发布会·中文精读

Tesla Q4 与 2025 全年财务业绩及 Q&A 直播

Tesla Q4 and full year 2025 Financial Results and Q&A Webcast

中文精读由 AI 基于原始视频与文稿改写,非官方翻译,可能存在偏差或错误;表述以原视频为准。说明

财报电话会开到一半,马斯克宣布了一件事:Model S 和 Model X 要停产了。

他用的词是「honorable discharge」,光荣退役。弗里蒙特工厂里原本生产这两款车的厂房,接下来要改成年产 100 万台 Optimus 人形机器人的产线。

一款曾经把电动车从边缘爱好变成主流选择的车型,就这样让位给机器人。这件事本身比任何财务数字都更能说明特斯拉现在的位置:车可以不造,方向盘可以不要,算力和机器人必须拿下。

同一场会上,马斯克还把公司使命改了。改完之后,整场会听起来就不像财报了,更像一份施工计划。

从「加速世界向可持续能源转变」到「惊人的富足」

特斯拉原来的使命是加速世界向可持续能源转变,用了很多年。这次马斯克把它换成「amazing abundance」,惊人的富足。他说这是为了向外界传递一种对未来的乐观。

我认为我们最有可能走向一个令人兴奋的、惊人的富足时代。AI 和机器人技术持续增长,我们实际上正走向一个普遍高收入的未来。(马斯克)

他特意区分了两个词。政府发钱兜底的那套叫普遍基本收入(universal basic income),他不认为那是终点。他要的那个词叫普遍高收入(universal high income),东西便宜到不需要兜底。

两个词只差一个英文单词,背后是两套完全不同的世界观。一套是「你失业了,我养你」,另一套是「生产成本趋近于零,你不需要被养」。

马斯克把这条路描述得很具体:持续提升安全性,压低商品成本,让人们不妥协地获得任何需要的东西,同时保住自然和地球的美感。他的原话大意是,这大概是你所能想到的最好未来,包括极好的医疗,也包括还在的森林和海洋。

听着像愿景,但接下来一小时里,他讲的几乎全是实现路径上的具体障碍:产线、芯片、精炼厂、监管、地缘政治。这些东西一条都绕不过去。

Model S 的体面退役,和一条年产百万台的机器人产线

停产的消息说得很干脆。想要 Model S 或 Model X 的人现在还可以下单,预计下个季度逐步收尾,基本停止生产。售后支持会继续,只要路上还有这两款车在跑。

弗里蒙特那块地不会空着。马斯克的说法是,长期目标是在原来 S 和 X 的空间里做到每年 100 万台 Optimus 机器人。

这有点令人伤感,但 S 和 X 项目该结束了,我们要转向自动驾驶的未来。(马斯克)

他提到自己在 X 上的头像已经挂了好几个月,上面写着「the future is autonomous」,未来是自动驾驶的。

这件事的性质是产能重新分配,比产品线优化严重得多。一家公司的厂房、供应链、工程团队都是有限的,把一条成熟车型的产线拆掉换成机器人产线,意味着他认为后者的长期回报远高于前者。这个决定可以错,但错得很贵。

还有个细节值得留意:S 和 X 是特斯拉历史上唯二证明过「高端车也能卖出去」的产品。砍掉它们,等于放弃高端乘用车这条腿,把资源全押在两件还没跑通规模的事上。这种取舍在成熟车企里几乎不会发生。

奥斯汀街头,车里已经没有人了

Robotaxi 的进展是这场会里信息密度最高的部分。

马斯克说,在奥斯汀已经开始做没有安全员的付费载客,乘客是随机选择的。他还提到一个细节:就在会前一两天,车上连跟随车都不需要了,就是一辆空车自己在跑,后面没有人跟着。

他对节奏的描述是:

你会看到自动驾驶的程度每个月都在大幅提升,基本上是每个月。(马斯克)

覆盖范围上,他预计到今年年底,美国大概四分之一到一半的地区会有完全自动驾驶车辆在运营,前提是拿到监管批准。这里有个变量:如果联邦层面出现对自动驾驶的优先立法(federal preemption),推进速度会快得多;没有的话,就只能一个城市一个城市、一个州一个州地谈。

即便只能逐城推进,他的目标是年底进入几十个主要城市。

至于为什么这么慢,马斯克的解释是有些路口本身就危险:

有些路口人类司机就经常出事。洛杉矶 Wilshire 和 Santa Monica 交汇那一段,差不多有 20 个红绿灯,那里天天有人撞车。你得确保 FSD 能处理这些奇怪的路口。(马斯克)

所以策略是每发布一版 FSD,验证过安全,就相应减少一点驾驶员监控的要求。用他的话说,是被偏执驱动的谨慎。这个词他用得很准,因为后面讲芯片和地缘政治时,他又用了一次。

被问到目前有多少辆出租车在载客,他给的数字是湾区和奥斯汀加起来超过 500 辆,并且大概每个月翻一番,处在指数曲线上。

把私家车挂进车队,像 Airbnb 一样

车主把车加入或移出自动驾驶车队这件事,容易被低估。模式类似 Airbnb 的房源上下架:车不用的时候让它自己出去跑单,需要的时候收回来。

马斯克认为市场对它的定价严重不足:

我觉得很多人低估了特斯拉车主把车加入自动驾驶车队这件事的价值,因为我们已经有几百万辆搭载 AI4 硬件的车可以做这件事。(马斯克)

他还给了一个很具体的算术:把车租给车队的收入,可能超过车主付给特斯拉的租赁月供。他的原话大意是,你相当于因为拥有一辆特斯拉而拿到报酬。

这个模式如果跑通,特斯拉的资产负债表会变得很奇怪。它既是卖车的公司,又是运营车队的平台,还从自己卖出去的车里分成。车主变成了微型运营商,车变成了能产生现金流的资产。

对车主来说,这意味着买车的决策逻辑变了。以前算的是折旧和用车成本,以后还要算一笔潜在收入。当一辆车的资产属性强过消费品属性,它对销量的拉动会非常直接。

当然前提是监管允许,以及车队调度真的能跑起来。这就引出了后面关于 Grok 的讨论。

CyberCab 没有方向盘,产量要超过其他所有车型之和

CyberCab 是一款两座专用自动驾驶车,没有方向盘,也没有踏板。

马斯克特意强调了这个设计的含义:

这里没有备用机制。这辆车要么自己开,要么就不开。(马斯克)

预计 4 月开始生产。产能爬坡仍然是标准的 S 曲线,先慢,然后指数上升,进入线性,最后逼近目标产能的平台期。

关键的论据来自工程高级副总裁 Lars Moravy 提到的数据:大约 90% 的行驶里程是一到两名乘客,其中大约 80% 只有一个人。所以两座车覆盖了绝大多数真实使用场景,把它当成缩水版来看就完全错了。

马斯克的推论很直接:长期看,CyberCab 的年产量会是特斯拉其他所有车型加起来的好几倍。

另一个被反复提到的概念是使用率。普通私家车一天平均开一个半小时,一周大概 10 到 11 小时,而一周有 168 小时。自动驾驶车的使用率预计是普通车的五倍左右,每周跑 50 到 60 小时。

这个差别会改变整车的设计逻辑:

你得按单位时间内更大的磨损来设计这辆车,要有更强的耐久性。它更像一辆几乎不停运转的商业卡车。自动驾驶车就该这么设计。(马斯克)

他还顺带说了 Cybertruck 的走向:整条线会转向完全自动驾驶,城市内几百英里范围的本地货运是个明确的市场。

以及一个彩蛋:下一代 Roadster 计划 4 月亮相,马斯克的评价是「something out of this world」。他还说,长期来看,除了 Roadster,特斯拉造的每一辆车都会是自动驾驶车。

顺便说一句命名上的麻烦。他承认 robo taxi 和 CyberCab 这两个词容易混,而且在有些州法律不允许使用 cab 或 taxi 这样的词,所以以后可能得叫 cyber vehicle 或者 cyber car。一个产品名字被监管逼着改,说明这类车在法律上还处在真空地带。

未来只有 1% 的里程由人自己开

被问到人类驾驶会不会消失,马斯克给了一个数字:

未来绝大多数的行驶里程会是自动驾驶。我猜可能不到 5% 的里程会由人自己开,也许低到 1%。(马斯克)

这个预测没法验证,但它解释了前面所有决策的逻辑。如果人类驾驶真的萎缩到个位数百分比,那为人类驾驶优化的车、为人类驾驶设计的产线、为人类驾驶建立的成本结构,长期看都是错的资产。

反过来,如果这个预测错了,特斯拉付出的代价是巨大的:S 和 X 停产、产线改造、CyberCab 这种没有方向盘的车上路。这是一次方向性下注,属于把整张桌子推上去的打法。

判断这类预测有个简单办法:看说话的人有没有为此改变自己的资产负债表。马斯克砍掉了两款还在赚钱的车,这是真金白银的投票。

Optimus 3 几个月后亮相,难的是造出一百万台

马斯克说 Optimus 3 大概几个月后发布,而且会让人吃惊。他用了「incredibly capable」来形容。

关于产能,他讲了一段很实在的工程逻辑。Optimus 是一条完全新的供应链,现有的汽车供应链里几乎没有能直接拿来用的东西,一切都是从物理第一性原理重新设计。

结果是 S 曲线会被拉长:

当一切都是新的,生产速率就取决于整条供应链里最不走运、最没底气的那个环节。如果有一万件事必须做对,只要有一件慢了,整条线就慢下来。(马斯克)

这段话值得多看两眼。它说的是一种典型的量产困境:新产品最怕的环节往往最不起眼,可能是一个密封圈,可能是一种特殊润滑剂,可能是一家只有几十人的小供应商。所有环节都新,意味着这类风险点成百上千地叠加。

他把 Optimus 定位成通用机器人,能通过观察学习:你可以演示一个任务,可以用语言描述一个任务,可以给它看一段视频,它就能做。

至于现在的实际使用情况,他说得很坦白:还处在研发阶段,工厂里让机器人做过一些基础任务,但每迭代一版新设计就淘汰旧版本,所以还没有实质性地投入使用。真正有意义的产量,大概要到今年年底。

他对长期影响的表述很夸张,但也给了具体的锚点:

长期看,Optimus 会对美国 GDP 产生非常显著的影响,会真正推动美国 GDP 变化。(马斯克)

这个说法听着大,但拆开也不难算。美国劳动力缺口集中在制造、物流、护理、建筑这些需要物理动作的行业,人形机器人如果能覆盖其中一部分,对产出的影响是以万亿计的。

顺带一条和行业气氛相反的信息:市面上很多科技公司在宣布裁员,但马斯克说弗里蒙特工厂反而预计会随时间增加人头,因为产出要大幅提升,没有裁员计划。

人形机器人最难的三件事,特斯拉说三件都在自己手里

马斯克把这个问题拆得很清楚。人形机器人有三个真正的硬骨头:

  • 造出一只和人类手部自由度、灵巧度相当的机械手,这是极难的工程挑战
  • 真实世界里的 AI,也就是机器人在物理世界里理解和行动的能力
  • 规模化量产

他的判断是,特斯拉是唯一同时具备这三块能力的公司。

手排在第一位不是随口说的。人手的自由度、触觉反馈、精细力控,是几十年生物进化的产物,机器人领域在这上面卡了很久。马斯克把 hand design 单独拎出来说它是机器人里最难的部分,说明特斯拉在这上面的投入权重很高。

这一点的产业含义很大。做机器人的公司很多,做灵巧手的很少,能把灵巧手做到量产良率可接受的更少。如果你的供应链里有一环是别人做不了的,你就在定价上有话语权。

马斯克现在每个周六都在画芯片

被问到最近把时间花在哪,马斯克的回答很具体:AI5 芯片。

完成 AI5 芯片设计并让它成为一颗好芯片,可以说是当前最重要的一件事。这也是为什么我现在在特斯拉花的时间比在任何其他事情上都多。基本上每个周六我都在这上面,每个周二也有一部分时间。(马斯克)

他补了一句判断标准:如果他在周六做某件事,那这件事一定足够重要。

他对 AI5 的设计表示有信心,并且说 AI6 会在这之后跟进,理想情况下间隔不到一年,会是相对 AI5 的又一次大跨越。

关于是否对外卖芯片,答案是目前不卖。逻辑很实在:先要保证自己车和 Optimus 的用量。特斯拉的数据中心里已经在用 AI4 芯片,训练时是 AI4 加英伟达硬件的组合。

真正的问题在三四年后。马斯克说,如果往前看三四年,特斯拉增长的限制因素很可能就是芯片产能,包括 AI 逻辑芯片、内存和 RAM 的供应量。这意味着不仅不能外卖,还得自己想办法扩产。

这里有个反直觉的地方。一家车企的 CEO 说自己的增长瓶颈是内存,听起来很不搭。但如果未来产品的主体是机器人和自动驾驶车,那本质就变了:这些产品卖的是算力和智能,钢铁和橡胶只是外壳。

TerraFab:美国本土连一座先进存储厂都没有

这是整场会里最重的一段,也是马斯克自己承认会引起争议的一段。

他的结论是特斯拉需要自建一座 TerraFab,一座包含逻辑、存储和封装的大型晶圆厂,而且要在美国本土。

即便把我们所有关键供应商的最佳产能都算上,包括三星、台积电、美光这些战略伙伴,我们问他们最多能产多少,答案是不够。为了解除三四年后这个大概率的限制,我们可能必须建一座特斯拉 TerraFab。(马斯克)

他给了一个很多人不知道的事实:美国目前没有任何一座规模化运营的先进存储晶圆厂,字面意义上是零。台积电在亚利桑那有厂,三星在德州有厂,先进存储这块是空白。他顺带开了个玩笑,希望美光几年后能有点产出,毕竟他们总部在爱达荷州,那里盛产土豆。

地缘政治是他反复提的动机:

很多人可能低估了几年后地缘政治风险的分量。很多公司对这件事是睡着的,或者把头埋进沙子里希望坏事别发生。我比那偏执得多。(马斯克)

他还回应了「建晶圆厂太难」的质疑:

很多人会说这太疯狂了,晶圆厂非常难。是的,我知道晶圆厂很难,我没说它容易。但我们以前也没有电池工厂、锂精炼厂、Megapack 工厂,这些都是我们自己搞出来的。(马斯克)

有一句话把这件事的严重性讲透了:Optimus 没有 AI 芯片就是一个假人模特,比《绿野仙踪》里的铁皮人还惨,铁皮人至少还能走,Optimus 连坐着都做不到。而汽车至少还能装回方向盘和踏板,或者做后装改造。

所以芯片对特斯拉是存亡问题,成本问题排在后面。目前他们对未来三年左右的规模有解决方案,三年之后会受制于供应商,必须提前想好对策。他说得更重的一句是:这事关乎特斯拉的存续,不是省钱。

每 GB 能装多少智能

马斯克提了一个他自己很看重的指标:intelligence per gigabyte,每 GB 的智能。

在内存受限的场景下,这个指标决定了一切。同样多的内存,你的模型能输出多少功能,能有多聪明。

我认为特斯拉在 AI 的智能密度上领先世界其他地方一个数量级甚至更多。这话听起来很狂,但我大致知道 Grok 和一些其他大模型的智能效率是什么水平。在内存效率上,特斯拉的 AI 要好一个数量级以上。(马斯克)

这句话的价值不在真假,在于它解释了特斯拉为什么敢自己设计芯片。车端和机器人端的算力、功耗、内存都是硬约束,云端那种堆卡的思路搬不过来。同样的内存里塞进更多能力,是唯一可行的路。

而如果这个密度优势是真的,那自研芯片加上高密度模型,会让特斯拉在成本曲线上和对手拉开距离。这也解释了为什么 AI5 值得他每个周六亲自盯。

Grok 是机器人乐团的指挥

特斯拉投资了 xAI,马斯克给的理由很朴素:很多股东要求这么做,那就做了。

但他接着讲了 Grok 在特斯拉业务里的实际位置,这段比投资本身更有信息量。

第一个场景是车队调度。未来如果有一千万辆甚至几千万辆自动驾驶车,怎么高效调度这些车,Grok 会远好于任何启发式算法或者人工管理方案。

第二个场景更远:用 Optimus 机器人去建工厂、建精炼厂。他举了个假设的例子,比如稀土矿精炼厂,美国非常缺这种产能。那么问题来了,谁来组织这一千台 Optimus 干活?

你需要一个乐团指挥。Grok 就是那个指挥,指挥 Optimus 机器人去建那座精炼厂。(马斯克)

这个比喻值得停一下。它换掉了主语:AI 的位置从替代某个具体岗位,变成整个生产系统的调度层。人形机器人负责物理动作,大模型负责拆解任务、分配工作、处理异常。如果这条路走通,工厂的组织形态会变。

他还补了个限定,说稀土精炼厂这个例子目前不在公司计划里。但后面那句感叹是真的:美国确实需要多得多的矿石精炼产能。

最大的对手在中国

被问到人形机器人的竞争格局,马斯克的回答没有任何外交辞令:

人形机器人最大的竞争毫无疑问来自中国。中国非常擅长规模化制造,AI 也相当强,你可以看到他们免费发布的那些模型质量相当好,而且在不断变好。(马斯克)

他还说了一句提醒:

我一直觉得中国以外的人有点低估中国。中国是另一个层级。(马斯克)

据他所知,中国以外没有明显的竞争对手。但中国会是硬仗,没有第二种可能。

至于特斯拉凭什么觉得自己能赢,他给的理由还是那三条:真实世界智能、机电灵巧度,尤其是手的设计,以及量产能力。他认为 Optimus 会比他在中国看到的任何在研机器人都更强。

这段表态的诚实程度在财报会上并不多见。承认对手强,通常意味着内部已经把对手当作真实威胁来评估,而不是公关层面的敷衍。

全美唯一的锂精炼厂,和一句「求求你们也建一个」

聊到供应链,马斯克有一段情绪很外露的发言。

他说特斯拉在 Corpus Christi 建的锂精炼厂没有照抄别人的方案,走的是一套全新工艺,效率和先进程度都超过世界上现有的任何同类设施。奥斯汀的正极材料厂也一样。

然后他连问了几遍:为什么这些事只能我们来做?

其他人能不能,看在上帝的份上,也来建这些厂?我们是被逼出来的。我们基本上是全美最大的,也是唯一的锂精炼厂和正极材料厂。(马斯克)

这段话听起来像抱怨,其实是判断。一家做车的公司被逼着自己去建精炼厂,说明上游的资本投入长期不足。而如果自动驾驶和机器人真的把电池需求推上去,这个短板会变成整个行业的瓶颈。

他最后落回到那句老话上:

我总是想起安迪·格鲁夫那句名言,只有偏执狂才能生存。他为什么在英特尔说出这句话?我想这里面有很深的智慧。我们会保持偏执,确保无论发生什么,我们都能继续造电池、造机器人、造 AI 芯片。不这么做的公司,一批会死掉。(马斯克)

这句引用放在这里挺有意思。格鲁夫说这话的背景是英特尔的存储业务被日本厂商打穿,被迫全面转型做处理器。那是一次被逼出来的生死转向。马斯克引用它,大概是在给自己现在的选择找坐标。

这场会真正在说什么

把整场会拆开看,特斯拉在做四件事,而且每一件都在同时进行:

  • 把车从「人开的机器」改造成「自己跑的运力」,代价是砍掉 S 和 X 这样的经典产品线
  • 把机器人从实验室推到量产,代价是一条完全从零搭建的供应链和一条被拉长的 S 曲线
  • 把算力从「买」改成「自己造」,代价是可能要建一座美国本土的大型晶圆厂
  • 把上游材料从「买」改成「自己炼」,代价是干别人不愿意干的脏活

这四件事的共同点是:都需要巨额资本开支,都要几年才见效,都在赌一个人类驾驶和人工劳动被重新定价的未来。马斯克自己说今年会是一个资本开支大年,而且是刻意的。

这场会最有价值的地方,是他把每一个赌注的失败模式也讲了出来:Optimus 没有芯片就是假人模特,美国没有先进存储厂,逐城推进会被监管拖慢,S 曲线取决于最慢的那个供应商。

一个愿意公开讲清楚自己会被什么卡死的人,通常比只讲愿景的人靠谱一些。

给从业者的几条可操作建议

  • 如果你在做自动驾驶或机器人相关的事,把「手」单独拿出来看。马斯克把它列为人形机器人第一难题,说明这块的工程门槛和时间成本被普遍低估,谁在这里有积累,谁就有议价权。
  • 关注「每 GB 的智能」这类效率指标,不要只看模型跑分。车端、机器人端、手机端都是内存和功耗受限的场景,密度决定能不能用,跑分决定不了。
  • 如果你是硬件或供应链从业者,把锂精炼、正极材料、先进存储这三块列进观察名单。前两块是全美只有一家在做,第三块是全美一家都没有,这种缺口不会长期存在。
  • 如果你是投资者,盯住三个可验证的信号:CyberCab 4 月是否真的投产、年底 robotaxi 覆盖的城市数量、Optimus 年底前是否有实际产量。这三个数字比任何发布会上的形容词都有用。
  • 如果你只是普通车主,可以算一个具体问题:你的车一周有多少小时在动。10 小时和 168 小时的差距,就是自动驾驶车队模式存在的全部理由。
  • 面对任何「十年后只剩 1% 人类驾驶」这类预测,先看说话的人有没有为此调整自己的产线和资本开支,再决定信不信。特斯拉砍掉 S 和 X,就是一次公开的投票。
第三方整理文稿,仅保留马斯克本人发言。已按语句合并成段落,并去掉 um、uh 等口头填充词。 文稿来源:The Motley Fool (transcription)。

Thanks, Travis. So I have updated the Tesla, Inc. mission to amazing abundance, and this is intended to send a message of optimism about the future. I think we are most likely headed to an exciting, amazing era of abundance. I think with the advent or continued growth of AI and robotics, I think we actually are headed to a future of universal income, not universal basic income, but universal high income. There is going to be a lot of change along the way, but that is what I see as the most likely outcome. So I think it makes sense to update Tesla, Inc.'s mission to reflect that goal.

Obviously, along that way, we are going to keep improving safety, driving down the cost of goods, and getting people access to anything they need without compromise. Making sure that the environment is great, nature is great, and people can have whatever they want, which seems like probably the best future. Open to other ideas, but that sounds like the best future you could possibly imagine. I guess it would be that everyone can have whatever they want, including amazing medical care, but we still keep the beauty of nature and Earth. I think that's probably the best outcome. We are seeing the first steps along that way this year for Tesla, Inc., post major steps.

As we increase vehicle autonomy and begin to produce Optimus robots at scale, we are making very big investments. This is going to be a very big CapEx year, as we will get into. That is deliberate because we are making big investments for an epic future. I think all these investments make a lot of sense. We will continue to make sure that when we do spend capital, it is spent very efficiently. But it's a lot of things. Major investments in batteries and the entire supply chain of batteries. We are also going to be significant manufacturers of solar cells, and we are making massive investments in AI chips. I think these all make a ton of strategic sense.

I guess I have one, not exactly bad news, but it's time to bring the Model S and X programs to an end with an honorable discharge. We are really moving into a future that is based on autonomy. If you are interested in buying a Model S and X, now would be the time to order it because we expect to wind down S and X production next quarter and basically stop production of Model S and X next quarter. We will obviously continue to support the Model S and X programs for as long as people have the vehicles.

We are going to take the Model S and X production space in our Fremont factory and convert that into an Optimus factory with a long-term goal of having a million units a year of Optimus robots in the current S and X space in Fremont. That is slightly sad, but it is time to bring the S and X programs to an end and shift to an autonomous future. As my profile picture on X for a few months there, the future is autonomous. With respect to self-driving and robotaxi, people are obviously following with very close attention the progress of FSD, and you can experience it for yourself. If you have a Tesla, you notice with every software update, the car gets better and better at autonomy.

We were able to do our first rides with no safety monitor in the car in Austin. These are paid rides, randomly selected paid rides with no safety monitor. I think maybe as of yesterday or so, we actually do not even have a chase car or anything like that. These are just cars with no people in them and no one following the car in Austin. We are obviously being very cautious about this because we want to have no injuries or serious accidents along the way. I think it makes sense to be very cautious. You will see the amount of autonomy increase dramatically, I think, every month, essentially.

There will also be an opportunity, something we have talked about for a long time, for existing owners of Teslas to add or subtract their cars to the fleet, kind of like how Airbnb works where you can add or subtract your house to the Airbnb inventory. I think probably the value of the Tesla, Inc. partial people adding or subtracting the cars to Tesla, Inc.'s autonomous fleet is probably a little underweighted by a lot of people because we have millions of cars with AI4 that can do this. It might potentially provide an opportunity for a lot of customers to earn more lending their car to the fleet than their lease cost to Tesla, Inc. It's kind of like you get paid to own a Tesla, Inc.

It's quite a good scenario. We expect to have fully autonomous vehicles in probably somewhere between a quarter and half of the United States by the end of the year, pending regulatory approval. A big factor would be if there's some kind of federal preemption for autonomous vehicles. In the absence of that, you have to go on a city-by-city or state-by-state basis. Nonetheless, even if it is city-by-city, state-by-state, we expect to be in dozens of major cities by the end of the year. With respect to energy, the Tesla, Inc. energy team has done incredible work. The growth rate on that work is continuing to be very strong.

We are building more manufacturing capacity and expect that energy will have very high growth for as far into the future as we can imagine. The solar opportunity is underestimated. We think the best way to add significant capability to the grid is solar and batteries on Earth and solar in space. That's why we are going to work towards getting 100 gigawatts a year of solar cell production integrating across the entire supply chain from raw materials all the way to finished solar panels. Maybe a bit more about Optimus. We will probably unveil Optimus 3 in a few months. I think it's going to be quite surprising to people. It's an incredibly capable robot.

As I mentioned, we are replacing the S and X line in Fremont with a million unit per year line of Optimus. Because it is a completely new supply chain, there's really nothing from the existing supply chains that exist in Optimus. Everything is designed from physics first principles. That means the normal S curve of manufacturing ramp will be longer for Optimus than it is for products that have at least some portion of an existing supply chain. When everything's new, the production rate will be proportionate to the least lucky, least confident part of the entire supply chain. If there's 10,000 things that need to go right, it only takes one to be slow to lag that.

It will be a stretched-out S curve. I'm confident that we'll get to a million units a year in Fremont of Optimus 3. This Optimus really will be a general-purpose robot that can learn by observing human behavior. You can demonstrate a task or verbally describe a task or show it a task. Even show it a video, it will be able to do that task. It's going to be a very capable robot. I think long-term Optimus will have a very significant impact on the US GDP. It will actually move the needle on US GDP significantly. In conclusion, there are still many who doubt our ambitions for creating amazing abundance.

We are confident it can be done, and we are making the right moves technologically to ensure that it does. Tesla, Inc. has never been a company to shy away from solving the hardest problems. I think that's how you build value in a company, you solve hard problems. I don't know how you create value by solving easy problems. There are a lot of hard problems that the Tesla, Inc. team is going to solve, but it's an incredibly talented, hardworking team. I'd like to thank everyone at Tesla, Inc. for their incredible hard work. It's an honor to work with such a talented group. Thank you to everyone who is supporting this mission. The future is more exciting than you can imagine.

Yeah. Thanks, Travis. As Elon said, the future is autonomous.

Just to add to what Lars said there. The point that Lars made, which is that 90% of miles driven are with one or two passengers or one or two occupants, essentially, is a very important one. Because that implies that the CyberCab, which is a dedicated two-seater dedicated robo taxi, is a little confusing with the terms robo taxi and CyberCab. Sorry about the confusion. In fact, in some states, we're not allowed to use the word cab or taxi, so it's going to get even more strange. It's going to be like cyber vehicle or something cyber car. But the CyberCab, which is a specific model that we're making, does not have a steering wheel or pedals.

So this is clearly, there's no fallback mechanism here. It's like this car either drives itself or it does not drive. We expect to start production in April. As always, it's just an S curve of the production rate is an S curve, so it starts off very slowly and then grows exponentially, then you hit the linear, and then ultimately it asymptotes at whatever your target volume is. We would expect over time to make far more CyberCabs than all of our other vehicles combined. Given that 90% of distance driven or distance being distance traveled exactly, no longer driving, is one or two people. I think it's like 80% is just one.

It would mean that long-term CyberCab would make several times more CyberCabs per year than all of our other vehicles combined.

Yeah. The vast majority of miles traveled will be autonomous in the future. I would say probably less than I'm just guessing, but probably less than 5% of miles driven will be where somebody's actually driving the car themselves in the future, maybe as low as 1%.

Yeah. Like, the CyberCab, the whole design of CyberCab was to optimize the fully considered cost per mile of autonomous driving. It's a different design problem than if you're trying to design cars for people who will be driving versus being driven. So CyberCab is super optimized for minimum cost per mile and also for a much higher duty cycle. You would expect CyberCab to be used probably fifty or sixty hours a week instead of the ten or eleven hours a week that a driven vehicle is used. Typically, people might drive their car for an hour and a half a day on average, so it's like ten hours a week out of a hundred and sixty-eight.

But I think an autonomous vehicle is likely to be used probably five times as often. Which means that you need to design the vehicle for much more wear and tear per unit time and much more resilience. It's more like a commercial truck that's in continuous operation or close to continuous operation. That is how you design an autonomous vehicle.

I really think long-term, the only vehicles that we'll make will be autonomous vehicles with the exception of the next generation Roadster which we're hoping to debut in April.

It's gonna be something out of this world. Fantastic.

But yeah. We will transition the Cybertruck line to just a fully autonomous line. There's obviously a market there for cargo delivery, like you say, localized cargo delivery within a city, within a few hundred miles, something like that. There's a lot of cargo that needs to move locally within a city, and an autonomous Cybertruck could be very useful for that.

Well, we are still very much at the early stages of Optimus. It's still in the R&D phase. We have had Optimus do some basic tasks in the factory. But as we iterate on new versions of Optimus, we deprecate the old versions. It's not in usage in our factories in a material way. It's more so that the robot can learn. We wouldn't expect to have any kind of significant Optimus production volume until probably the end of this year.

Yeah. It's an awesome robot that minimizes any differences, but it looks like a human. People could be easily confused that it's a human. This helps our strategy for the AI too because you can learn how humans do these tasks, and it's very easy to teach the robot in the same way. As opposed to previous robots.

I guess one thing I should say, there's a lot of news of various companies announcing layoffs and whatnot. But at our Tesla, Inc. factory in Fremont, we actually expect to increase headcount over time. To significantly increase output from our factories. We don't have any layoff plans. We expect to actually increase headcount.

Well, it is 100% unsupervised. FSD is 100% unsupervised. We obviously have cars operating with no one in them and no safety monitor and no follow car or anything like that in Austin right now. For customers, we're just being very cautious with the rollout. With each successive version as we prove it out and we make sure that there are no unique issues in particular cities. Sometimes you get some very difficult intersections. It will be an intersection where a lot of humans have accidents, by the way. There are some pretty nutty intersections where a lot of humans make mistakes and have accidents in various cities. We want to make sure that FSD can handle those unusual intersections.

If you take LA, for example, where Wilshire and Santa Monica combined is like there's about 20 traffic lights. People are constantly having accidents there. You want to make sure that FSD can handle things in a particular city. We're actually just being paranoid about safety. With each successful release of FSD, we will reduce the amount of driver monitoring that's needed proportionate to the safety of the FSD build.

Yeah. In terms of taxi vehicles carrying paid customers, I think we're well over 500 at this point between the Bay Area and Austin.

This will probably double every month type of thing. It's on an exponential curve.

Well, I tend to spend time on whatever the most critical issue is for the company. Completing the AI5 chip design and having it be a great chip is arguably the number one most critical thing to get done, which is why I'm spending more time on that than currently anything else at Tesla, Inc. I spend pretty much every Saturday on this, and a chunk of every Tuesday. If I'm spending my Saturdays on something, it's going to be something pretty important. I do think AI5 will be a very good chip. I feel quite confident about the design at this point. AI6, which will follow that, aspirationally would follow that in under a year. It will be yet another big leap beyond AI5.

I feel pretty good about our chip strategy right now. In terms of selling it outside of Tesla, Inc., we first need to make sure we have enough chips for our vehicle production, all of our Optimus production. We will actually use the AI5 chips in our data centers. We already use the AI4 chips in our data centers. When we do training, it's a combination of the AI4 chips and NVIDIA hardware. Primarily that we do training with. By the end of the decade, things are changing so fast that it's hard to imagine what happens at the end of the decade. When I look ahead at what's the limiting factor for Tesla, Inc. growth if you go three or four years out, I think it actually is chip production.

Is there enough AI logic and enough memory, enough RAM for volume? Right now, I see that as being the thing that probably limits our growth in three or four years. Which implies that we're not selling chips outside of Tesla, Inc. because we need them. In fact, I think it's going to make sense. This is definitely going to be a controversial thing, but I think Tesla, Inc. needs to build a TerraFab. I mentioned this at the shareholder meeting. Even when you look at the output of the best-case output of all of our key suppliers, and I'd say even they're beyond suppliers are like strategic partners like Samsung, TSMC, and Micron. We say, what's the most you could possibly make?

Then it's not enough. I think in order to remove the constraint, the probable constraint in three or four years, we're going to have to build a Tesla, Inc. TerraFab, a very big fab that includes logic, memory, and packaging. Domestically. That's actually also going to be very important to ensure that we are protected against any geopolitical risks. I think people may be underweighting some of the geopolitical risks that are going to be a major factor in a few years. A lot of people will say that's crazy. Fabs are really hard. Yes, I know fabs are really hard. I don't think they're easy.

Or we didn't used to have battery cell factories or lithium refineries or MegaPack factories or all these other things. We figured it out. I think if we don't do the Tesla, Inc. TerraFab, we're going to be limited by supplier output of chips. I think maybe memory is an even bigger limiter than AI logic. For example, we have chip supply deals with TSMC in Arizona and Samsung in Texas, but currently, there are no advanced memory fabs at scale in the United States. There are zero, literally zero. Hopefully, Micron will have something going in a few years because they are headquartered in Idaho, where they make a lot of potato chips. We don't really need to make computer chips too.

We're with our strategic partners on the chip front, memory and logic. I think we gotta also try our hand at building a large-scale fab that integrates logic memory and packaging. If we don't do that, we're just going to be fundamentally limited by the supply chain. Especially if there's some worst-case geopolitical situation, it would be quite a severe situation. I think quite frankly, it'd be crazy not to try out the TerraFab.

We'll have a bigger announcement on this in the future.

And then for training. Training. Yeah.

We just had a lot of investors ask us to do this. There was a lot of investor but Tesla, Inc. shareholders said we should invest in XAI. So we're just doing what shareholders would ask us to do pretty much. But Grok will be, I think, very helpful in maximizing the efficiency of the management of a large autonomous fleet. If you've got an autonomous fleet that's in the future, 10 million vehicles or tens of millions of vehicles, then optimizing the efficient use of that fleet Grok will be way better than any heuristic solution. Or manually managed solution.

If you set up managing a large team of Optimus robots to build a factory or build a refinery, and say a rare hypothetical like this is a hypothetical example. A rare earth ore refinery. Which we do desperately need in America. Then you'd say, well, what's going to organize the Optimus robots to build that ore refinery? You kind of need an orchestra conductor. So then Grok would be kind of the orchestra conductor. For the Optimus robots to build the hypothetically, an it might not be hypothetical in the future. I'm just saying it's not currently on our plans. But we do need a lot more ore refining capacity in the US. So then what's going to manage, let's say, a thousand Optimus robots?

Well, at Tesla, Inc., the Tesla, Inc. AI is very compute efficient and very memory efficient. One of the metrics I want you to consider for any given AI model is the intelligence per gigabyte. Especially when you're constrained on RAM. Having an AI that has very high intelligence density per gigabyte, you can say, for a given number of gigabytes, how much functionality can you get out of it? I actually think Tesla, Inc. is ahead of the rest of the world in intelligence density of AI by an order of magnitude or more. This is going to sound like a pretty bold statement, but I kind of know what the intelligence efficiency of the big models are like Grok.

The honest and a bunch of the other models. Tesla, Inc.'s AI is, in terms of memory efficiency, I think, more than an order of magnitude better. So that puts us in a pretty good position for scaling. We do think that there's we do have a solution for logic and memory for, let's say, the next roughly three years. But if you start going beyond three years, and we look at the scaling plans, and how many fabs are getting built, especially if you factor in geopolitical uncertainty, there's always risk that maybe those chips don't arrive that people were expecting to arrive. That's why I think we need to have more fab capacity in the US. Just in case. Chips don't stop arriving for any reason.

This is really existential for Tesla, Inc. because Optimus is completely useless without an AI chip. It's not like at least the cars we can put steering wheels and pedals in. Or retrofit them if need be. But Optimus is just a mannequin without, it's like the Tin Man or whatever The Wizard of Oz. But even worse, at least the Tin Man could walk. Optimus won't even be able to just sit there without an AI chip. We've got a good solution for significant scale through for the next roughly three years. Beyond that, we will be supplier limited and so we've got to figure out some game plan to not be supplier limited.

Well, I do think that by far, the biggest competition for humanoid robots will be from China. China is incredibly good at scaling manufacturing. Actually quite good at AI as you can see from the open-source, not the open-source, but the sort of I guess, some of them are open, actually. But, basically, the models that China's distributing for free are actually quite good and they keep getting better. China is very good at AI, very good at manufacturing, and will definitely be the toughest competition for Tesla, Inc. To the best of our knowledge, we don't see any significant competitors outside of China. But China will definitely be tough competition. There's no two ways about it.

I always think people sort of outside of China kind of underestimate China. China's next level. We think Optimus will be much more capable than any robot that we are aware of under development in China. We think we'll be ahead in terms of the real-world intelligence, the electromechanical dexterity, especially the hand design, which is by far the hardest thing in the robot. In fact, I'd say there's really three hard things about humanoid robots. Building an incredible hand that has the same degrees of freedom and dexterity as a human hand is an incredibly difficult engineering challenge. Then there's the real-world AI and scaling production.

Those are the three hardest problems by far for humanoid robots. I think Tesla, Inc. is the only company that actually has all three of those components.

Well, really, all we're trying to do is make sure that we can scale to a very high volume with autonomous vehicles, with humanoid robots. That we address geopolitical risk. There are so many companies out there that are asleep with the switch with regard to geopolitical risk. Or they just have their head in the sand and hope nothing bad will happen. I'm way more paranoid than that. Always think of Andy Grove's famous statement, only the paranoid survive. Why did he come up with that statement at Intel? Let's think. I think there's a lot of wisdom in that statement. We're going to be paranoid. Make sure that we can continue to build batteries and robots and AI chips no matter what happens.

Companies that don't do that, a bunch of them will cease to exist.

Yeah. We're building the most advanced lithium refinery in the world, by the way. It's not just like, our lithium refinery in Corpus Christi is not just a copy of what others have done. It's an entirely new process that is fundamentally more efficient and more advanced than anything else in the world. The same is true of our cathode refinery here in Austin. We wish others would build this. Can other people please, for the love of God, help? In the name of all that is holy, can others please build this stuff? It's not the first time you host. Exactly.

Why do we have to build these things? Why can others not also please build these things? It's very hard to build these things. We build them out of desperation. Not because nobody else is building lithium refineries and cathode refineries. We're pretty much the not just the largest, but also the only lithium refinery and cathode refinery in America. We're making moves to make sure that no matter what happens, Tesla, Inc. will prosper.

从现场发言里挑出的 10 句原话,附中文译文(AI 翻译,以英文原话为准)。点时间戳可跳到视频对应位置。

我认为,随着人工智能和机器人技术的出现或持续增长,我们实际上正走向一个全民收入的时代,不是全民基本收入,而是全民高收入。

I think with the advent or continued growth of AI and robotics, I think we actually are headed to a future of universal income, not universal basic income, but universal high income.

未来绝大多数行驶里程都将是自动驾驶。

The vast majority of miles traveled will be autonomous in the future.

你可以预期 CyberCab 每周大概使用五六十个小时,而不是有人驾驶车辆每周使用十到十一个小时。

You would expect CyberCab to be used probably fifty or sixty hours a week instead of the ten or eleven hours a week that a driven vehicle is used.

我真的认为,长期来看,我们唯一会制造的车辆将是自动驾驶汽车,例外是下一代 Roadster,我们希望它在四月首次亮相。

I really think long-term, the only vehicles that we'll make will be autonomous vehicles with the exception of the next generation Roadster which we're hoping to debut in April.

我们预计在今年年底之前,Optimus 不会有任何可观的产量。

We wouldn't expect to have any kind of significant Optimus production volume until probably the end of this year.

我们没有任何裁员计划。我们实际上预计会增加员工人数。

We don't have any layoff plans. We expect to actually increase headcount.

当我展望三四年后 Tesla, Inc. 增长的限制因素时,我认为实际上就是芯片生产。

When I look ahead at what's the limiting factor for Tesla, Inc. growth if you go three or four years out, I think it actually is chip production.

我认为内存可能是比 AI 逻辑更大的限制因素。

I think maybe memory is an even bigger limiter than AI logic.

我实际上认为,Tesla, Inc. 在 AI 的智能密度上领先世界其他地方一个数量级甚至更多。

I actually think Tesla, Inc. is ahead of the rest of the world in intelligence density of AI by an order of magnitude or more.

我确实认为,人形机器人领域最大的竞争将来自中国,而且遥遥领先。

I do think that by far, the biggest competition for humanoid robots will be from China.

资源与出处

相关视频

同频道、同主题或相邻年份