$TSLA

$TSLA · Solving The Money Problem · 2026-11-18

AI 摘要 · Solving The Money Problem

Elon Musk 认为 Tesla 的 robotaxi 与 Optimus 项目进展良好:Cybercab 已在 Austin 投入运营,并正扩展至德州更多城市和佛罗里达州,加州或于明年初获批;出于安全考虑,推进节奏保持缓慢,以免被监管机构叫停。他预计 Optimus 将在 12 至 18 个月内承担家务,并在一两个月内进行演示,同时提醒不要过度自信。

看涨他说的是几个月内
《Elon Spills The Beans In New Interview (Supercut)》
Solving The Money Problem引用来源:Elon Musk发布 2026-11-18 · 11 段内容

播放时,YouTube 可能接收设备与播放信息并使用 Cookie。 隐私说明 · Google 隐私政策

05:16/ —打开原视频

逐段跳播

11 段内容
5:1616:14

我确实对事情的走向感觉不错。我们有很多来自 SpaceX 本部的员工,以前负责 Starlink 软件和 Starship 软件的人,现在在主导重写,主导新的 AI 软件栈,他们非常出色。

AI 训练有一个很有意思的地方:做训练时,你必须让大量 GPU 协同一致地工作,几十万块 GPU,而这会变成一个网络问题。就是如何让所有这些 GPU 以尽可能高效的方式互相通信?

所以这本质上是一个网络软件问题,而这恰恰也是 Starlink 在做的事——当你在太空中重建整个互联网时,这显然是一个网络问题。所以我们发现,这项技术和技能可以很好地迁移到 AI 训练上。

Um, and uh I do feel good about where things are going. Um we've got um uh a lot of the people from SpaceX proper um the the the you know we've got the the people that used to run Starlink software and uh Starship software now uh leading the uh rewrite you know leading the new um AI software stack um and and they're extremely good and and sort of an interesting thing about AI training is that when you do the training you you've got to have a large number of of GPUs work together coherently um hundreds of thousands of them and and it interestingly it becomes a network networking problem.

It's like how do you get all of these GPUs to talk together in in the most efficient manner possible? And this that so it's it's it's a yeah like a networking software problem essentially which is exactly what Star Link is also uh you know when you're rebuilding the entire internet and space uh it's it's a it's obviously a networking problem.

Um and so we found that that technology and skill is very transferable to uh AI training.

再说一次,我不想过度自信,也不想自满,但这么说吧,如果现在有谁的训练软件比我们更好,我会非常意外。而且显然,对于 Tesla AI,我们的目标不是达到某个正确率百分比,我们的目标是做到完美无瑕,让它不会撞车。

Um, again, I don't I don't want to be overconfident or or or complacent, but I I would be let's just say I would be very surprised if anyone has better training software than we do at this point.

Um, and and and obviously with with Tesla AI, our goal is not to get some percentage correct. Our goal is to get to make it flawless so it does not crash.

所以这个回答有点长,但我们目前正在做 GR 4.8。它正处于 RL 阶段,我们会通过 Grokbot 做一些周期性的软发布。所以你在 Grokbot 上体验到的,实际上是我们在后台测试的各种 Grok 版本,用来判断它是否有用,比如它是否解决了用户的问题,然后再利用它是否解决了问题的数据来改进 Grok。

所以我们有一个很好的数据飞轮,让 Grokbot 随着时间越来越好。

So this is kind of a a long way long it's kind of a long answer but um but we are working on on on GR 4.8 right now. Uh it's in the it's in the RL phase um and and we are we we do sort of periodic like soft releases under Grockbot.

So um you know so and and and so what you'll experience with rock bottoms is actually um you know various various versions of Grock under the hood that we're we're testing to determine its usefulness like is it is it solving the user's problem um and and then and then and then using that data of whether it did or did not solve the problem to to improve to improve Grock.

So, we've got we've got a good data flywheel uh to keep making rockbot better and better over time.

所以,是的。我认为 Optimus 正在取得不错的进展。等它准备好时我们会做一次演示,我认为这个时机就是当我们能做出一次我们觉得真实、而不是摆拍或预编程的演示时,因为你看到的人形机器人演示其实都是摆拍和预编程的,不是真的。

实际上没有任何机器人能让你在真实世界里给它任意任务,然后让它对从未见过的东西直接去完成这些任务。所以就像我说的,你迄今看到的机器人演示,就像如果你让它去拿一个苹果,然后你移动桌子,它就再也拿不到那个苹果了。

所以你至少应该能想明白,好吧,即使苹果在桌上的位置被移动了,我也得去把苹果拿起来,诸如此类。所以我们会做一次演示,可能是在一两个月后,我不确定,但那会是一次真实的演示,你看到的就是你日常与 Optimus 互动时实际会遇到的表现。

>> So, yeah. Um Optimus I think is making good progress. Um and and we'll do a demo, you know, when it's ready, which is which is I think when we can do a demo that we feel that the demo is authentic and not just staged or pre-programmed because really the the the humanoid robot demos that you see are are staged and pre-programmed.

They are not real. um you you you can't actually there's there's no robot where you can um ask it arbitrary tasks in the real world and have it just go do those tasks in with something it's never seen seen before.

Um so like like I said the robot demos that you've seen thus far is like if you ask like go pick up an apple and then you and then you move the table it won't be able to pick up the apple anymore.

Um so you should at least be able to figure out okay I need to go even if they move the move the where the location of the apple on the table still pick up the apple you know type of thing.

Um so so we are going to do a demo and it might be I don't in a month or two but but it's going to be an authentic demo um where what what you see is what you would actually uh find with Optimus if you were to interact with it dayto-day.

但人形机器人,一个有用的人形机器人,是一个非常非常难的问题。尤其是手,你必须达到人类完成生产性任务所需的精细运动能力,这极其困难。围绕手和前臂——主要执行器所在的位置——的工程,比机器人其余所有部分加起来都更难。

Um but a humanoid robot a useful humanoid robot is a very very hard problem. you've you've you've especially with the hands, you've got to to achieve the the fine motor skills that that humans have to do productive tasks is incredibly difficult.

Um the the engineering around the the hands and the and the forearms where the primary actuators are um is is it's more difficult than the than the entire rest of the robot.

我认为 Optimus 大约在 12 到 18 个月内就能做到你描述的那些事情。太好了。是的。但你描述的那件事,比如拿起一个盘子,一个还留着食物残渣的盘子,一个脏盘子,然后把食物从盘子上擦掉,把它放到水槽下,洗一洗,擦干,然后安全地放回柜子里,这实际上非常——没有任何人形机器人能做到这一点。根本没有任何机器人能做到这一点。是的。

>> I think Optimus will be capable of doing the things you describe in in about uh 12 to 18 months. >> Awesome. Um yeah. Um but but the thing you described there of like picking up a say a plate that you know plate that's got still got food remnants on it.

So dirty plate and then wiping the food off the plate uh you know putting it under the sink, washing washing it, drying it and then putting it you know um safely back in in the closet is actually very there's there's no human robot that can do that. There's no human can can do that at all. Yeah.

但我认为 Octopus 能在 12 到 18 个月内做到。现在,如果你问最高、最好的用途是什么,大家都为此兴奋,老兄。不只是我,大家都为此兴奋。是的。是的。显然 Optimus 可以有很多用途。

它可以是你孩子的教学辅导老师,也可以当保姆,照看、保护你的孩子,确保他们没事,并教他们东西;它可以修剪草坪,可以去做晚饭,等等。我认为可以公平地说,大多数人都会想要拥有自己的私人 R2-D2、C-3PO,但那比那还要有用得多。

拥有它会非常非常酷。但如果你问 Optimus 最初最高、最好的用途是什么,那大概率是重工业。比如采矿和矿石精炼,精炼有很多阶段,因为在美国,我们在采矿和矿石精炼上对其他国家的依赖程度非常高。

我认为为了国家安全,我们需要确保能在国内完成这些。然后,对于任何 24/7 的任务,也就是每周使用 168 小时的任务,Optimus 也会是最经济合理的首选。所以,一旦你度过了那些 24/7 的重工业工作,到了某个时点,人们在家里拥有一台就会变得合理。

而且我认为几乎每个人都会想要一台,如果不是的话。

>> Um so, um but I think Octopus will be able to do it in 12 to 18 months. >> Now, if you say like what is the highest and best everybody's excited for that, man. I'm not alone.

Everybody's excited for that. >> Yeah. Yeah. And and and obviously Optimus could could be many things. It could it could be a a a teaching tutor for your for your kids and you know um you know protect it could you know be a babysitter watch over protect your kids make sure they're they're okay um and and teach them things um it could u yeah mow the lawn it could go you know whatever cook dinner um and um and I think it's it's fair to say that um I think most people would love to have their own personal R2-D2 C3P PO but that but that's even more useful than that.

Um it would be very very cool to have that and but if you but but you say like what's the initial highest best use for Optimus? It would be probably heavy industry. Um so like um mining and refining ore um and uh and uh and there are many stages of refinement uh because in in the like we we in the United States has a very uh high degree of dependence on other countries for mining and refining ore.

Um and and I think we we need to make sure for our national security that that we can do so with within the country. Um and and and and then and then for [clears throat] Optimus for for any kind of 24/7 task where that it's in use 168 hours a week that that will also be the you know the the sort of most economically sensible thing for to do at first.

Um so um but then once you get through of those those sort of 24/7 um like kind of heavy industry jobs uh then at some point it's it um it it'll make sense for people to have one in their in their home.

Um and um and and I think pretty much everyone's going to want one if not

是的,我们在 CyberCab 的推出上非常谨慎,因为它没有方向盘和踏板,所以没有后备方案,没有其他控制方式。所以我们想非常小心,迄今为止,我们的 robotaxi 推出——robotaxi 是包含使用 Model Y、Model 3 和其他车辆以及专门用于完全自动驾驶的 CyberCab 的广义系统——没有发生任何严重伤害或事故。

我们只是想非常小心,不要发生任何伤害到人的事故,进而导致整个项目被监管机构叫停。但它运行得非常好。你现在去 Austin,几分钟内就能叫到一辆 CyberCab。它们遍布 Austin,我们正在 Texas 其他主要城市大规模扩展 CyberCab。

然后是 Florida。我认为也许明年初我们会获得在 California 运营的批准。所以我们正在尽可能快地扩展 CyberCab 项目和整个 robotaxi,同时保持非常高的安全水平,并取得各项监管批准。但它将深刻改变城市的面貌。

yeah we are being very careful with the roll out of CyberCap because it has no steering wheel and pedals so there's no no fallback. there's no other way to control it. Um uh so we want to be very careful and thus far we've had no um you know no serious injuries or or or accidents with uh robot roll out of robo taxi or robo taxi being the broad system that includes use of model wise and model 3s and other cars or the cyber cab which is specifically for um fully autonomous driving.

Um, we just want to be very careful and not have any accident that causes that harms someone and then causes the whole program to be shut down by regulators. Um, but it is working very well.

You can go to Austin and and get a a cyber cab in minutes right now. Uh, they're all over Austin and we're we we're expanding to a large large numbers of cyber cabs in other major Texas cities.

Um, and then Florida. I think uh perhaps early next year we'll get approval to operate in California. Um and uh and so we're expanding the savicap program as and and robot taxi in general as fast as we can uh while maintaining very high levels of safety um and getting the various regulatory approvals.

Um but it it is uh going to profoundly change the look of cities.

是的。所以我的意思是,Ter 仍处于起步阶段。显然,我们今年早些时候才宣布了它。但我想我们已经招募了一支相当强的团队,而且我们还在寻找更多优秀的工程师加入 Terafab 项目。

这是世界上最令人兴奋的项目之一,要在地球上建造最大、最先进的芯片制造工厂,事实上它将是地球上任何类型中最大的建筑。所以我们预计它最终会达到1亿平方英尺,并密集地配备世界上最先进的设备。

我们一开始会使用相对标准的设备,但之后我们会推动供应商制造特殊版本的机器。在某些情况下,如果我们认为为了改进生产工艺必须这样做,我们可能会自己制造机器。所以是的,我认为进展相当顺利。

我们预计这个研究型晶圆厂大约在明年6月左右投入运营,并希望能造出我们的第一颗测试芯片。我想距离那大概还有9个月,或者8到9个月。如果一切顺利,我们应该会在2028年底左右实现量产。

至少是 Sarafab 的第一阶段,然后我们会进入第二阶段及以后。

Yeah. So I mean Ter is still in its infancy. Obviously, we, you know, we only announced it um earlier this year. Um but we've already recruited, I think, quite a strong team and we're looking for more uh great engineers to join the the Terafab project.

Um it is one of the most exciting projects in the world uh to create the largest most advanced chip fabrication facility uh on earth and in fact it'll be the largest building of any size uh largest building of any kind on earth.

So, you know, we expect it ultimately be 100 million square ft and um densely packed with the most advanced equipment in the world. Um and you know, we'll we'll use kind of the relatively standard equipment at first, but then we'll be pushing suppliers to um do special versions of the machines.

In some cases, we we may make our own machine if we feel that um that that uh that has to be done in order to improve the production process. Um so yeah so yes I think it's going quite well.

Um we we expect to have the uh the re the sort of research uh fab uh in in operation and and hopefully make our first uh test chip around June of of next year or so. Guess we're about like 9 months or yeah eight or nine months away from that.

Um and and and if if all goes well, we we should have um we will achieve volume production um towards the end of uh 2028. Um at least for phase one of of um Sarafab and and then we'll go to phase two and and onwards.

但我认为你会看到芯片制造方式出现一些非常迅速的演变。芯片制造公司对他们非常尊重,他们能力极强,但那里存在一种风险规避,考虑到他们必须投入巨额资本支出来建造晶圆厂,这是可以理解的。

所以当你有那么多在线产能时,就不愿承担太多风险。我们将承担大量风险,快速迭代,最终我认为我们会拥有比任何人都高效得多的芯片生产工艺。这是我们的假设。我们需要让它真正实现。但此时此刻我有信心,我不想乌鸦嘴。

敲敲木头。但我谨慎乐观地认为,到28年我们将拥有显著更好的芯片生产工艺。

Um but I think you'll see some very rapid evolution in the way that that chips are made. Um I the chip making companies have massive respect for them and they're they're incredibly competent but there there is a risk aversion there which is understandable given the massive amount of capex that they have to devote to uh to building a fab.

So that when you have that much online there's a reluctance to take too many risks. Um we're going to take a lot of risks iterate fast and um and ultimately I think have a much more efficient production process for chips than anyone else.

That's our hypothesis. Um, we need to make actually have that come true. Uh, but I I I at this point I'm I'm confident I I don't want to jinx it. Knock on wood. I'm but I I I I am cautiously optimistic um that we will um have a significantly better uh uh chip production process by 28.

你看到那些关于它将把我们带到10 G算力的估计,我认为其中大多数都差了两倍,有时甚至超过两倍。>> 是的。所以你以人们预期一半的成本建造,并且你在将其变现。>> 以某些人、某些人预期一半的成本 >> 当我看到行业共识数字时,我想我们大概在,也许拉到那个数字的60%左右。

这大概会是一个合理的猜测。而且,如果你看我的过往记录,我有在资本支出上极其高效的记录。我的意思是,我认为人们,我可能是资本支出效率最高的,但如果不是,我也非常接近最好。而且正如你提到的,我们确实有 SpaceX 不可思议的人才优势。

然后当然,我们与 Tesla 在 mega packs 上密切合作。而且 Tesla 可以说拥有世界上最好的电气工程人才,我认为也许毫无疑问是最好的。所以我们将它与 mega packs 结合起来,这对电网稳定或总体上的涡轮机稳定很重要,因为 GPU 的负载变化相当大,除非你有某种东西能在规模上、在很宽的频率范围内做功率平滑,否则例如进行训练运行就非常困难。

所以你必须平滑功率,否则你会弄坏涡轮机的曲轴,或者弄坏涡轮机的轴。所以

You see these estimates for where it will take us for 10 G of compute and I think they're off by a factor of two most of them. and some sometimes more than a factor of two. >> Yeah.

So you're building at half the cost that people expect and you are monetizing it. >> Half the cost that some people some people >> when I look at the industry consensus number I you know I I think we we we're somewhere around maybe pull it 60% of that number.

It would be probably a reasonable guess. Um and uh and it's you know if you look at my track record I I I have a track record of being extremely efficient with capex. I mean it's I think people you know I might be the best at capex efficiency but if not I'm pretty close to being the best.

Um um and and we do have the advantage as you alluded to of of incredible talent at SpaceX. Um and then of course we work closely with Tesla with the mega packs. Um and uh and Tesla has you know arguably the best electrical not I think maybe unequivocally the best electrical engineering talent in the world.

Um, so we combine it with the mega packs u which are important for uh grid stabilization or just in general turbine stabilization because the the loads that the GPUs do uh vary considerably and unless you've got something that's going to um do power smoothing u at scale um at at a wide range of frequencies then uh it's very difficult to do training runs for example.

Um, so you you've got to smooth out the power otherwise you'll you'll break the crankshaft on the turbines uh or break the shaft on the turbines. So

观察点

加州监管机构对Tesla robotaxi运营的批准
获得在加州运营的批准
明年初加州监管机构的批准决定
支持扩张逻辑
Optimus演示处理其此前未见过的任意家务任务
真实、非脚本化的演示成功
一两个月已宣布的Optimus演示
支持Optimus时间表

这个频道对 $TSLA 的历次判断

Solving The Money Problem 在这只股票上共 36 条判断,这里只列相邻的几条。

跨洋听风