Industry Insider Speaks Out / Wall Street Drops Robotaxi Research / Optimus Production Estimates ⚡️
发布时间 2026-01-17 01:27:11 来源
Welcome to Electrified, it's your host Dylan Umas, quick shout out to Paul G, thank you for using my Tesla referral link and congratulations on your new Tesla. Pierre Faragou from New Street Research finally put out his Robotaxi report for 2026, but before I touch on just a few slides, I want to put a disclaimer there about all of this modeling. Because back in the day, it's something I used to do myself and I really enjoyed it and I thought it was valuable and don't get me wrong, there can be instances where it is. I just want people to be cautious because we know Elon has gotten plenty of timelines wrong, so of course all of the analysts are ultimately just guessing as well.
欢迎来到《Electrified》，我是你的主持人Dylan Umas。首先感谢Paul G，感谢您使用我的特斯拉推荐链接，并祝贺您购买了新的特斯拉。New Street Research的Pierre Faragou终于发布了他关于2026年Robotaxi的报告，但在我讨论其中几个幻灯片之前，我想先做一个免责声明。因为在过去，我也曾从事这类建模工作，并非常享受其中，也认为它很有价值。当然，它在某些情况下确实有意义。但是我想提醒大家要谨慎，因为我们知道埃隆·马斯克有过很多时间预估错误，所以所有的分析师最终也只是在猜测。
And as you'll see, it's not like Waymo's dominating the markets where it's competing against Uber and Lyft. Now granted, their costs are still higher, which is certainly a problem. But my point is determining what percentage of the population will remain skeptical of cars driving themselves with no human is going to be very difficult and I actually think it'll be very different market to market. A very tech forward market like San Francisco is going to be very different than a more rural setting, but really my biggest problem is just understanding how the market actually works.
正如你所看到的，Waymo 在与 Uber 和 Lyft 竞争的市场上并没有占据主导地位。当然，他们的成本仍然较高，这确实是个问题。但我的观点是，要确定有多少人会对无人驾驶汽车保持怀疑态度是非常困难的，而且我实际上认为这会因市场而异。一个非常科技前沿的市场，比如旧金山，将与更乡村的环境有很大不同。但实际上，我最大的困惑是如何真正理解这个市场是如何运作的。
For example, even if Pierre's model was spot on with every single number, that doesn't mean that the stock market is going to apply this perfect expected multiple to any of these companies at any given time. With so much of the stock market capitalization being driven by AI, it's not really driven by fundamentals and valuations like it once was. And we've watched Tesla go from $400 billion to a one and a half trillion dollar valuation and back down where all during that time not much has changed with Tesla's profitability. So what I'm saying is that company valuations very rarely actually reflect real world achievement in real time.
例如，即使皮埃尔（Pierre）的模型在每个数字上都完全准确，这也不意味着股市会在任何时候将这个完美的预期倍数应用到这些公司上。由于现在股市市值很大程度上受人工智能的驱动，它不再像过去那样主要靠基本面和估值推动。我们曾目睹特斯拉的市值从4000亿美元涨到1.5万亿美元，又跌回去，而在此期间，特斯拉的盈利能力几乎没有发生大的变化。所以，我的意思是，公司估值很少真正反映实时的现实成就。
And then of course you always have to consider the macro. You could have this perfect model built out where Tesla's robot taxi hits every milestone as expected, but if that happens during a bear market or during wartime or during a time where there's any major negative economic event, then you're back to just waiting and riding it out even if you have the perfect model built out. So ultimately I kind of view them as these intellectual mirages. They can be exciting and enticing, but is it really something that we can change our behavior based on?
当然，你总是需要考虑宏观经济因素。你可能设计了一个完美的模型，预测特斯拉的机器人出租车能够按照预期达到每一个里程碑，但如果在熊市、战争时期或任何重大负面经济事件期间发生这一切，即使模型再完美，你也只能等待并熬过去。所以，我有点把它们看作是智力上的海市蜃楼。它们可能让人感到兴奋和诱人，但这真的是我们能据此改变行为的东西吗？
And in the intro, Pierre said against conventional wisdom, Uber is likely to retain a solid market position as robot taxi will not meet all demand and use cases. And as we'd expect robot taxi challengers to do best partnering with Uber. But look, as I've said, I certainly don't think Uber will become irrelevant overnight and this chart would prove as much. Here we're looking at Waymo's trips in the San Francisco market dating back to August 2023.
在开头，皮埃尔提到，与传统观点相反，Uber可能会保持稳固的市场地位，因为机器人出租车并不能满足所有的需求和使用场景。同时，我们预计机器人出租车的挑战者最好能与Uber合作。然而，正如我所说，我并不认为Uber会在一夜之间变得无关紧要，而且这张图表也证明了这一点。这张图显示的是Waymo在旧金山市场自2023年8月以来的行程数据。
So in the two years that this chart is showing, you'll see Waymo's market share of ride hailing trips is still only at 15% after two years. Meaning companies like Uber and Lyft still retain the majority of the market share. And of course it's not just because people are skeptical of self-driven cars. It's driven at least in part because in that market, Waymo's cost per mile is $3.30 compared to Uber at $2.50. Now I will say I definitely think with Waymo's new O-Hi, it's Gen 6 vehicle, there'll be a lot lower than $170,000 per unit.
在这张图表所显示的两年时间里，你会看到Waymo在打车业务中的市场份额仍然只有15%。这意味着像优步（Uber）和Lyft这样的公司仍然占据市场的大部分份额。当然，这不仅仅是因为人们对自动驾驶汽车持怀疑态度。至少有部分原因是Waymo在该市场的每英里成本为3.30美元，而优步为2.50美元。我要说的是，我确实认为随着Waymo的新款O-Hi第六代车型推出，他们每辆车的成本将会比17万美元低得多。
But as you'll see, this report was put together as of December 17th last year. And right here is Waymo's problem. They say pricing 30% below Uber would likely lead to an over 10x increase in volume, but lead to a half a billion dollar annual loss per city. And as we know, Waymo is not yet profitable even at this higher pricing structure and Uber today has a margin of roughly 9%. And so for Waymo to hit a number like that, being 30% cheaper than Uber, they're saying they would need a 40% reduction in their cost of good sold.
但是，如你所见，这份报告是在去年的12月17日编写的。Waymo的问题就在这里。他们表示，如果定价比Uber低30%，可能会导致业务量增加超过10倍，但每个城市则会造成每年5亿美元的损失。而且我们知道，即便在当前较高的定价结构下，Waymo也尚未实现盈利，而Uber目前的利润率约为9%。因此，为了让定价比Uber低30%，Waymo认为需要将其成本削减40%。
And of course for this purpose, that would include operating costs as well. And the report said, Robotaxi adoption remains limited to riders attracted by the tech or premium experience factors that justify higher pricing but can strain market share gains. So again, it's one thing to tell us that Waymo has to cut costs by 40%, but it's another thing to tell us how they're going to do that. And I hope you would agree, I'm really not a fan of oversimplification, but this is just what we've been saying for years without having a report like this.
当然，为了这个目的，还应该包括运营成本。报告指出，Robotaxi（无人驾驶出租车）的采用目前仍然局限于那些对技术或高端体验因素感兴趣的乘客，这些因素虽然可以支持较高定价，但可能会限制市场份额的增长。因此，说Waymo需要削减40%的成本是一回事，但告诉我们他们将如何实现这一目标则是另一回事。希望你能同意，我并不喜欢过度简化，但这正是我们多年来一直所说的，只是现在有了这样一份报告来支持。
If Waymo can't find a way to bring its cost down and ultimately start making money, it doesn't have a chance of surviving into the future. And if they've been struggling to do that, where Tesla's not competing in the market, how in the world is it going to happen when Tesla is? And just to be clear, the report does say that $170,000 per unit number is Waymo's Gen 5 fleet. This report then predicts Waymo's long-term costs but they still have the long-term projection at a $70,000 cost per car. That's 35K for the car, 32K for hardware, and then 8K for integration and calibration.
如果Waymo无法找到降低成本并最终开始盈利的方法，它在未来就没有生存的机会。而且，如果它们在特斯拉尚未进入市场时就已经很难做到这一点，那要是特斯拉加入市场竞争，该怎么办？需要说明的是，报告中提到的每台17万美元的数字指的是Waymo的第五代车队。报告还预测了Waymo的长期成本，但即便如此，他们的长期成本预估仍是每辆车7万美元。其中，车价为3.5万美元，硬件成本为3.2万美元，整合和校准费用为8千美元。
Whereas the Cyber Cab cost per car is going to be around $25,000 and hopefully a lot lower. Meaning, Waymo's long-term projections for what they might be able to get their cost down to is still multiple's higher than what Tesla's Cyber Cab will be this year. It says each Waymo market takes about three years to reach maturity, a year of testing and mapping, a pilot phase for limited users, and then a full commercial rollout that expands gradually to 15-20% share.
网络出租车（Cyber Cab）的每辆车成本将约为25,000美元，并且有望大幅降低。这意味着，Waymo在长期预测中，他们能够降低成本的可能性仍然比特斯拉今年的网络出租车要高出数倍。Waymo的每个市场大约需要三年才能成熟，包括一年的测试和地图绘制、为有限用户的试运行阶段，然后是逐步扩展至15-20%市场份额的全面商业推广。
So, Pierre's analysis means that until Waymo can materially change something with its cost structure, it's going to cap out at a market share between 10-20% in any given market. And who knows what Waymo thinks internally, but we know publicly that they're targeting a new 10-plus cities to launch in for 2026, which makes perfect sense if they're capping out at that 15% market share in each city. The report said Tesla is structurally better positioned than Waymo on cost, benefiting from a simpler sensor suite, lower manufacturing costs, and full vertical integration.
所以，Pierre 的分析表明，除非 Waymo 能实质性地改变其成本结构，否则它在任何市场的市场份额都会限制在10-20%之间。我们不知道 Waymo 内部是怎么想的，但他们公开表示计划在2026年前在10个以上的新城市推出业务。如果它们在每个城市的市场份额限制在15%，这种计划就很合理。报告指出，特斯拉在成本方面的结构性优势比 Waymo 更明显，因为特斯拉拥有更简单的传感器套件、更低的生产成本以及完全的垂直整合。
Tesla also avoids diminishing capital returns, like Uber it can tap a flexible supply base, meaning owners vehicles to meet peak demand, allowing the owned fleet to remain highly utilized at all times. These advantages combined with Waymo's long-cost curve give Tesla years to ramp a competitive robotaxi fleet. Then Pierre adds Tesla into the mix and it sounds like it's just his firm's estimates and analysis, but he puts Tesla in there at $2.20 per mile, meaning it's $0.30 cheaper per mile than Uber already, and $1.10 per mile cheaper than Waymo.
特斯拉也避免了资本回报递减的问题。类似于优步，特斯拉可以利用一个灵活的供应基础，这意味着它可以利用车主的车辆来满足高峰需求，从而使自有车队始终能高效利用。 这些优势加上Waymo的长期成本优势，使特斯拉有多年的时间来发展一个有竞争力的机器人出租车车队。 然后，Pierre将特斯拉纳入考虑，他的公司进行估算和分析，发现特斯拉每英里的成本为2.20美元，这意味着特斯拉的每英里成本已经比优步便宜0.30美元，比Waymo便宜1.10美元。
Which to me would likely confirm what I've been saying all along, that Tesla is most likely building its pricing algorithm as if the drivers are not part of the equation. Point being, when unsupervised finally ships in Austin, I don't think we see another huge 50% reduction in the cost per mile from Tesla. They're already pricing quite aggressively and low enough to take meaningful market share from the competition. Here they say what we've already known, Tesla's cost per mile could fall to around $0.50 per mile over time, 40% below Waymo's long-term potential.
这段话大致意思是：
我一直以来的观点可能会得到证实，即特斯拉在制定其定价算法时，很可能没有将司机因素考虑在内。重点是，当无人监管的驾驶技术最终在奥斯汀推出时，我认为我们不会看到特斯拉的每英里成本再大幅下降50%。特斯拉的定价已经相当激进，并且足够低，以从竞争对手那里夺取有意义的市场份额。这里提到的内容我们早已知道，即特斯拉的每英里成本可能会逐渐降至大约0.50美元，比Waymo的长期潜力低40%。
So yeah, Tesla has a structural cost advantage over Waymo, Waymo buys cars from OEMs paying the full price, including the manufacturer's gross profit, then buys LIDAR's, radars, sonars, cameras, and pays again to integrate all of that into the vehicle. And slides like this show where Tesla has another advantage with the customer owned fleet, and as we've been saying, this might not happen for a while because Tesla's going to have to validate all of those vehicles and it won't really need to do that until it starts to build out its own Tesla-owned network.
好的，特斯拉在结构成本上比Waymo有优势。Waymo需要从原始设备制造商（OEM）那里购买汽车，并支付全价，这个价格包括了制造商的毛利。随后，Waymo还需要购买激光雷达、雷达、声呐和摄像头，然后再支付费用将这些设备整合到车辆中。而一些展示还表明，特斯拉在自有车队方面也有优势。不过正如我们所说的，这种优势可能暂时还体现不出来，因为特斯拉需要验证所有这些车辆，它只有在开始建设自己的特斯拉车队网络时才真正需要这样做。
But the main takeaway is when that time comes for any company to achieve 100% of the market in San Francisco for the current ride-hailed demand. It would cost Waymo in the neighborhood of $700 million of CapEx, whereas it would only cost Tesla in the neighborhood of $100 million. Because again, during times of peak demand, Tesla can just open up the network to these customer-owned vehicles, where the CapEx is already accounted for.
主要的信息是，当任何公司在旧金山达到当前叫车需求的100%市场份额时，Waymo需要投入大约7亿美元的资本支出，而特斯拉只需大约1亿美元。这是因为在需求高峰期，特斯拉可以让那些顾客自有的车辆加入网络，而这部分的资本支出已经计算在内。
This slide shows new streets estimates have Tesla at $2.20 per mile without a safety driver today. But my pushback here would be anecdotally out in the real world, we've seen plenty of trip comparisons where Tesla is coming in at like 50% of what Uber and Waymo are offering. So if you kept Waymo and Uber the same using new streets research, that would have Tesla in the neighborhood of $1.25 per mile.
这张幻灯片显示，根据新的市场估计，特斯拉如今在没有安全驾驶员的情况下每英里成本为2.20美元。但我要反驳的是，根据我们在现实世界中的观察，很多旅行比较中，特斯拉的成本只有优步和Waymo的一半左右。所以，如果按照新的市场研究保持Waymo和优步不变，特斯拉的成本大约是每英里1.25美元。
And obviously to date, Tesla has done that with a safety driver, which is exactly why I think so far on such a small scale Tesla is fine taking a loss just including or writing off the cost of the safety monitor. And again, building their pricing algorithm as if they won't be there. The next slide, Tesla generates with FSD nearly 10 times more miles per quarter than all the miles accumulated by Waymo since inception.
显然，到目前为止，特斯拉都是使用安全员来实现这一目标。这也正是为什么我认为在这样小的规模上，特斯拉愿意接受损失，将安全员的成本计入或注销。特斯拉还在构建他们的定价算法，假设这些安全员最终不会存在。接下来的内容提到，特斯拉通过全自动驾驶功能（FSD）每季度产生的行驶里程几乎是Waymo自成立以来总行驶里程的10倍。
This slide also tells us what we've already known that Tesla has the potential to blitz past Waymo from a scale perspective, especially with CyberCab coming online. So by the end of 2027, Tesla could and theory have hundreds of thousands of robot taxis where Waymo is likely to be stuck under 50,000. NPR said there are good reasons to believe robot taxis will co-exist with ride sharing for a long time. And guess what, one of those reasons consumer trust and adoption remain uncertain. Surveys consistently show hesitation. According to triple A, six and ten Americans are afraid to ride in a self-driving vehicle. The share willing to ride in a robot taxi is rising but still low. 13% today, up from 9% last year.
这张幻灯片告诉我们，我们已经知道特斯拉有潜力在规模上迅速超越Waymo，尤其是在CyberCab上线后。因此，到2027年底，特斯拉理论上可能拥有数十万辆自动驾驶出租车，而Waymo可能还停留在不到5万辆的水平。美国国家公共电台（NPR）表示，有充分理由相信自动驾驶出租车将长期与共享出行共存。其中一个原因是消费者的信任和接受度仍不确定。调查数据显示，人们仍然持有犹豫态度。根据汽车协会（AAA）的数据，有六成美国人害怕乘坐自动驾驶汽车。愿意乘坐自动驾驶出租车的人虽然有所增加，但比例仍然较低。如今为13%，相比去年9%的比例有所上升。
So there you have it, I did leave out a few slides but I'll have the full report linked below. And look, don't give me wrong, I don't think modeling is useless. It can absolutely be a tool to think through potential outcomes and to play with numbers. But in an AI autonomous world where the uncertainties far outweigh the predictable outcomes, to me it just becomes more of an intellectual exercise than anything actionable. And as long as you have a long enough time frame and believe that Tesla 1 has a material cost advantage, 2, a safety advantage, and 3, a scaling advantage, then all you have to do is wait patiently for Tesla to deliver and for the market to then price all of that in.
所以这里就是这样，我确实省略了一些幻灯片，但我会在下面链接完整报告。听我说，我并不是说建模没有用。建模绝对可以用来思考潜在结果，并玩转数字。但是在一个不确定性远远超过可预测结果的人工智能自主世界中，对我来说，这更像是一种智力练习，而不是任何可操作的东西。而且，只要你有足够长的时间框架，并相信特斯拉在成本、安全和规模上都具有优势，那么你只需要耐心等待特斯拉交付，然后市场将会反映这一切。
MKBHD did review the new Model Y performance and it was pretty underwhelming in my opinion. He said Tesla has the best self-driving but he didn't demonstrate it at all or go any further than that. He didn't say anything about the continuously variable shock absorbers on the new performance and the adaptive suspension to contrast it with the non-performance models. He at one point said other cars are more fun because they have traditional buttons and it just seemed like someone forced him to do this video as if he was doing it against his will.
MKBHD对新的Model Y性能版进行了评测，但在我看来，这次评测表现得比较平淡。他提到特斯拉的自动驾驶是最好的，但并没有对其进行任何演示或进一步探讨。关于新性能版上配备的连续可变减震器和自适应悬挂，以对比非性能版车型，他也没有提到。他还提到有些车因为有传统按钮而更有乐趣，这让人感觉像是他被迫做这个视频，似乎是在违心完成任务。
He did praise Tesla for still having the best software but he caveated that with if you prefer something other than car play or android auto, implying those are still better than Tesla's software. So overall I was not impressed at all but I did want to touch on it if you want to watch it for yourself because over a million people will likely see this video but as I've said long ago he lost any Tesla credibility he had putting out bad info about wireless charging and reviewing outdated FSD software and still refusing to review the new FSD 14.
他确实称赞了特斯拉的软件仍然是最好的，但他也提到，如果你更喜欢CarPlay或Android Auto，那么这些仍然比特斯拉的软件更好。所以总体来说，我对他并没有太深的印象，但如果你有兴趣自己看看这个视频，我还是想提一下，因为可能会有超过一百万人观看。但正如我很久以前所说的，他已经失去了对特斯拉的信誉，因为他曾发布关于无线充电的不实信息，评论过时的FSD软件，并且仍然拒绝评测新的FSD 14版本。
So it is a shame because he has such a big audience but it might not get better because he also implied that Tesla is not talking to him and he had to get this car from a friend to review it. I hate to be the bearer of bad news but David Moss said the record setting FSD intervention free streak is over. His intervention free streak will end at 12,961 miles and what caused the intervention rural Wisconsin roads snow and weather in the teens was just a bit too much. He said no accident or anything big to note just found myself in a tricky situation where it was time to take over.
因此，这真是可惜，因为他有如此多的观众，但情况可能不会好转，因为他还暗示特斯拉并没有与他沟通，而他不得不向朋友借车来进行评测。我不想当坏消息的传递者，但大卫·莫斯说，他创造的“全自动驾驶（FSD）无干预驾驶”的记录已经结束。他的无干预记录停留在12,961英里，而导致此次干预的原因是威斯康星州乡村道路上的积雪以及个位数的低温，实在是太难应付了。他说没有事故或其他大问题，只是发现自己处于一个复杂的状况，不得不接管驾驶。
But FSD 14.2 took him to 30 states in 7.5 weeks of driving without any human input whatsoever so that remarkable streak has finally come to an end which just means that's the new bar and it's time for someone else to go beat it. And I do hope Tesla does something along the lines of the odometer with Tripe A and B so you can reset your stats anytime you want to maybe before a trip or to go after a record and so on. So remember years ago when Elon talked about this trip going from LA to New York City that's about 2,500 miles.
FSD 14.2 在7.5周的驾驶中，毫无人工干预地带他去了30个州，这一令人惊叹的记录终于结束了。这意味着这是新的标杆，接下来就看谁能打破它了。我希望特斯拉能像设计A和B档里程表那样，也给我们一个可以随时重置统计数据的选项，比如在出发前或挑战记录前重置。还记得几年前Elon谈到的那次从洛杉矶到纽约的旅行吗？这段距离大约是2,500英里。
What that means is David just did that trip effectively 5 times without intervening once. And I don't care if it's mostly highway driving because he still had to get off to park in supercharge so effectively going across the country 5 times without intervening is incredible. So look I know we still have things we need to get sorted in the parking lots and especially with navigation. But I really just hope everybody can stop and appreciate this achievement because it really is a representation of where we are with this technology today and how far we've come in just the last few years.
这意味着大卫实际上完成了5次这样的旅行，没有进行任何干预。我不在乎这大部分是高速公路驾驶，因为他仍然需要停下来在超级充电站充电，所以没有干预地穿越国家5次是非常了不起的。所以，我知道我们在停车场和导航方面还有一些问题需要解决。但我真的希望每个人都能停下来欣赏这一成就，因为这确实代表了我们今天所拥有的技术水平，以及在过去几年中我们取得的巨大进步。
This account I'll call Genma JP has been putting out some very interesting articles as of late on autonomy. Sadly it's an anonymous account so I can't confirm this person is who they're saying but unless they're excellent at prompting AI it really does seem like they're actually working in the industry. Assuming it's a heat for over 15 years I've been an expert in sensor fusion for ADAS. He said I've personally developed an integrated ultrasonic sensors radars worked with automotive light hours in production and fused data from surround view cameras as well as third party computer vision systems.
这个我称之为Genma JP的账号最近发布了一些非常有趣的关于自动驾驶的文章。遗憾的是，这个账号是匿名的，所以我无法确认这个人是否如他所说的那样，但除非他非常擅长引导AI，否则他看起来确实在这个行业工作。据推测该账号的主人在这个行业已经有超过15年的经验，是ADAS（高级驾驶辅助系统）中的传感器融合专家。他说自己亲自开发了整合超声波传感器雷达，与汽车激光雷达合作生产，并融合了全景摄像头以及第三方计算机视觉系统的数据。
In the world of classical ADAS I've seen and built it all and yet I've put every spare euro I own into Tesla shares. Why? Because after a career spent perfecting multi sensor fusion I'm now convinced it's fundamentally the wrong path to full self driving. What follows is not armchair speculation it's the conclusion of someone whose professional life has been dedicated to making fusion work in the real world. Someone who knows the strengths shortcuts and ultimate limitations. He said after years of building I've concluded adding radars light hours and other sensors to cameras does not meaningfully advance us toward FSD in many ways it impedes progress and I'll just share a few key points. He said radar and light are primarily provide depth and relative velocity. Data that modern neural networks can already derive sufficiently from camera images alone.
在传统高级驾驶辅助系统领域，我曾见过并参与过所有类型的开发，但我却把所有闲置的资金都投入了特斯拉的股票。为什么？因为在从事多传感器融合技术工作的职业生涯后，我现在确信这是实现完全自动驾驶的根本错误路径。接下来我要说的并非是空谈，而是一个把职业生涯奉献给让传感器融合在现实世界中发挥作用的人得出的结论。一个熟知其优势、捷径和最终局限性的人说道，经过多年的研发，我得出的结论是，给摄像头添加雷达、激光雷达和其他传感器实际上并没有实质性地推动我们迈向完全自动驾驶，反而在很多方面阻碍了进展。我要分享几点关键看法。他指出，雷达和激光雷达主要是提供深度和相对速度这类数据，而现代神经网络已经能单靠摄像头图像充分推导出这些信息了。
Light hours fundamental weaknesses it performs poorly in rain fog and on reflective surfaces and lacks the angular resolution for reliable classification at distance. Radar's limitations, despite better weather penetration it delivers extremely sparse detections, suffers from clustering and classification challenges and often masks weaker objects behind stronger reflectors particularly problematic for static infrastructure in low speed scenarios. Radar and light are cannot detect critical semantic info like traffic signs, lights, lane markings or pedestrian intent cues. Stellar computer vision is mandatory anyway the other sensors cannot compensate for its absence. Cameras are robust enough modern images match or exceed human eye performance and things like wipers and airflow can handle issues like rain drops.
翻译成中文，并尽量使其易于理解：
光传感器的基本弱点在于它在雨天、雾天以及反光表面上的表现不佳，缺乏可靠的远距离分类所需的角度分辨率。尽管雷达在恶劣天气下的穿透力更好，但其检测结果非常稀疏，容易受到聚类和分类方面的挑战影响，且经常在强反射物体后隐藏较弱的物体，这在低速情境中的静态基础设施中尤其成问题。雷达和光传感器无法检测到关键的语义信息，比如交通标志、信号灯、车道标记或行人意图。卓越的计算机视觉是必不可少的，因为其他传感器无法弥补其缺失。相机足够强大，现代图像的表现可以媲美甚至超过人类眼睛，而且像雨刷和气流这样的工具可以解决诸如雨滴之类的问题。
In truly degraded visibility the safe response is to slow down something an AV can do systematically and despite the camera heaters and everything testless come out with I still have problems where I am at with dirt and road salt in these snowy conditions. But to put it into context for this entire winter season for me I've maybe had the red wheel three or four times that I can remember. And if you only remember one thing from this let it be this right here fusion is a crutch. Multi-sensor approaches deliver quick early winds by patching vision weaknesses but they mask the need for true mastery of computer vision through massive data and compute. Companies end up over investing in complex fusion logic instead of solving the hard problem.
在能见度极差的情况下，安全的做法是减速，这一点自动驾驶汽车可以系统地做到。尽管有摄像头加热器等设备帮助，我在雪天的脏污和路盐环境下仍然遇到问题。但为了让大家了解整个冬季，我记得可能只有三四次遇到了红色警报。而如果你只能记住一件事，那就是：传感器融合只是权宜之计。多传感器方法通过弥补视觉弱点快速取得早期胜利，但它掩盖了通过大量数据和计算实现计算机视觉真正掌握的需求。公司最终会过多地投资于复杂的融合逻辑，而不是解决这个棘手问题。
You might need to double tap on the left of your screen and go back and listen to that a few times and really commit it to memory. And this is exactly what I mean when I say that these other companies are likely to get stuck in local maximums. It is likely to be faster out of the gate but in the end it is not going to be the path that takes you to generalize autonomy. So even if it's true that solving the sensor fusion problem continually gets easier thanks to the advancements with neural networks like what RIVian is deciding to do that still does not mean it's the optimal long term approach. So I thought that was worth sharing because we so often hear from people outside the industry and yes I can read studies and listen to podcasts and pick up cues from a show in Yun Tassai and Elon over the years but I'm not in the trenches.
你可能需要在屏幕左侧点击两次，然后返回去多听几遍，这样才能真正把它记住。这正是我所说的，其他公司可能会陷入局部最大值的问题。虽然它们可能会一开始跑得更快，但最终这条路径并不会带你走向广泛的自动化。因此，即便解决传感器融合问题随着神经网络技术的进步变得越来越容易，例如RIVian正在做的事情，但这并不意味着这是最优的长期方法。我觉得值得分享这个观点，因为我们经常听到来自行业外的人的看法。是的，我可以阅读研究文章、听播客，并从多年来关于Elon等的节目中获取信息，但我毕竟不是身处一线的人。
I'm not working on these technical challenges day in and day out. So to hear from someone that presumably is to say what we've been saying here on the channel for years I really do think validates the arguments we've been making and not that it was needed but I do think it's encouraging especially at a time when some people are a bit more uncertain with Nvidia developing alpomayo. So it's likely to be a noisy race with a lot up for debate for another six to twelve months but it's not how you start it's how you finish and this is just another data point that the winner has already been chosen it's just a matter of it playing out and then the world realizing it and ultimately assigning a value to it.
我并不是每天都在处理这些技术挑战。因此，听到像似乎专注于这个领域的人表达与我们在频道中多年来所说的一致的观点，我确实认为这证明了我们一直以来的论点。尽管我们并不需要更多的验证，但在Nvidia研发Alpomayo时，一些人感到不确定的情况下，这样的支持确实很鼓舞人心。未来六到十二个月可能还是一个争论激烈、竞争激烈的阶段，但重要的不在于起点，而在于终点。这只是另一个数据点，表明赢家已然显现，只是时间的问题，直到全世界意识到并赋予其相应的价值。
Canada and China have struck an initial trade deal that will slash tariffs on EVs and canola and some other agricultural items. This coming from Canadian Prime Minister Mark Carney. Canada will initially allow in up to 49,000 Chinese EVs at a tariff rate of 6.1 percent which yes is coming down from the prior 100% tariffs on Chinese EVs but I think it's important context that this really is just returning to the previous tariff rate before some of these trade wars between Canada and China ensued. So it's not like this is the first time Chinese EVs will be entering Canada remember Tesla shipped cars from Shanghai into Canada for some time.
加拿大和中国达成了一项初步贸易协议，将大幅削减电动汽车、油菜籽和一些其他农产品的关税。此消息来自加拿大总理马克·卡尼。根据协议，加拿大将允许最多49,000辆中国电动汽车进入，关税税率为6.1%。虽然比之前对中国电动汽车征收的100%关税有所降低，但实际上这只是恢复到两国贸易战前的关税水平。因此，这并不意味着中国电动汽车首次进入加拿大市场，之前特斯拉就曾经从上海向加拿大出口汽车。
However the difference is how good Chinese EVs have gotten in the past few years. Carney said for Canada to build its own competitive EV sector we need to learn from innovative partners access their supply chains and increase local demand. The problem with that is how are these Chinese companies making EVs going to be partners if they're just selling their products to your market. They're not setting up auto factories in Canada and I certainly don't think Canada will be getting access to the Chinese IP of its EVs. But for what it's worth, Carney said he expected the EV pact would drive considerable Chinese investment into Canada's auto sector.
然而，区别在于中国的电动汽车在过去几年里取得了多大进步。卡尼表示，如果加拿大想要建立自己的竞争力电动车产业，我们需要向创新的合作伙伴学习，进入他们的供应链，并增加本地需求。问题在于，如果这些中国公司只是把产品卖到你的市场，它们怎么会成为合作伙伴呢？它们并没有在加拿大设立汽车工厂，而且我确实不认为加拿大能获得中国电动车的知识产权。不过，据卡尼所说，他预计这项电动车协议将推动中国在加拿大汽车行业的大量投资。
And again in return Canada is expecting China to lower tariffs on certain agricultural products but a lot of that is outside of the scope here. There's been very mixed feedback to this news. US trade representative James and Greer said there's a reason we don't sell a lot of Chinese cars in the US. It's because we have tariffs to protect American auto workers and Americans from those vehicles. He said I think in the long run they're not going to like having made that deal. And they're saying that 49,000 annual EV limit will go up by about 6% every year until it hits 70,000 vehicles in about 5 years.
加拿大期望中国降低某些农产品的关税，但这方面超出了这里的话题范围。对此消息的反馈不一。美国贸易代表詹姆斯和格里尔表示，我们在美国不怎么卖中国汽车是有原因的，因为我们设有关税以保护美国汽车工人和消费者。他说，我认为从长远来看，他们不会喜欢达成这样的协议。据说每年49,000辆电动车的限额将在大约5年内每年增加约6%，直到达到70,000辆。
And any imports exceeding that quota would still face the full 100% tariff. So I really don't want to step out of bounds and speak as a Canadian here but for my Canadian friends I would love to hear your perspective on all of this. Of course as a consumer I think it's safe to say this would be a win having access to some high-quality EVs that are quite affordable. And over time it should help to motivate the local Canadian auto industry to innovate and try to compete. And it's true that the Canadian auto sector is not as important to overall GDP as it is in the United States. I believe in Canada the auto industry is under 1% of total GDP whereas in the US it can fluctuate between 3 and 5%.
超出配额的任何进口仍将面临100%的关税。所以，我并不想越界，以一个加拿大人的身份发表意见，但我很想听听我加拿大朋友们对此事的看法。作为消费者，我觉得可以说这是一个好消息，因为这样可以获得一些高质量且价格合理的电动车。从长远来看，这也应该能激励加拿大本地汽车行业进行创新和竞争。确实，加拿大汽车行业在整体GDP中的重要性不如美国。我相信在加拿大，汽车行业占总GDP的比例不到1%，而在美国，这一比例可以在3%到5%之间波动。
And then we have Doug Ford the premier of Ontario said make no mistake China now has a foothold in the Canadian market and will use it to their full advantage at the expense of Canadian workers. The federal government is inviting a flood of cheap made in China EVs without any real guarantee of equal or immediate investments in Canada's economy. And I really don't think this will change too much for the United States. Trump has stayed steady he says that the Chinese are welcome to set up factories in the US if they hire American workers. But for the most part no one's really taken them up on that offer.
然后我们看到安大略省省长道格·福特表示，不要误会，中国现在已经在加拿大市场站稳脚跟，并将充分利用这一点，牺牲加拿大工人的利益。联邦政府正在引进大量廉价的中国制造电动汽车，却没有真正保证对加拿大经济的平等或立即的投资。我觉得这对美国的影响不会太大。特朗普一直态度明确，他表示欢迎中国在美国建立工厂，只要他们雇用美国工人。然而，大多数情况下，没有人真正接受这个提议。
And with the back and forth in the US political landscape I mean can we really blame them. So just like most things this deal comes with trade-offs but I know the Chinese market is huge for Canadian farmers and the lower tariffs in the agricultural sector are very important and who knows maybe more of a partnership between Canada and China does actually materialize and the consumers win and the Canadian auto industry gets help innovating or the Canadian auto industry gets out competed over the next 5 plus years and the domestic auto sector collapses in Canada. Truthfully I could see either outcome but I'd rather hear from you Canadians actually living there in the comments.
鉴于美国政治局势的反复无常，我觉得我们不能真的责怪他们。因此，就像大多数事情一样，这项协议也有其权衡之处。但我知道，中国市场对于加拿大农民来说非常重要，降低农业领域的关税也非常关键。也许，加拿大和中国之间会形成更多的合作，消费者会因此受益，加拿大汽车产业得到创新帮助；或者，加拿大汽车产业在未来5年多时间里被竞争对手超越，导致国内汽车行业崩溃。说实话，我认为这两种结果都有可能发生，但我更想听听在加拿大生活的本地人的看法，请在评论中分享你的观点。
I'm sure most of you heard J. Cal's comments about Optimus V3 already but in case you haven't. Yeah I don't want to name drop but I went two weeks two Sundays ago I went to Tesla with Elon and I went and visited the Optimus Lab. There were a large number of people working on a Sunday at 10am and I saw Optimus 3. I could tell you now nobody will remember that Tesla ever made a car. They will only remember the Optimus and that he is going to make a billion of those and it is going to be the most transformative technology product ever made in the history of humanity because what LLM is going to enable those products to do is understand the world and then do things in the world that we don't want to do.
我确信你们大多数人已经听过J. Cal对Optimus V3的评论，不过如果你还没听过的话。我不想故意炫耀名字，但两个星期前的星期天，我和Elon去了趟特斯拉，参观了Optimus实验室。周日上午10点，我看到有很多人在那里工作，并看到了Optimus 3。我可以告诉你，未来没有人会记得特斯拉曾经制造过汽车。他们只会记得Optimus，以及它将成为人类历史上最具变革性的技术产品之一。因为大型语言模型（LLM）将赋予这些产品理解世界的能力，然后去做一些我们不想做的事情。
I believe there will be a one-to-one ratio of humans to Optimus and I think he's already one. So there you go. J. Cal clearly saw something special and he's declaring Tesla has already won. Then Elon posted feeling optimistic about the future. Spelled like Optimus and shared this AI generated video. The fact that so many people did not know this was an AI generated video is already alarming and makes my point that we're headed to a very dangerous world. But the first frame could actually be a real photo that was used to make that video.
我相信，将来人类和Optimus机器人会达到一对一的比例，而且我觉得他已经是其中之一。所以，你懂的。J. Cal显然看到了某些特别之处，并宣称特斯拉已经赢了。然后，Elon发布了一个信息，表达了对未来的乐观态度，并分享了这个由AI生成的视频。很多人不知道这是AI生成的视频，这一点令人担忧，也证明了我所说的我们正走向一个非常危险的世界。不过，这个视频的第一帧可能确实是真实照片，用来制作这个视频的。
But listen the truth is even if that's the case we have no idea what this is we don't know what we're looking at. There's some sort of concealing clothing on this Optimus spot. The arm around Elon doesn't look anything like a robot arm. It literally looks like a pool noodle covered in a black sleeve and this gold colored hand or glove or whatever we're looking at. I could sit here and speculate for the next 20 minutes but the truth is I really don't know what we're looking at here.
但是听我说，事实是，即使情况是这样，我们也不知道这是什么，我们不知道自己在看什么。这个Optimus上的某种遮盖物让我们难以看清。抱着Elon的那只手臂看起来一点也不像机器人的手臂。它看起来就像一个穿着黑色袖子的泳池漂浮条，而我们看到的这个金色的手或手套，更是让人摸不着头脑。我可以坐在这里猜测20分钟，但事实是我真的不知道我们在看什么。
But long before J. Cal said what he said I've been feeling that it's felt eerily quiet in Tesla land as of late both on the autonomy front and when it comes to Optimus. So for me this is the calm before the storm. And whenever Tesla comes out with its Optimus V3 demo I do think it's going to be a loud reminder that no one is on Tesla's level when it comes to building a generalized humanoid bot.
早在J. Cal发表言论之前，我就一直觉得最近特斯拉领域有些诡异的安静，无论是在自动驾驶方面，还是在Optimus项目上。所以对我来说，这种安静就像是暴风雨前的宁静。我认为，特斯拉一旦推出Optimus V3演示，将会响亮地提醒大家，在制造通用人形机器人方面，没有人能与之匹敌。
Because just like the sensor fusion problem much of the competition is coming out with these hand designs that can start doing a bunch of different jobs. But companies like Boston Dynamics and Figure and all of these Chinese companies are not gearing and preparing to do things like surgery. And while I think Elon's timelines are way off for that the point is the engineering team has that end goal in mind. That's what they're optimizing for the long tail. They're not racing the market. They're not racing to do basic batteries sorting in the factory. They're building a humanoid bot that is generalized in time capable of doing everything humans can do.
因为就像传感器融合问题一样，许多竞争对手正在推出能够开始处理各种不同任务的手部设计。但是像波士顿动力公司、Figure和许多中国公司并没有在朝着做手术这样的事情做准备。而我认为埃隆的时间表对于这些事情来说非常不准确，但重点是工程团队有这样一个最终目标。那是他们为实现长远目标而优化的方向。 他们并没有急于抢占市场，也没有急于在工厂里进行基础的电池分类。他们正在构建一个类人机器人，其目标是最终能够在某个时间点完成所有人类可以做的事情。
And yes I do think Tesla is going to keep everything close to the chest from here on out until they get into production. So even with the demo they might be concealing the hand they might be concealing the body. And I'll sound like Candace Owens here but I just have this feeling that Tesla is going to come out and show us something that makes it nearly impossible not to overreact.
是的，我确实认为特斯拉从现在开始会更加保密，直到他们开始生产。所以即使在演示中，他们可能也不会完全展示他们的计划或者所有细节。虽然这样说可能有点像坎迪斯·欧文斯，但我有一种感觉，就是特斯拉会推出一些让我们无法不大肆反应的东西。
And I just wanted to clarify there's a bunch of people in articles out there saying that Elon's predicting 50 to 100,000 optimists produced this year. As if he made that statement recently when the truth is he actually said that January 8th of 2025 so it was over a year ago. And of course much has changed since then. Which is why personally I'm not expecting anything close to that this year.
我想澄清一下，有很多文章和人声称埃隆预测今年将生产5万到10万台Optimus机器人，似乎是他最近做出的声明。但实际上，他是在2025年1月8日说的这番话，那已经是一年多以前的事情了。当然，自那以后发生了很多变化。这也是为什么我个人并不指望今年能达到那个数字。
The 2025 S&P Global Loyalty Awards came out. Last year we had again GM and Tesla earning the two highest awards for customer retention. Tesla was the recipient of the overall Loyalty to Make award for the fourth consecutive year. These awards are based on 13.6 million new retail vehicle registrations in the US from October 2024 to September 2025. Tesla's multiple awards highlight a strategy of targeted market appeal and consistent disruption. Its win for ethnic market loyalty to Make was driven by remarkable retention among Asian and Hispanic households at 64 and 62% respectively.
2025年标普全球忠诚度奖公布了。去年，通用汽车和特斯拉再次获得了客户保留率的两项最高奖项。特斯拉连续第四年获得了“整体品牌忠诚度奖”。这些奖项基于美国从2024年10月至2025年9月期间的1360万份新零售车辆注册数据。特斯拉获得多项奖项，彰显了其精准的市场吸引力和一贯的创新策略。在归属品牌忠诚度方面，特斯拉在亚裔和西班牙裔家庭中的保留率分别达到了64%和62%，这也是其获奖的关键因素。
Rates that significantly outpaced the national averages for those groups. Tesla also secured the highest conquest percentage award for the sixth consecutive year. And boiling it down that just means Tesla is continually winning new buyers from other automakers. So of the seven overall Loyalty awards, Tesla is the winner of three of them.
这个句子翻译成中文是：“这些增长率远远超过了这些群体的全国平均水平。特斯拉还连续第六年获得了最高客户转化率奖。这基本上意味着特斯拉不断地吸引其他汽车制造商的顾客。因此，在七个综合忠诚度奖项中，特斯拉获得了其中的三个。”
In a new drone fly over in Austin, Joe Tetmire saw these model wise in the parking lot with covers on the wheels. And these large sticker looking things on the trunk where typically the performance badge would be. In the video you'll see one of the cars had the coverings fall off so it looks like that could be the new gray helix wheels. And for these new stickers on the back, I don't have a confident view but figured I'd pass it along.
在奥斯汀的一次无人机飞行中，乔·泰特迈尔看到停车场里的这些车型，它们的车轮上盖着罩子。车子的后备箱上通常放置性能徽章的地方贴着大块贴纸。在视频中，你会看到其中一辆车的罩子掉了下来，所以看起来这可能是新的灰色螺旋轮子。至于这些新贴纸，我没有很确定的看法，但还是想和大家分享一下。
Nitsa just granted Tesla a five week extension to respond to an investigation into whether its cars violated traffic laws went on FSD. The new deadline is February 23 after Tesla requested more time to manually review thousands of records to identify incidents that may be relevant. And Tesla just said it has 8,313 records that remained to be reviewed and it could process about 300 per day.
NHTSA刚刚同意给特斯拉五周的延期，回应关于其汽车在启用完全自动驾驶（FSD）时是否违反交通法规的调查。由于特斯拉请求更多时间手动审查数千条记录以确定可能相关的事件，新截止日期定为2月23日。特斯拉表示，目前还有8,313条记录需要审查，他们每天可以处理大约300条。
Tesla also cited the burden of responding simultaneously to multiple Nitsa investigations, including separate probes into delayed crash reporting and inoperative door handles, saying the volume of requests could affect the quality of its responses. And who set fire to a Tesla cyber truck and dealership in Mesa last spring will spend five years behind bars.
特斯拉还提到，同时应对多个国家公路交通安全管理局（NHTSA）调查的负担，包括对延迟报告事故和车门把手失灵的独立调查，表示请求的数量可能会影响其回应的质量。而去年春天在梅萨市纵火焚烧特斯拉Cybertruck和一家经销店的人将被判五年监禁。
In April, 35 year old Ian William Moses put fire starter logs along the outside wall of the Tesla dealership that Southern and Saucerman flames destroyed a cyber truck and damaged part of the building. He also spray painted the word thief as you see there misspelling it. US Attorney General Pam Bondy called the arson in act of domestic terrorism. So nice to see some justice being served and just FYI no domestic terrorism charges were actually filed.
在四月，35岁的伊恩·威廉·摩西（Ian William Moses）在一特斯拉（Tesla）经销店的外墙放置了引火木，Southern和Saucerman的火焰烧毁了一辆Cybertruck，并损坏了建筑的一部分。他还在墙上喷涂了“thief”这个词，不过拼错了。美国司法部长帕姆·邦迪（Pam Bondy）称这一纵火事件为本土恐怖主义行为。不过，实际上并没有提出本土恐怖主义的指控。很高兴看到有正义得以伸张。
The other day we got Euro NCAP results and today we get NCAP for Australia and New Zealand. They said the Model Y achieved the highest overall weighted score of any vehicle assessed by ANN CAP in 2025. Recording strong performance across all areas of occupant protection and active safety tech. This is the second time the Model Y has taken the lead position. And the Model 3 was right there alongside the Model Y 90% adult 95 child 89% VRUs and 88% safety assist. So nothing we don't already know but every time it's still great to hear and have it reiterated with new versions. The Model Y the safest vehicle tested by ANN CAP for all of 2025.
那天我们得到了欧洲NCAP的结果，今天我们看到了澳大利亚和新西兰的NCAP结果。他们表示，Model Y在2025年获得了由ANN CAP评估的所有车辆中总体评分最高的成绩。它在乘员保护和主动安全技术的各个领域都表现出色。这是Model Y第二次取得领先地位。Model 3的表现也很出色，成人保护评分为90%，儿童保护评分为95%，弱势道路使用者保护评分为89%，安全辅助评分为88%。虽然这些我们之前已经知道了，但每次听到这样的好消息，尤其是与新版本一起重申时，依然令人欣慰。Model Y是2025年由ANN CAP测试过的所有车辆中最安全的。
Tesla is now allowing people to reserve the Sky Pad the upper level balcony area at the Tesla diner. So if you want to host an event request must be submitted a minimum of two weeks before the date. And the Tesla charging account just shared that a fourth business has decided to white label Tesla superchargers this time it's Wawa in Alachua, Florida. And that did not sound right coming off the tongue so I looked it up it's Alachua. And there are 16 stalls at this location.
特斯拉现在允许人们预订特斯拉餐厅的Sky Pad，这是一处上层阳台区域。所以，如果你想在这里举办活动，需至少提前两周提交申请。此外，特斯拉充电账户刚刚分享了一则消息：又有一家企业决定使用特斯拉的超级充电桩，这次是在佛罗里达州的阿拉楚瓦（Alachua）的Wawa。这名字念起来有点怪，所以我查了一下，是Alachua。这个地点有16个充电位。
Tesla stock closed the day at $437 and 50 cents down 0.24% while the NDX was down 0.07%. The volume was 14% below the average. Hope you all have a safe and wonderful weekend and a huge thank you to all of my Patreon supporters.
特斯拉股票当天收盘价为437美元50美分，下跌了0.24%，而纳斯达克100指数下跌了0.07%。成交量比平均水平低14%。希望大家度过一个安全美好的周末，非常感谢我的所有Patreon支持者。