发布会·中文精读

Tesla Q1 2024 财报与问答网络直播

Tesla Q1 2024 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

马斯克在财报电话会快结束时说了一句话,语气很硬:

如果你不认为特斯拉能解决自动驾驶,那你就不该投资这家公司。(马斯克)

这句话放在 2024 年 Q1 的背景下,很难不被当成嘴硬。当季营收承压,全球电动车渗透率放缓,一堆传统车厂转头去做插电混动,特斯拉自己也在 Fremont 工厂爬坡新款 Model 3。电话会开场的定调是「我们穿越了几个意料之外的挑战」。

但如果你把整场问答听完,会发现马斯克从头到尾只在讲一件事:特斯拉不是一家汽车公司,它的估值框架应该按 AI 和机器人来算。他甚至搬出了凯西·伍德(Cathie Wood)的话来背书。剩下所有内容,无论是新车、储能、FSD 还是 Optimus,都是这个论点的论据。

这篇文章把这场电话会里真正有信息量的部分拆开讲。数据、逻辑、他自己承认的不确定性,都摆出来,你自己判断。

储能业务先破了纪录,而且他说这是确定的

开场第一个数字来自能源业务:Megapack 在 Q1 的部署量创下历史新高,能源板块的盈利能力也创了纪录。

这件事容易被忽略,因为大家的注意力都在车上面。但马斯克对这块业务的措辞非常罕见,他用了「我们其实知道它一定会增长」这种表述,并且明确说增速会显著快于汽车业务。

汽车业务的增长有天花板,受制于产能、需求、价格战、各国的补贴政策。储能业务的逻辑不一样,它绑的是电网改造和可再生能源并网这条长期曲线。马斯克的意思是,这条曲线的斜率现在比车陡。

他还顺带提了一句 AI 训练算力:Q1 的训练计算量环比翻了一倍以上。这个数字后面会展开讲,因为它牵扯到整场电话会最核心的叙事。

新车提前了整整半年,而且不用新建工厂

过去几周外界一直在猜特斯拉那款低价车型到底什么时候来。马斯克这次给了明确答复:

原计划是 2025 年下半年开始生产,现在提前到 2025 年初,乐观的话今年年底就能看到。

关键在下面这句:这些新车,包括更便宜的车型,会同时使用下一代平台和现有平台的技术,并且在现有的生产线上生产。

这句话的分量比发布日期更大。新建一座工厂意味着几百亿美元的资本开支和好几年的建设周期,任何一次延期都会直接打击现金流。而在现有产线上改造,等于把资本开支压缩到最低。

马斯克给出的产能数字是:完全实现后,特斯拉的总产能可以超过 300 万辆。目前特斯拉的年产能大概在 200 万辆上下,也就是说这套「混线生产」的方案能多榨出接近 50% 的产能,而不用多盖一座超级工厂。

这是典型的工程思维:与其赌一个新平台从零到一的成功,不如把现有资产的价值榨到极致。

FSD V12 到底变了什么,为什么他敢说「不可逆」

整场电话会里,马斯克最用力推销的东西是 FSD V12。他的原话是:

如果你还没体验过,我强烈建议你去试试,它是深刻的。而且改进的速度非常快。(马斯克)

V12 和之前版本的本质区别在于技术路线。之前是规则驱动的,工程师写代码告诉车「遇到红灯要停」「遇到行人要让」。V12 是端到端的神经网络,输入是摄像头画面,输出是方向盘转角和油门刹车,中间没有人工规则。

马斯克解释为什么这条路是对的,用了一个很朴素的类比:

整个人类的道路网络,本来就是为生物神经网络和眼睛设计的。路标、车道线、红绿灯,全都是给人眼看的。所以摄像头加数字神经网络,天然适配现有的道路系统。这也是他坚决不用激光雷达(LiDAR)、不用雷达、不用超声波传感器的理由。

推送范围上,V12 已经推给了北美所有配备摄像头和推理计算机的车辆,从 Hardware 3 开始算,大概 180 万辆。目前约一半的车主在用,而且这个比例每周都在涨。

关于累计行驶里程,官方口径是「FSD V12 上线以来超过 3000 亿英里」。这个数字大概率是现场口误或语音识别的偏差,因为地球到太阳的距离也才 9300 万英里,3000 亿英里是一个不可能的量级。但不管具体数字是多少,重点在于他在强调数据飞轮的规模,这一点后面讲车队的时候还会回到。

为了让更多人试,订阅价格降到了每月 99 美元。这个动作的意图很清楚:扩大用户基数,收集更多真实路况数据,继续喂模型。

训练算力:从卡脖子到不卡脖子,只花了几个月

这一段是技术听众最该留意的部分。

马斯克承认,过去一段时间特斯拉的训练进度是被算力卡住的。现在情况变了:

已经安装并投入运行的 H100 大约 3.5 万张,年底预计达到 8.5 万张左右,全部用于训练。

他在这里插了一句吐槽,说 GPU 这个词用错了,G 代表 graphics,但 H100 根本不干图形的事,「每次说 GPU 我都觉得别扭,这行得有个新词」。这个小细节其实说明他对算力硬件的认知是深入到芯片层面的,不是外行看热闹。

更重要的信息是下面这句:光堆 H100 的数量没用,怎么高效地用起来才是关键。他甚至说,特斯拉的 AI 推理效率远超任何其他公司,「没有一家公司接近特斯拉的推理效率」。

为什么特斯拉能做到?因为车端的推理硬件是被约束死的,功耗、散热、成本都有硬上限,工程师只能在有限的算力里榨出最多的性能。这种约束逼出来的优化能力,反哺到训练侧和机器人侧,就是别人花钱也买不到的东西。

Optimus:他赌的是整个经济规模的上限

关于人形机器人,马斯克给了两个时间点,并且自己标注了这是猜测:

  • 今年年底前,Optimus 会在特斯拉自己的工厂里做有用的任务,进入有限量产
  • 明年年底前,有可能对外销售

然后是他那句每次都讲、但每次讲都值得琢磨的话:

我认为 Optimus 会比所有其他业务加起来的价值还要高。因为如果你有一个有感知能力的人形机器人,能在现实中导航、按需完成任务,那经济规模就没有实质性的上限了。(马斯克)

这个逻辑要成立,前提是机器人真的能干活。而马斯克认为特斯拉在这件事上有一个别人没有的优势:在机器人本体上做高效推理的能力。前面说的车端推理效率,会直接迁移到机器人身上。机器人不能拖根电线,不能背个机柜,功耗和散热约束比车还狠。

他还补了一句,语气比平时谨慎:

我会对 Optimus 更保留一些。如果我们有一个超级智能的人形机器人,能跟着你进室内,而你逃不掉,那就是终结者级别的风险。如果我对它如何部署没有某种实质性的影响力,我会不舒服。(马斯克)

这段话放在财报电话会上其实挺奇怪。但它是真话,也解释了为什么他一直在强调自己要对公司保持影响力。后面讲股东投票和重组的时候,这条线索会接上。

机器人出租车不是 Uber,是 Airbnb 加 Uber

关于 8 月 8 日要展示的 Cybercab,马斯克这次给了一个更清晰的车队运营模型。

他说特斯拉会自己运营一部分车队,同时允许用户把自己的车加入或移出车队,随时可以收回。车主可以设置只让家人朋友用、只让五星用户用,或者对所有人开放。他明确对比了 Airbnb:你家客房想租就租,不想租就空着。

这个模型的意义在于,它把资产所有权和运营权解耦了。特斯拉不需要自己买下几千万辆车才能撑起一个巨型车队,用户手里的车本身就是供给。

支撑这个模型的,是数据飞轮。马斯克说得很直白:

我们已经有 700 万辆车,很快 900 万辆,最终会到几千万辆。每一次出问题,都会被加入训练数据,你就有了这个训练飞轮。这跟 Google 搜索是一样的,你很难跟 Google 竞争,因为人们不停在搜索、在点击,Google 一直在拿反馈。特斯拉也一样。(马斯克)

这也是他反驳「特斯拉只是车厂」的核心论据。车厂卖完车,关系就基本结束了。特斯拉卖完车,关系才刚开始,每一辆车都是一个持续产出数据的节点。

100 吉瓦的算力,藏在几千万辆车的后备箱里

这是整场电话会里最有想象力、也最容易被当成吹牛的一段。但它的算术其实很朴素。

马斯克说,未来可能出现一个类似 AWS 的业务:当车停着不动的时候,车上的推理计算机可以跑分布式推理。

算一笔账:假设车队规模到 1 亿辆,平均每辆车有 1 千瓦的可用推理算力,那就是 100 吉瓦的分布式算力。他补了一句,地球上要凑出 100 吉瓦的 AI 算力非常困难。

有人会问,手机不也是分布式算力吗?马斯克的回答是:手机不行,因为满功率跑起来电池就没了。但车不一样,一台车带 50 到 60 千瓦时的电池包,跑 1 千瓦的推理负载 10 个小时,也就用掉 10 度电,对续航的影响可以忽略。

而且车还自带两个已经付过钱的东西:分布式的供电和分布式的散热。这两样在数据中心里是最贵的成本项之一。

他把这个类比直接套到了亚马逊身上:AWS 起步的时候,没人觉得它会成为亚马逊最值钱的业务,当时亚马逊就是个卖书的。AWS 之所以诞生,是因为亚马逊自己的算力需求在一年里有几个尖峰,其余时间大量闲置,于是他们把闲置算力租出去。特斯拉的逻辑一模一样。

硬件迭代的节奏也支持这个设想:现在路上跑的车大量是 Hardware 3,新车都是 Hardware 4,Hardware 5 已经基本设计完成,希望明年年底上车。等到车队规模上亿的时候,车上装的可能是 Hardware 6 或 7。

电梯和诺基亚:两个类比讲透监管和授权

关于自动驾驶的监管,马斯克的判断是:只要数据足够有说服力,监管不会是障碍。他的原话很直接:如果大规模数据能明确证明自动驾驶的事故率是人类驾驶的一半,那阻止自动驾驶就等于在杀人。

他用的类比是电梯。早期电梯有专门的 operator,拿着继电器开关操作。但人会累、会喝醉、会犯错,一不小心就把人夹在两層楼之间。后来电梯全自动了,你进去按个按钮,根本不会想它是怎么运作的。马斯克说,将来车也是这样:手机叫车,上车,到地方,下车,全程不用想。

另外他提了一句,Robotaxi 公司们在前面蹚监管的路,对特斯拉其实是好事。旧金山已经跑了一段时间, approvals 在加速。

FSD 授权:那个诺基亚时刻

关于对外授权 FSD,马斯克给了一个时间尺度:现在签的协议,大概三年后才能装到别人的车上。他管这个叫「闪电速度」,因为在汽车行业里,这确实算快。

但他判断这件事会从「可选」变成「必须」,理由是这个:

我还记得诺基亚如日中天的时候,手机卖得飞起。然后我看到他们出的智能手机,基本就是块砖,功能有限。再然后 iPhone 和安卓出来了,人们还是不明白所有手机都会变成那样,不会再有翻盖机了。(马斯克)

他的意思是,一旦消费者意识到「没有智能驾驶的车不值得买」,授权就不再是商业选择,而是生存问题。他原话是:授权,否则没人买你的车。

目前已经在和一家主要汽车制造商谈。他说今年有不错的概率签下来,可能还不止一家。

35 万亿个细胞:为什么要重组公司

有人问组织重组,马斯克的回答用了生物学类比,而且讲得相当完整。

他的逻辑是:一家公司在不同规模下必须有不同的组织结构。如果你是 10 个细胞、100 个、100 万个、10 亿个、1 万亿个,你不可能用同一套结构。人类大约有 35 万亿个细胞,但你感觉自己是「一个人」,不是一堆细胞。公司也一样,是个不断生长的生物体,长到某个阶段必须重组,否则就长不大。

他还给了一个量化说法:从 2019 年到现在,公司经历了一段很长的繁荣期。如果一家公司每年在组织上偏差 5%,累积下来就是很大的低效。他们一路做过一些修正,但现在是时候为下一个增长阶段重新组织。

这段话其实是在回应外界的裁员和人事动荡。他的框架是:这不是衰退信号,是生长痛。

被忽略的可负担性,和一分钟买车

有一段话值得单独拎出来,因为它讲的是一个常被硅谷忽略的问题。

马斯克说,对一个年入几十万美元的人来说,车价贵几千块无所谓。但绝大多数人是月光族,每个月租金或贷款的月供差 10 美元,就是实实在在的差别。所以定价不只是「性价比」问题,还是一个「根本可负担性」问题。这两件事必须同时解决。

他还提到,特斯拉在销售流程上把自己搞复杂了,过去一两周刚刚做了大幅简化。他的目标是:一分钟内买完一辆特斯拉。之前的流程太繁琐了。

Q1 交付承压的原因,他给了两个:一是车压在船上,物流复杂,Model 3 和 Model Y 在某些市场过剩、某些市场缺货;二是销售流程本身太复杂。

关于 Q2,他的判断是「会好很多」。关于全年,他说销量会高于去年。

他还顺带回应了中国市场的竞争:如果比销量下滑的幅度,特斯拉下滑得比竞争对手少。

电池和 4680:一个对冲工具,不是一个执念

4680 电池经常被当成特斯拉「什么都自己做」的证据。马斯克这次的解释不太一样。

他说 4680 的初衷是对冲。前几年所有大车厂都在下巨额电池订单,导致锂离子电池每千瓦时的价格涨到疯狂的水平,是预期的两倍。这种情况下,如果自己没有电芯产能,就只能任人宰割。所以做 4680 是为了应对需求冲击,不是因为「我们喜欢给自己找麻烦」。

现在的局面变了:其他车厂的电池订单大幅下降,供应商的产能过剩,报价变得比以前有竞争力得多。所以 4680 的战略价值在下降,但电池行业本身会像 DRAM 一样周期波动,今天的情况不代表明天。

他还吐槽了 IRA(通胀削减法案)这个名字,说每次听到都以为是爱尔兰共和军,或者俄罗斯的互联网研究机构,或者罗斯 IRA 退休账户,「像个蜘蛛侠的梗」。这个玩笑背后是个真问题:法案对美国本土生产的电芯有额外需求,但法案能持续多久,没人知道。

最后那句关于外星人的话

有人问他个人对公司的依赖程度。马斯克的回答是:

即使我明天被外星人绑架了,特斯拉也会解决自动驾驶,可能慢一点,但车辆这块的自动驾驶会解决。我不确定 Optimus 或者未来产品上我们还能不能赢,但就车辆自动驾驶而言,特斯拉有足够的动能。(马斯克)

这句话的潜台词是:车辆自动驾驶已经过了「靠某个人拍板」的阶段,进入了工程和数据的惯性轨道。而 Optimus 还在早期,还需要他本人的判断和推动。这也解释了为什么他在 Optimus 那段特别强调「我要对部署有实质影响力」,以及为什么股东投票这件事对他很重要。

整场电话会其实只讲了一个估值框架

把所有内容串起来看,马斯克的论证链条是这样的:

第一,FSD V12 证明了纯视觉端到端路线可行,这是自动驾驶的技术底座,也是所有后续业务的入口。

第二,车队规模带来数据飞轮,飞轮一旦转起来,后来者很难追。这是护城河。

第三,车队本身就是算力网络,闲置时间可以变现,这是第二条增长曲线。

第四,同样的推理能力迁移到人形机器人,打开的是经济规模的上限。

在这个框架下,卖车只是获取数据和部署硬件的手段,不是终点。所以他才会说:用汽车公司的框架去估特斯拉,是问错了问题,问错了问题就不可能得到正确答案。

你可以不同意这个框架,但至少要知道他在用什么框架跟你对话。这场电话会的信息密度,比一个季度的交付数字重要得多。

如果你想自己验证他的说法,最直接的办法不是看财报,是去开一次 FSD V12。他自己在电话会里反复说了三遍,大意是:不开一次,你根本不可能理解这家公司。

  • 去试一次 FSD V12。 这是马斯克在电话会上唯一重复三次的建议。如果你在北美,订阅价格已经降到每月 99 美元,试一个月,用自己的体感判断他说的「深刻」是不是真的。不在北美的,找找身边有没有体验过的朋友,问具体场景,别问感受。
  • 把特斯拉的估值拆成三块看。 汽车业务按制造业估值,储能业务按能源基建估值,FSD 和 Optimus 按期权估值。混在一起算,怎么算都算不对。分开算,你至少知道自己付的钱买的是哪一块。
  • 盯住 8 月 8 日的 Cybercab 发布。 这是检验车队运营模型是否成立的关键节点。重点看三件事:车队的所有权和运营权怎么分、单车成本、以及有没有给出可验证的时间表。
  • 关注 FSD 授权的那份协议。 马斯克说今年有不错的概率签下来,可能不止一家。一旦签了,这是纯视觉路线被传统车厂认可的第一个硬证据。但记住他自己的话:签完到装车,大概要三年。
  • 对「100 吉瓦分布式算力」保持好奇,但别急着下注。 这个算术成立的前提是车队规模到 1 亿辆、单车算力到 1 千瓦、且监管允许车在闲置时跑第三方负载。三个条件缺一不可。把它当成一个长期期权,而不是近期催化剂。
  • 电池和储能两条线分开跟踪。 储能是当下已经被验证的增长曲线,有收入有利润;4680 是对冲工具,战略价值随供应商报价波动。别把这两件事混为一谈。
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, Martin. To recap, in Q1, we navigated several unforeseen challenges as well as the ramp of the updated Model 3 in Fremont. As we all have seen, the EV adoption rate globally is under pressure and a lot of other order manufacturers are pulling back on EVs and pursuing plug-in hybrids instead. We believe this is not the right strategy and electric vehicles will ultimately dominate the market. Despite these challenges, the Tesla team did a great job executing in a tough environment and energy storage deployments of Megapack, in particular, reached an all-time high in Q1 leading to record profitability for the energy business.

And that looks likely to continue to increase in the quarters and years ahead. It will increase. We actually know it will, so significantly faster than the car business as we expected. We also continue to expand our AI training capacity in Q1 more than doubling our training compute sequentially. In terms of the new product road map, there's been a lot of talk about our upcoming vehicle line in the next -- in the past several weeks. We've updated our future vehicle lineup to accelerate the launch of new models ahead, previously mentioned start of production in the second half of 2025. So, we expect it to be more like the early 2025, if not late this year.

These new vehicles, including more affordable models, will use aspects of the next-generation platform as well as aspects of our current platforms, and we'll be able to produce on the same manufacturing lines as our current vehicle lineup. So, it's not contingent upon any new factory or massive new production line, it will be made on our current production lines much more efficiently. And we think this should allow us to get to over 3 million vehicles of capacity when realized to the full extent. Regarding FSD Version 12, which is the pure AI-based self-driving, if you haven't experienced this, I strongly urge you to try it out, it's profound. And the rate of improvement is rapid.

And we've now turned that on for all cars with the cameras and inference computer everything from Hardware 3 on in North America. So, it's been pushed out to, I think, around 1.8 million vehicles, and we're seeing about half of people use it so far and that percentage is increasing with each passing week. So, we now have over 300 billion miles that have been driven with FSD V12 since the launch of full self-driving -- supervised full self-driving. It's become very clear that the vision-based approach with end-to-end neural networks is the right solution for scalable autonomy. And it's really how humans drive. Our entire road network is designed for biological neural nets and eyes.

So, naturally, cameras and digital neural nets are the solution to our current road system. To make it more accessible, we've reduced the subscription price to $99 a month, so it's easy to try out. And as we've announced, we will be showcasing our purpose-built robotaxi or Cybercab in August. Yeah. Regarding AI compute, over the past few months, we've been actively working on expanding Tesla's core AI infrastructure. For a while there, we were training constrained in our progress. We are, at this point, no longer training-constrained, and so we're making rapid progress. We've installed and commissioned, meaning they're actually working 35,000 H100 computers or GPUs. GPU is wrong word.

They need a new word. I always feel like a wince when I say GPU because it's not. GPU stand -- G stands for graphics, and it doesn't do graphics. But you know, roughly 35,000 H100S are active, and we expect that to be probably 85,000 or thereabouts by the end of this year and training, just for training. We are making sure that we're being as efficient as possible in our training. It's not just about the number of H100s, but how efficiently they're used. So, in conclusion, we're super excited about our autonomy road map. I think it should be obvious to anyone who's driving Version 12 and it is only a matter of time before we exceed the reliability of humans in not much time with that.

And we're really headed for an electric vehicle and autonomous future. And I go back to something I said several years ago that in the future, gasoline cars that are not autonomous will be like riding a horse and using a flip phone. And that will become very obvious in hindsight. We continue to make the necessary investments that will drive growth and profits for Tesla in the future, and I wanted to thank the Tesla team for incredible execution during this period and look forward to everything that we have planned ahead. Next.

We are able to do simple factory tasks or at least, I should say, factory tasks in the lab. We do think we will have Optimus in limited production in the natural factory itself, doing useful tasks before the end of this year. And then I think we may be able to sell it externally by the end of next year. These are just guesses. As I've said before, I think Optimus will be more valuable than everything else combined. Because if you've got a sentient humanoid robots that is able to navigate reality and do tasks at request, there is no meaningful limit to the size of the economy. So, that's what's going to happen.

And I think Tesla is best positioned of any humanoid robot maker to be able to reach volume production with efficient inference on the robot itself. I mean, this, perhaps, is a point that is worth emphasizing Tesla's AI inference efficiency is vastly better than any other company. There's no company even close to the inference efficiency of Tesla. We've had to do that because we were constrained by the inference hardware in the car. We didn't have a choice. But that will pay dividends in many ways.

Yeah. It's actually been pretty helpful that the autonomous car companies have been cutting a path through the regulatory jungle. So, that's actually quite helpful. And they have obviously been operating in San Francisco for a while. I think they got approval for City of Vale. So, these approvals are happening rapidly. I think if you've got at scale, a statistically significant amount of data that shows conclusively that the autonomous car has, let's say, half the accident rate of a human-driven car, I think that's difficult to ignore because at that point, stopping autonomy means killing people.

So, I actually do not think that there will be significant regulatory barriers provided, there was conclusive data that the autonomous car is safer than a human-driven car. And in my view, this will be much like elevators. Elevators used to be operated by a guy with relay switch. But sometimes, the guy would get tired or drunk or just make a mistake and send somebody in half between floors. So, we just get an elevator and press button, we don't think about it. In fact, it's kind of weird if somebody is standing there with a relay switch. And that will be how cars work. You just summon the car using your phone. You get in. It takes you to a destination. You get out.

You don't even think about it, just like an elevator. It takes you to your floor. That's it. Don't think about how the elevator is working or anything like that. And something I should clarify is that Tesla will be operating the fleet. So, you can think of like how Tesla, think of it as combination of Airbnb and Uber meaning that there will be some number of cars that Tesla owns itself and operates in the fleet. There will be some number of cars and then there'll be a bunch of cars where they're owned by the end user.

That end user can add or subtract their car to the fleet whenever they want, and they can decide if they want to only let the car be used by friends and family or only buy five-star users or by anyone at any time they could have the car come back to them and be exclusively theirs like an Airbnb. You could rent out your guest room or not, any time you want. So, as our fleet grows, we have 7 million cars going -- 9 million cars going to eventually tens of millions of cars worldwide.

With a constant feedback loop, every time something goes wrong, that gets added to the training data and you get this training flywheel happening in the same way that Google Search has the sort of flywheel, it's very difficult to compete with Google because people are constantly doing searches and clicking, and Google is getting that feedback loop. So, the same with Tesla. But at the scale that is maybe difficult to comprehend, but ultimately, it will be tens of millions. I think there's also some potential here for an AWS element down the road where if we've got very powerful inference because we've got a Hardware 3 in the cars, but now all cars are being made with Hardware 4.

Hardware 5 is pretty much designed and should be in cars, hopefully toward the end of next year. And there's a potential to run -- when the car is not moving to actually run distributed inference. So, kind of like AWS, but distributed inference. Like it takes a lot of computers to train an AI model, but many orders of magnitude less compute to run it. So, if you can imagine future, perhaps where there's a fleet of 100 million Teslas, and on average, they've got like maybe a kilowatt of inference compute. That's 100 gigawatts of inference compute distributed all around the world. It's pretty hard to put together 100 gigawatts of AI compute.

And even in an autonomous future where the car is, perhaps, used instead of being used 10 hours a week, it is used 50 hours a week. That still leaves over 100 hours a week where the car inference computer could be doing something else. And it seems like it will be a waste not to use it.

Yeah. And we do have some insight into how good the things will be in like, let's say, three or four months because we have advanced models that our far more capable than what is in the car, but have some issues with them that we need to fix. So, they are there'll be a step change improvement in the capabilities of the car, but it will have some quirks that are -- that need to be addressed in order to release it. As Ashok was saying, we have to be very careful in what we release the fleet or to customers in general.

So, if we look at, say, 12.4 and 12.5, which are really -- could arguably even be Version 13, Version 14 because it's pretty close to a total retrain of the neural nets in each case are substantially different. So, we have good insight into where the model is, how well the car will perform in, say, three or four months.

Yeah. We'll talk about this more on August 8. But really, the way to think of Tesla is almost entirely in terms of solving autonomy and being able to turn on that autonomy for a gigantic fleet. And I think it might be the biggest asset value appreciation history when that day happens when you can do unsupervised full self-driving.

Yeah. It will be 7 million cars in a year or so and then 10 million, and then eventually, we're talking about tens of millions of cars -- not eventually, it's like before the end of this year. Yeah. But in the decade, it's several tens of million cars, I think.

We're in conversations with one major automaker regarding licensing FSD.

I think we've said we were on that front. So, what's your follow-up?

Well, as it constitutes a majority of my work time, and I work pretty much every day of the week, it's rare for me to take a Sunday afternoon off. I'm going to make sure Tesla is quite prosperous. And it is -- like it is prosperous, and it will be very much so in the future.

No, I think we'll have higher sales this year than last year.

I mean, I don't know what our competitors can do, except we've done relatively better than they have because if you look at the drop in our competitors in China sales versus our drop in sales, our drop was less than theirs. So, we're doing well. But I think Cathie Wood said it best. Like really, we should be thought of as an AI or robotics company. If you value Tesla as just like an auto company, you just have to -- fundamentally, it's just the wrong framework and if you ask the wrong question, then the right answer is impossible. So, I mean, if somebody doesn't believe Tesla is going to solve autonomy, I think they should not be an investor in the company.

Like that is, but we will and we are and then you have a car that goes from 10 hours of use a week, like an hour and a half a day to probably 50%, but it costs the same.

Yeah. If you've not tried the FSD 12.3, and like I said, 12.4 is going to be significantly better and 12.5 even better than that. And we have visibility into those things. Then you really don't understand what's going on. It's not possible.

Yeah. I mean, we're putting the actual auto in automobile. So, sort of we go like, well, sort of like tell us about future horse carriages you're making. I'm like, well, actually, it doesn't need a horse that's the whole point. That's really the whole point.

Well, I think no matter what Tesla -- even if I get kidnapped by aliens tomorrow, Tesla will solve autonomy, maybe a little slower, but it would solve autonomy for vehicles at least. I don't know if we would win on with respect to Optimus or with respect to future products, but it would that -- that there's enough momentum for Tesla to solve autonomy even if I disappeared for vehicles. Yes, there's a whole range of things we can do in the future beyond that. I'll be more reticent with respect to Optimus, if we have a super-sentient humanoid robot that can follow you indoors and that you can escape, we're talking terminator-level risk. And yes, I'd be uncomfortable with.

If there's not some meaningful level of influence over how that is deployed. And there's shareholders have an opportunity to ratify or reratify the sort of competition because I can't say that. That is a fact. They have an opportunity. And yes, we'll see. If the company generates a lot of positive cash flow, we could obviously buy back shares.

Yeah. We're not giving up anything that is significant that I'm aware of. We've had a long period of prosperity from 2019 to now. And so, if a company sort of organizationally is 5% wrong per year, that accumulates to of inefficiency. We've made some corrections along the way. But it is time to reorganize the company for the next phase of growth and you really need to reorganize it, just like a human when we start off with one cell and kind of zygote, kind of blastocyst and you start growing arms and legs and briefly, you have a tail.

You shed the tail, hopefully. And then you're a baby, and basically, you have to be the organism -- a company is kind of like creature growing. And if you don't reorganize it for different phases of growth, it will fail. You can't have the same organizational structure if you're 10 cells versus 100 versus 1 million versus 1 billion versus 1 trillion. Humans are around 35 trillion cells, doesn't feel like it feels like one person. But you're basically a walking cell colony of roughly 35 trillion depending on your body mass and about three times that number in bacteria. So, anyway, you've got to reorganize the company for a new phase of growth or will fail to achieve that growth.

Well, I think we just need to -- it just needs to be obvious that our approach is the right approach. And I think it is. I think we've now with 12.3, if you just have the car drive you around, it is obvious that our solution with a relatively low-cost inference computer and standard cameras can achieve self-driving. No LiDARs, no radars, ultrasonic. Nothing.

Yeah. So, it really just be a case of having them use the same cameras and inference computer and licensing our software. Once it becomes obvious that if you don't have this in a car, nobody wants your car. It's a smart car. I still remember in fact, when Nokia was king of the hill, yeah, and the cellphone is crushing. And then I saw them come out with a smartphone that was basically a brick with limited functionality. And then the iPhone and Android, people still do not understand that all the phones are going to be that way. There's not going to be any flip phones. There will be a niche product or home phone. Yeah. Not even exactly. When's the last time you saw a whole book?

That's like an idea. Yeah.

Yes, the hotels have them. The people don't understand all cars will need to be smart cars, or you will not sell, or the car will not -- nobody would buy it. Once that becomes obvious, I think licensing becomes not optional.

Yes, absolutely, it is. License it, or nobody will buy your car.

Yeah. I mean, really, a deal signed now would result in it being in a car probably three years.

That's like lightning, basically.

Yeah. So, I wouldn't be surprised if we do sign a deal. I think we have a good chance we do sign a deal this year, maybe more than one. But yes, it would be probably three years before it's integrated with a car. even though all you need is cameras and our inference computer. So, just talking about a massive design change.

Yeah. I think we can be free cash flow positive meaningfully.

Yeah. I mean, at the end of the day, like for any given company, if you sell a great product at a great price -- if you have a great product at a great price, the sales will be excellent. That's true of any area. So, over time, we do need to keep making sure that we're -- that it's a great product at a great price. And moreover, that price is accessible to people. So, it's not -- you have to solve both the value for money and the fundamental affordability question. The fundamental affordability question is sometimes overlooked. If somebody is earning several hundred thousand dollars a year, they don't think of a car from a fundamental affordability standpoint.

But vast majority of people are living paycheck to paycheck. So, it actually makes a difference if the cost per month for lease refinancing is $10 one way or the other. It is important to keep improving the affordability and to keep making the price --

Yeah, exactly. Make the price more accessible, the value for money better, and to keep improving that over time.

Yeah. It's going to be a great product and at a great price. And the standards for what constitutes a great product at a great price keep increasing. So, there's like -- you can't just be static. You have to keep making the car better, improving the price but improving the cost of production, and that's what we're doing.

You mean like new markets? Yeah, we are -- there are a bunch of markets where we don't currently sell cars that we should be selling cars in. We'll see some acceleration of that.

Yeah. So, think about the end-to-end neural net-based autonomy is that just like a human, it actually works pretty well without modification in almost any market. So, we plan on -- with the approval of the regulators, releasing it as a supervised autonomy system in any market that -- where we can get regulatory approval for that, which we think includes China. So, yes, it's -- just like a human, you can go rent a car in a foreign country and you can drive pretty well. Obviously, if you live in that country, you'll drive better. And so, we'll make the car drive better in these other countries with country-specific training. But it can drive quite well almost everywhere.

It understands that it shouldn't hit things, no matter what the road rules are.

Yeah. We think Q2 will be a lot better. Yeah.

Yeah, exactly. It's just -- if you've got cars that are sitting on ships, they obviously cannot delivered to people. And if you've got the excess demand for Model 3 and Model Y in one market, but you don't have it there. It's quite a -- it's extremely complex logistics situation. So, I'd say also the -- we did overcomplicate the sales process, which we've just in the past week or so have greatly simplified. So, it became far too complex to buy a Tesla, whereas it should just be you can buy the car in under a minute. So, we're getting back to the -- you can buy a Tesla in under a minute interface from what was quite complex.

Ashok, do you want to say something?

That's my guess, like a third of the hours of the week.

I think it's analogous to Amazon Web Services, where people didn't expect that AWS would be the most valuable part of Amazon when it started out as a bookstore. So, that was on nobody's radar. But they found that they had excess compute because the compute needs would spike to extreme levels for brief periods of the year and then they had idle compute for the rest of the year. So, then what should they do to pull that excess compute for the rest of the year. That's kind of -- yeah, monetize it.

It seems like kind of a no-brainer to say, OK, if we've got millions and then tens of millions of vehicles out there where the computers are idle most of the time that we might well have them do something useful. And then I mean if you get like to the 100 million vehicle level, which I think we will, at some point, get to, then -- and you've got a kilowatt of useable compute and maybe your own Hardware 6 or 7 by that time. Then you really -- I think you could have on the order of 100 gigawatts of useful compute, which might be more than anyone, more than any company, probably more than any company.

We've already learned about deploying workloads to these compute nodes.

Well, you're just draining the battery on the phone, so like technically, I suppose like Apple would have the most amount of distributed compute, but you can't use it because you can't get the -- you can't just run the phone at full power and drain the battery. So, for the car, even if you're a kilowatt-level inference computer, which is crazy power compared to a phone. If you've got 50- or 60-kilowatt-hour pack, it's still not a big deal whether you plugged it or not. It could be plugged in or not like you could run for 10 hours and use 10 kilowatt hours of your kilowatt of compute power.

Exactly. It's distributed power generation -- distributed access to power and distributed cooling. That was already paid for.

Yeah.

You know, we're making good progress on that. But I don't think it's super important for at least in the near term. As Lars said, we think it will exceed the competitiveness of suppliers by the end of this year. And then we'll continue to improve.

Yeah. I mean, a big part of the 4680, Tesla doing internal cells was a hedge against what would happen with our suppliers because for a while they are it was very difficult because every big carmaker put in massive battery orders. And so, the price per kilowatt hour of lithium-ion batteries went to crazy numbers, crazy levels.

Yeah, just bonkers. So, like, OK, we've got to have some hedge here to deal with cost per kilowatt hours of numbers that were double what we anticipated. If we have an internal cell production, then we have that hedge against demand shocks, we have too much demand. That's really the way to think about it. It's not like we want to take on a whole bunch of problems just for the hell of it. We did the cell program in order to address the crazy increase in cost per kilowatt hour from our suppliers due to gigantic orders placed by every carmaker on Earth.

Well, I want to say again, we'd just like to strongly recommend that anyone who is, I guess, thinking about the Tesla stock should really drive 12.3. It really -- you can't -- it's impossible to understand the company if you do not do this.

Great. What seems to be happening is that the nets are missing something, the orders for batteries from other automakers have declined dramatically. So, we're seeing much more competitive prices for sales from our suppliers, dramatically more competitive than in the past. It is clear that a lot of our suppliers have excess capacity.

Yeah. No, I mean, this is going to wax and wane, obviously. So, there's going to be a boom and bust in battery cell production where production exceeds supply and then supply exceeds production and back and forth kind of like, I don't know, DRAM or something. But it's like what is true today will not be true in the future, there's going to be somewhat of a boom-and-bust cycle here. And then there are additional complications with government incentives like the Inflation Reduction Act, the IRA -- I always found it like a funny name or a comical name. Yeah, it is like Irish Republican Army, the Internet Research Agency from Russia.

Yeah, exactly. Roth IRA. It's like a Spider-Man situation, which IRA wins. But it is complicate the incentive structure. So, that is there's the stronger demand for cells that are produced in the U.S. than outside the U.S. But then how long does the IRA last, I don't know.

Yeah. Well, I mean, the reason I reached out to you was because I thought your analysis of Tesla was the best that I had seen. Yes, thank you for helping Tesla to get to where it is today over seven years. It's been a pleasure working with you.

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。

现在已经非常清楚,基于视觉的端到端神经网络方案,是实现可规模化自动驾驶的正确解决方案。

It's become very clear that the vision-based approach with end-to-end neural networks is the right solution for scalable autonomy.

我认为Optimus将比其余所有业务加起来更有价值。

I think Optimus will be more valuable than everything else combined.

所以我其实不认为会有重大的监管障碍,前提是有确凿数据表明自动驾驶汽车比人类驾驶的汽车更安全。

So, I actually do not think that there will be significant regulatory barriers provided, there was conclusive data that the autonomous car is safer than a human-driven car.

所以,如果你能想象未来,也许有1亿辆Tesla组成的车队,平均每辆车大约有1千瓦的推理算力。那就是分布在全球各地的100吉瓦推理算力。

So, if you can imagine future, perhaps where there's a fleet of 100 million Teslas, and on average, they've got like maybe a kilowatt of inference compute. That's 100 gigawatts of inference compute distributed all around the world.

但真正看待Tesla的方式,几乎完全在于它能否解决自动驾驶,并让一个庞大的车队启用自动驾驶。

But really, the way to think of Tesla is almost entirely in terms of solving autonomy and being able to turn on that autonomy for a gigantic fleet.

我们正在与一家大型车企就授权FSD进行谈判。

We're in conversations with one major automaker regarding licensing FSD.

所以我的意思是,如果某人不相信Tesla会解决自动驾驶,我认为他就不该投资这家公司。

So, I mean, if somebody doesn't believe Tesla is going to solve autonomy, I think they should not be an investor in the company.

我认为,现在有了12.3,你只要让车带你四处开,就会明显看出:我们用相对低成本的推理计算机和标准摄像头就能实现自动驾驶。没有激光雷达,没有雷达,没有超声波。什么都没有。

I think we've now with 12.3, if you just have the car drive you around, it is obvious that our solution with a relatively low-cost inference computer and standard cameras can achieve self-driving. No LiDARs, no radars, ultrasonic. Nothing.

我们做电芯项目,是为了应对供应商每千瓦时成本疯狂上涨,而这是因为地球上每家车企都下了巨额电池订单。

We did the cell program in order to address the crazy increase in cost per kilowatt hour from our suppliers due to gigantic orders placed by every carmaker on Earth.

所以,我们看到供应商给出的售价竞争性强得多,比过去强得多。很明显,很多供应商产能过剩。

So, we're seeing much more competitive prices for sales from our suppliers, dramatically more competitive than in the past. It is clear that a lot of our suppliers have excess capacity.

资源与出处

相关视频

同频道、同主题或相邻年份