安全管理局（NHTSA）仍然可能对特斯拉的解释提出质疑。
But for now, Tesla is saying, we meet all of the criteria and we will not be subject to that 2500 unit cap. Which as we were hinting at last week, Tesla has not filed for any part 555 exemption and with CyberCab production starting in a few weeks, we would have likely seen that if they thought they would need the exemption. So until further notice and unless NITSA is going to challenge Tesla's interpretation and self-certification here, then the CyberCab will not need that part 555 exemption and it will not be subject to the 2500 unit cap. Which yes, I don't need to explain is absolutely great news.
但就目前而言，特斯拉表示，他们满足了所有标准，因此不会受到2500辆汽车上限的限制。正如我们上周所提到的，特斯拉并没有申请任何第555部分的豁免，而随着CyberCab生产将在几周内启动，如果他们认为需要豁免，我们可能已经看到了申请。因此，除非国家道路交通安全管理局（NITSA）挑战特斯拉的解释和自我认证，否则CyberCab将不需要该第555部分的豁免，也不会受到2500辆上限的约束。是的，我不需要解释这绝对是个好消息。
So Tesla is now free to scale the CyberCab. Now I do think a lot of this debate around if FSD is solved or not comes down to semantics and what each person means by being solved. But I'm personally saying, I think we should stop saying that because in my eyes, it's not actually solved until Tesla is deploying hundreds and thousands of unsupervised rubble taxis. Until then, it will remain very close. But if it was truly solved, meaning the software was as safe as it needed to be, I'm pretty confident Tesla is not going to gatekeep this technology and only have one unsupervised rubble taxis riding around Austin if it was really ready to be scaled.
所以特斯拉现在可以自由发展CyberCab了。不过，我确实认为关于FSD（全自动驾驶）是否真正实现的争论，很多时候是说法上的问题，每个人对“实现”的定义不同。但我个人认为我们应该停止说它已经实现了，因为在我看来，特斯拉还没有大量部署无人监督的自动驾驶出租车之前，它并不能算真正实现。目前离这个目标很接近了，但如果它真的已经实现了，也就是说软件已经足够安全，我相信特斯拉不会限制这项技术的使用，仅仅让一辆无人监督的自动驾驶出租车在奥斯汀运行。如果它真的准备好进行大规模应用，情况就不是这样了。
Some people are saying regulations are holding it back, but in Austin that's not the case. Tesla is free to have as many unsupervised rubble taxis in Austin as it wants. And as far as we can tell, it's choosing to have one. Now I've said the third party rubble taxi tracker should not be taken as gospel, but whole Mars is arguing that 22% of Tesla's Austin fleet is now unsupervised. I asked him where he got that number, but he didn't respond directly. But I just wanted to be clear, the way I am defining solved is it's ready to actually scale. Because as we know, it's never going to be done. It's always going to continue getting better. More data coming in, bigger fleets, more edge cases turned out, continually refined into perpetuity. Models becoming smarter and more efficient and so on. But truly, I would love to hear from you guys how are you defining solved for FSD? Do you think it's solved or are we not there yet? For me, of course, I think we're very close, but I am very much in the camp that we are not there yet.
有些人说监管阻碍了发展，但在奥斯汀并非如此。特斯拉在奥斯汀可以随意运营无人监督的自动驾驶出租车。据我们所知，他们选择了只运营一辆。我之前说过，不要把第三方的出租车跟踪数据当作权威，但Whole Mars主张特斯拉在奥斯汀的车队中有22%现在是无人监督的。我问过他这个数据从哪里来的，但他没有直接回应。不过我想明确的是，我定义的问题解决了，意思是它已经可以大规模推广了。因为我们知道，问题永远不会完全解决，它总是在不断变得更好。随着更多数据的进入，更大的车队，更多的极端情况被发现，持续精炼下去。模型变得更智能、更有效等等。但说实话，我很想听听你们怎么定义FSD（完全自动驾驶）的“问题解决”。你们认为问题解决了吗，还是我们还没到那一步？对我来说，我当然觉得我们非常接近了，但我仍然认为我们还没到那个阶段。
Now, if we get to 100 unsupervised vehicles where any random person can go to Austin and get an unsupervised vehicle, relatively easy, without 60 tries or without knowing all of the hacks and the specific geofence it's operating in right now, you know, the original geofence that's inside of the bigger geofence. And listen, even if Omar's number that he threw out there of 22% being unsupervised is true, if the fleets around 50, that's only roughly 10 unsupervised vehicles. To me, that's very much still more of a testing beta validation phase than actually saying, hey, we're done, it's good, we can absolutely scale now. Because again, for me, that's what solved means. I just keep looking at my screen and realizing how brutal the lighting is right now, so apologies, but all the new stuff is in transit. Let's do some speed running of our own here.
现在，如果我们能达到100辆无人监管的车辆，而任何人都可以相对轻松地在奥斯汀获取一辆这样的车，而不是尝试60次或掌握所有的窍门以及目前它运行的特定地理围栏（有个更大的地理围栏里还有个原始的地理围栏），即使奥马尔提到的22%的车辆是无人监管的这一比例是对的，如果车队有50辆车，那么也只有大约10辆是无人监管的。对我来说，这仍然更像是测试或验证阶段，而不是说"好了，我们可以完全大规模推行了"。因为在我看来，这就是解决问题的意思。我注意到屏幕上的光线太差了，所以抱歉，不过所有的新设备都在运输中。让我们也来快速处理一下自己的事情吧。
So this Tesla converting XAI investment into SpaceX steak. So many people were framing this news as some events that means a Tesla and SpaceX merger is now much more likely. In my opinion, that's absolutely not the case. This was effectively just administrative. All this is was that $2 billion that Tesla already invested into XAI months ago being converted to reflect a steak in SpaceX, aka the new SpaceX AI entity. So yes, Tesla investors now officially own a portion of SpaceX under 1%, really not going to be material for Tesla investors. So really, this was effectively a nothing burger. There's just a lot of people in the community right now that are reaching for anything to make it seem like Tesla and SpaceX are about to merge. Plenty of people really would love to see that happen and look, I totally understand why. But again, this specifically, effectively, a non-event.
所以这件事是特斯拉将对XAI的投资转变为SpaceX的股份。很多人将这一消息解读为特斯拉和SpaceX合并的可能性大大增加。但在我看来，绝对不是这么回事。这实际上只是一个行政上的调整。事情的来龙去脉是：特斯拉几个月前投资XAI的20亿美元，现在被转成了SpaceX的一部分股份，也就是新的SpaceX AI实体。所以现在特斯拉的投资者确实拥有了不到1%的SpaceX股份，但这对特斯拉的投资者来说几乎没有实质影响。所以，这基本上是一件无关紧要的小事。现在社区里有很多人试图抓住这些信息，让它看起来像特斯拉和SpaceX即将合并。很多人确实希望看到这样的事情发生，我完全可以理解。但仅就此事而言，这实际上是一个无足轻重的事件。
Tesla did get the green light to start supplying electricity to British homes, so Tesla is now officially approved as an electricity supplier in the UK. The S&P 500 is considering changes to rules governing how companies join the index. This could potentially fast track SpaceX's entry after IPO. The index provider is engaging with stakeholders to determine whether there's demand for changing the rules, but for now, no decision has been made. And further, the S&P would still have to launch a formal consultation that would last several weeks before any changes can be made. Honda is facing its first annual loss in almost 70 years as a listed company after its $15.7 billion restructuring charge tied to its EV business.
特斯拉已经获得批准，可以开始向英国家庭供电，因此特斯拉现在被正式认定为英国的电力供应商。标准普尔500指数正在考虑修改企业加入该指数的相关规则，这可能会在SpaceX进行首次公开募股后，加速其进入标准普尔500指数的进程。目前，指数提供商正在与相关方沟通，以确定是否有必要修改规则。但目前尚未做出任何决定。此外，标准普尔还需要进行一个正式的磋商，该过程将持续数周，之后才能做出任何更改。由于与电动汽车业务相关的157亿美元的重组费用，本田公司作为上市公司，面临近70年来的首次年度亏损。
Some photos are coming out from the pop-up in Austin and Owen said that the Windows switches and the USB-C ports on the CyberCab have been moved to under the screen, which is odd because that means the cords would have to run from the screen across the leg area all the way to the seat, whereas previously they were right on the seat by the cup holder. So again, plenty of folks are throwing around production version and we still have a few weeks so Tesla might continue to make some final tweaks before production. Just FYI on the whole a-show situation, so here, financial time said Elon has redeployed Lswamy head of AI software at Tesla to reboot the macrohard effort and review the work done previously.
一些照片从奥斯汀的快闪活动中流出，Owen说CyberCab上的Windows开关和USB-C接口已经移到了屏幕下方，这很奇怪，因为这样的话电线就得从屏幕穿过腿部区域一直延伸到座位，而之前这些接口是直接在放杯架旁边的座位上的。所以，很多人都在猜测这是量产版本，不过我们还有几周时间，特斯拉可能会在生产前继续做一些最后的调整。关于整个展览情况，顺便说一下，《金融时报》表示Elon已经重新指派特斯拉AI软件团队负责人Lswamy，以重新启动那个项目并审核之前完成的工作。
So to me, that does not mean that a show is leaving the autopilot team and now just leading macrohard. It seems very temporary just using a show to help get things on track. We won't go into specifics today when it comes to what's going on at XAI, but Elon did say XAI was not built right the first time around, so it's being rebuilt from the foundations up. On one hand, you can argue Elon made a mistake with his hiring at XAI. That's not good. On the other hand, you can argue at least Elon is able to admit he made a mistake and can pivot when need be. Elon said many talented people over the past few years were declined in offer or even an interview at XAI, my apologies. Barrisoni are going through the company interview history and reaching back out to promising candidates.
对我来说，这并不意味着某个节目正在离开自动驾驶团队并开始领导宏硬公司。这似乎只是暂时利用某个节目来帮助事情走上正轨。我们今天不会详细讨论XAI的具体情况，但Elon确实表示XAI在初次建立时并不理想，因此正在从基础上重建。从一个角度来看，你可以说Elon在XAI的招聘上犯了错，这不好。但从另一个角度来看，你可以说至少Elon能够承认自己的错误，并在需要时进行调整。Elon提到在过去几年中，很多有才华的人在XAI都没有被给予机会或面试，对此我表示道歉。Barrisoni 正在回顾公司面试历史，并重新联系有潜力的候选人。
Tesla has a job listing up right now in Palo Alto for a senior program manager for Robotaxi customer operations. The focus will be on delivering exceptional rider experiences. Sparking the debate AI driver shared a photo from the cyber cab UI that does show a pedals and steering option. However, David Moss shared this photo of the steering wheel on the cyber cab. And I think most of us would agree this little number right here would not be something that makes it into production. But yes, that was a little setup on my part. We had Holmars saying it's just standard Tesla software that's running on all cars. The cyber cab does not have a frunk, but it still says open frunk. That shared in an image from David Moss and Eric, the lead cyber cab engineer today said this is correct. So the in-car software for the cyber cab at least for these vehicles is just standard. It has not yet been modified to be specific to cyber cab.
特斯拉目前在帕洛阿尔托发布了一则招聘信息，职位是Robotaxi客户运营的高级项目经理。该职位的重点是提供卓越的乘客体验。围绕这一点引发了一些讨论：有AI司机分享了一张从网络出租车用户界面上截取的照片，该照片显示了踏板和方向盘的选项。然而，David Moss分享了一张网络出租车方向盘的照片，我认为大多数人都会同意这个设计不太可能进入量产。这只是我的一点铺垫。Holmars说，这只是运行在所有车辆上的标准特斯拉软件。虽然网络出租车没有前备厢，但软件界面上仍显示“打开前备厢”。今天，网络出租车的首席工程师Eric证实了这一点是正确的。因此，至少对于这些车辆来说，网络出租车的车载软件仍然是标准软件，还没有针对网络出租车进行专门的修改。
Elon did give a new interview with Peter Diamandis. The biggest takeaway was that he said Optimus 3 production would begin this summer. And that could certainly be interpreted as a pulling forward of production. But as we know production start has changed numerous times. Early last year, the plan was to start production late last year. Then the plan was to start production late this year. Now it's summer of this year. And of course, I can't close this video without touching on Elon's tweet tariffab project launches in seven days. I'll admit, a part of me really thought that all of that public commentary from Elon was just to get all of these silicon makers to make more chips. But whether that was the plan and it failed or not, either way, the Tesla tariffab is apparently a real thing. I believe this is the biggest undertaking in Tesla's history.
埃隆·马斯克（Elon Musk）最近接受了彼得·戴曼迪斯（Peter Diamandis）的新采访。最大的亮点是他说Optimus 3的生产将于今年夏天开始。这可能意味着生产计划提前了。但我们知道，生产开始的时间已经多次更改。去年年初的计划是去年年底开始生产，然后改为今年年底，现在变成了今年夏天。当然，我不能不提到埃隆的推特，他提到tariffab项目将在七天后启动。说实话，我原以为埃隆所有的公开评论只是为了让所有这些芯片制造商生产更多芯片。但不管这是否是计划中的一部分并且失败了，特斯拉的tariffab显然是真实存在的。我相信这是特斯拉历史上最大的一次挑战。
Financially, the capex for this alone is likely to be somewhere in the neighborhood of 10 to 30 billion dollars. This factory will not be producing chips for at least two and a half to maybe four years. I want to be clear the success of this fab is certainly not guaranteed. Just like for a period of time, we thought Tesla wasn't going to crack the code with dry battery electrode tech. Elon said publicly that it was way harder than we thought and they were going back to the wet method. So it really is wild that we have a truly company altering post like this on a random Saturday just casually from Elon. And truly, the fact that Elon and Tesla believe this is required to not be bottlenecked by chip limitations tells us everything we need to know about the scale Tesla is planning to hit.
从财务角度来看，仅此项目的资本支出可能在100至300亿美元之间。这家工厂至少在两年半到四年内不会生产芯片。我想明确指出，这座芯片制造厂的成功并不是板上钉钉的。就像我们曾经一段时间认为特斯拉无法破解干电池电极技术一样。埃隆公开表示，这比我们想象的要困难得多，他们最终决定回到湿法。因此，在一个普通的星期六，埃隆随意发布如此影响公司的消息，实在让人惊讶。实际上，埃隆和特斯拉认为需要这样做才能突破芯片限制的瓶颈，这充分说明了特斯拉计划达到的规模。
And you guys are smart. You know long term if Tesla can pull this off, it's going to bring with it a long list of benefits. But before we get to that point, there's going to be a long list of costs and likely setbacks and challenges. So it'll be fascinating for our purposes here to get to document and cover all of this stuff that's going on at Tesla. But if you're a Tesla investor that came into the company for electric vehicles and energy storage and maybe a little bit of autonomy, we just all have to accept that that narrative alone has completely been blown up. And obviously it's still a part of the picture. But Optimus, full on autonomy, data centers in space, tariff abs, agentech AI, Tesla is 100% morphing right before our eyes.
你们都很聪明。你们知道，从长远来看，如果特斯拉能成功实现这一目标，将会带来一长串的好处。但在我们到达那个阶段之前，会有一长串的成本，可能还会有挫折和挑战。所以，从我们的角度来看，记录和报道特斯拉正在发生的一切将会是非常有趣的。但是，如果你是因为电动汽车、能源存储或是一些自动驾驶技术而投资特斯拉的投资者，我们都必须接受，仅仅依赖这些的叙述已经完全不切实际了。显然，这些仍然是特斯拉的一部分，但特斯拉正在我们眼前全面转型，包括Optimus、完全自动驾驶、太空数据中心、芯片生产、农业科技AI，特斯拉正在100%地变化。
So listen, I know that people are frustrated with timelines and expectations and what feels like distractions. I think to a certain degree that really is all valid. And if you're in a situation where you needed Tesla stock to be up by now or you need it up soon, truly I feel for you. But if you're somebody that's a long term investor and has another three five seven ten years, I would vehemently make the argument that Tesla's positioning and the addressable markets it's going after is 100% unmatched by any other company on planet earth. There's immense work to be done, but the culture at Tesla, the efficiency at Tesla, the leverage and the foundation and the software, the world models that will be used in hardware.
所以，听我说，我知道大家对时间表、预期以及那些看起来像是干扰的事情感到失望。在某种程度上，这些感受完全是合理的。如果你正处于一个需要特斯拉股票现在上涨或者即将上涨的情况中，我真的很同情你。不过，如果你是一位长期投资者，还能够等待三年、五年、七年甚至十年，我会坚定地认为特斯拉的市场定位和目标市场远远超过地球上任何其他公司。虽然还有大量工作要做，但特斯拉的企业文化、效率、杠杆、基础以及软件与硬件所使用的世界模型，都是无与伦比的。
Also vertically integrated at Tesla, the scaling will come. If you're somebody that actually doubts that Tesla's ability to scale FSD and or Optimus, then I've said many times, respectfully you should not own Tesla stock. So we all have a lot of new things to learn in the weeks ahead if we really want to continually understand Tesla incredibly well. Not everybody loves change, but I am indeed perpetually curious and I have to admit a lot of the time I really enjoy change.
在特斯拉，也是采用垂直整合的方式，规模化是会实现的。如果你对特斯拉扩展全自动驾驶（FSD）或Optimus机器人的能力有疑虑，我已经多次说过，那么恭敬地建议你不应该持有特斯拉的股票。因此，如果我们真的想持续深入了解特斯拉，在接下来的几周里我们还有很多新东西要学习。并不是所有人都喜欢改变，但我确实对任何事物充满好奇，并且不得不承认很多时候我非常享受改变。
But at the end of the day, Tesla is going through a transformation that really is unprecedented for a one and a half trillion dollar company to make changes on the fly like this, taking massive risks in entirely new industries like a silicon fab is just the reality of Elon staying at the forefront, being a couple steps ahead of everyone else and pivoting to do what he believes is required based on what the technology is enabling.
归根结底，特斯拉正在经历一场变革，对于一个市值达到1.5万亿美元的公司来说，能够如此迅速地做出改变实属前所未有。进入硅晶圆厂这样全新的行业并冒巨大风险，正是因为埃隆·马斯克始终保持在前沿，始终比其他人领先几步，并根据技术的发展调整方向，去做他认为必要的事情。
And that right there is the crux of everything. The technology, both from a hardware and software perspective, has progressed so much in just the past six to 12 months that what is possible and what our vision for the future is has pretty dramatically been changed. So you have to ask yourself, would you really prefer that Elon and Tesla sit on the sidelines for all of these new industries that are now possible for the first time in human history?
这就是一切的关键所在。从硬件和软件的角度来看，技术在过去六到十二个月内有了很大的进步，使得我们对未来的设想和可能性发生了显著的变化。因此，你得问问自己，你真的希望马斯克和特斯拉在这些人类历史上首次可能出现的新产业中袖手旁观吗？
Now again, if you're a swing trader or a short term investor, the answer for you is probably yes. But as we know, Tesla plays the long game. It is not concerned with the short term. It is concerned with building the future. And so I bet you can guess the one thing I have questions about is Tesla's tariff app.
如果你是波段交易者或短期投资者，答案很可能是肯定的。但我们都知道，特斯拉注重的是长远发展，它并不关注短期表现，而是专注于打造未来。所以我猜你可能猜到了我唯一有疑问的事情，就是特斯拉的关税应用。
However, if there is a silver lining and I think that there is, it's that Tesla will surely have help from companies like Samsung and TSMC. Tesla won't get free IP, but it's very clearly developed a partnership more than just a customer type relationship with Samsung. And if anybody has the ability to attract the top talent needed and to solve hard problems, we already know it's Elon Musk.
尽管如此，我认为事情还存在一些积极的因素，那就是特斯拉肯定会得到来自三星和台积电这样的公司帮助。特斯拉虽然不会免费获得知识产权，但它显然与三星建立了比普通客户关系更深的合作伙伴关系。而且，如果说有谁能够吸引顶尖人才并解决难题的话，我们已经知道那个人就是埃隆·马斯克。
So we might have to sit tight for another few months, or who knows, maybe for the rest of this year. But I really don't think it's hard to envision this future, where CyberCab is actually starting to scale to multiple cities, where Optimus enters production and starts doing work in Tesla's factory, where the semi is in production and company orders are coming in, where Tesla energy and mega block continue its generational run, where a GENTIG AI in some capacity is available on Tesla vehicles sometime later this year.
所以我们可能还需要再耐心等待几个月，或者，谁知道呢，可能要等到今年结束。但我真的不觉得很难想象这样的未来：在多个城市推广的CyberCab，Optimus投入生产并开始在特斯拉的工厂中工作，电动卡车进入生产阶段并且公司订单源源不断，Tesla Energy和巨型储能装置继续其跨时代的表现，某种形式的GENTIG AI可能在今年晚些时候出现在特斯拉的车辆上。
And I think a lot of you would agree with me that everything I just said has a very realistic chance of happening in the next 12 months. And listen, many of us from a stock perspective have already waited five years, so at this point, what's another 12 months? And as I always say, once those milestones happen and they give Wall Street the validation it needs to build new financial models and start pricing in some of that revenue and profitability, that will be the time Tesla stock finally moves.
我想很多人会同意我所说的，即这些事情在接下来的12个月内有很大可能会发生。还有，从投资角度看，我们中的许多人已经等待了五年，所以再等12个月也不算什么。正如我一直所说的，一旦那些里程碑事件发生，并给华尔街提供了构建新财务模型并开始将部分收入和利润纳入估价的验证，那将是特斯拉股票真正上涨的时候。
Of course, barring the macro that is the lid always. So the state of the union, for me right now, it is truly as strong as ever, arguably with more risk and uncertainty than ever. And I do think that's a big reason why not everybody loves this massive transition. But the truth is, if you're invested in a company like Tesla, you do it for the upside.
当然，从整体的大环境来看，情况始终存在一些限制。目前对我来说，所处的状态依然非常强劲，甚至可以说风险和不确定性比以往任何时候都要多。我认为这也是为什么并不是所有人都喜欢这个大规模转变的一个重要原因。但事实是，如果你投资像特斯拉这样的公司，你就是为了追求其巨大的收益潜力。
You don't invest in a company like Tesla to protect your capital. And so to boil it all down, in my humble opinion, there is no company on planet earth, with more upside over the next 10 years than Tesla. And frankly, I really do not even think it's close. But I hope you all enjoy the rest of your weekend and a huge thank you to all of my Patreon supporters.
你投资像特斯拉这样的公司，并不是为了保值。简单来说，以我个人的观点来看，未来10年内，没有哪家公司能比特斯拉有更大的增长潜力。坦白说，我认为这个差距非常明显。希望大家能享受接下来的周末，感谢我所有的Patreon支持者。