$NVDA · Benzinga · 2026-11-19
英伟达是机器人领域的军火商,向几乎所有人出售芯片、模型和仿真,但说话者将今天的NVDA定位为一家数据中心公司,附带一张非常昂贵的机器人彩票:物理AI占收入不到3%,近90%来自数据中心AI芯片。只有当人形机器人普及、这3%成为下一个增长引擎时才成立;如果物理AI再当五年科学项目,买家就是花一万亿美元买一个故事。特斯拉自己的大脑和中国国产芯片可能比乐观者预期更快地削弱其掌控力。
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
逐段跳播
Nvidia。它是机器人战争中的军火商。它不在乎谁赢,因为它向各方出售武器,而且市值已经超过4万亿美元。地球上最有价值的公司之一。
Nvidia. This is the arms dealer in the robot war. It doesn't care who wins because it sells weapons to all sides, and is already worth more than $4 trillion. One of the most valuable companies on the planet.
在上一集里,你看到特斯拉用驱动其汽车的同一种人工智能来构建自己的内部智能,但特斯拉是例外,而不是常态。从零开始构建自己的机器人大脑既困难又昂贵,以至于地球上几乎所有其他机器人公司都做出了同样的决定。
它们不自己造大脑,而是买。有一家无可争议的领先公司,几乎所有人都求助于它。Nvidia 不卖给你机器人,它卖给你机器人思考所需的一切,而且分为三层。
In the last episode, you saw that Tesla is building its own internal intelligence using the same artificial intelligence that drives its cars, but Tesla is the exception, not the rule.
Building your own robot brain from scratch is so difficult and expensive that virtually every other robotics company on Earth has made the same decision. They don't build the mind, they buy it.
There is an undisputed leading company that almost everyone turns to . Nvidia doesn't sell you a robot, it sells you everything a robot needs to think, and it comes in three layers.
第一,物理大脑,也就是一块名为“Jetson Thor”的芯片。可以把它想象成一台被缩小到能装进机器人胸腔的数据中心超级计算机。它刚刚开始出货,它为机器人提供了足够的实时 AI 处理能力,让它在移动中理解世界。
第二,大脑本身。Nvidia 正在构建名为“Isaac Grot”的人工智能模型,让人形机器人能够学习新技能,而不是为每项任务单独进行人工编程。第三,这也是秘密武器,Nvidia 正在构建整个虚拟世界,让机器人在其中训练。
First, the physical mind , which is a chip called "Jetson Thor". Think of it as a data center supercomputer that has been shrunk down to fit inside a robot's chest. It has just started shipping , and it gives the robot enough real-time AI processing power to understand the world as it moves.
Secondly, the mind itself. Nvidia is building artificial intelligence models called " Isaac Grot" that allow the humanoid robot to learn new skills instead of being manually programmed for each task individually.
Third, and this is the secret weapon, Nvidia is building entire virtual worlds where robots train.
这就是为什么这一点很重要。只在现实世界中训练的机器人学得非常慢,一次只能笨拙地迈出一步。但在 Nvidia 的模拟器里,你可以同时运行同一机器人的数千个副本,而且速度超越现实,让它们在几天内学会在真实工厂车间里需要数年才能学会的东西。然后这些训练成果被转移到真实的机器上。
Here's why that's important. A robot that is only trained in the real world learns very slowly, one clumsy step at a time. But within the Nvidia simulator, you can run thousands of copies of the same robot at once and at speeds that surpass reality, and let them learn in days what would have taken years on the actual factory floor.
Then that training is transferred to the real machine.
Nvidia 不是在造单个机器人,而是在为所有人建造一座大脑工厂。这证明了这套东西确实行得通。看看谁在基于 Nvidia 的大脑进行构建。Boston Dynamics 把“Jetson Thor”芯片装进了“Atlas”机器人。
Figure,世界上最令人兴奋的新兴机器人公司之一,依赖它。Agility Robotics,其机器人已经在亚马逊仓库内运行,也依赖它们。亚马逊自己、Caterpillar、Meta,甚至“OpenAI”都在名单上。
再读一遍这份名单。这意味着几乎所有人。Nvidia 不必去猜哪款人形机器人会赢。所以无论谁赢,大概率都要归功于 Nvidia。
Nvidia is not building a single robot, but rather a brain factory for everyone. This is proof that this actually works. Look at who is building on the mind of Nvidia. Boston Dynamics places the "Jetson Thor" chip inside the "Atlas" robot.
Figur, one of the world's most exciting emerging robotics companies, relies on it. Agility Robotics, whose robots already operate inside Amazon warehouses, relies on them. Amazon itself, Caterpillar, Meta, and even "OpenAI" are on the list.
Read that list again. This means almost everyone. Nvidia doesn't have to guess which humanoid robot will win. So whoever wins, it's most likely thanks to Nvidia.
这就是基本思路。她是赌场的主人,而在这家赌场里,主人向每位玩家出售筹码。如果这个说法听起来耳熟,那是因为 Nvidia 以前就这么做过。它已经在数据中心的人工智能大脑领域近乎垄断。
仅凭这项业务,Nvidia 的市值就达到了数万亿美元。大约90%的收入来自它。现在它沿用同样的策略,并直接将其指向物理世界。
This is the basic idea. She is the owner of the casino, and in this casino the owner sells chips to each player. If this approach sounds familiar, it's because Nvidia has used it before.
It already has a near-monopoly on artificial intelligence minds within data centers. That work alone is the reason Nvidia's value has reached trillions of dollars. Approximately 90% of its revenue comes from it.
Now it follows the same strategy and directs it directly towards the physical world.
这就是让与它竞争变得极其困难的部分。这从来就不只是芯片的问题。多年来,Nvidia 让整整一代 AI 工程师依赖它的软件,一套名为“CODA”的工具。一旦一个机器人团队把整个系统建立在 Nvidia 的工具之上,转向竞争对手就意味着重写多年的工作。
这与让 Nvidia 在数据中心领域不可战胜的是同一种“绑定”策略,而它现在正在机器人领域扎根。
This is the part that makes competing with it extremely difficult. It was never just about chips. Over the years, Nvidia has made an entire generation of AI engineers rely on its software, a suite of tools called "CODA" .
Once a robotics team has built its entire system on Nvidia tools, switching to a competitor means rewriting years of work. This is the same "bondage" approach that made Nvidia invincible in data centers, and it is now taking root in the world of robotics.
Nvidia 首席执行官 Jensen Huang 把即将到来的这波浪潮称为“物理 AI”,并说机器人的“chat GPT”时刻已经到来。长期回报估计约为40万亿美元。这不只是空谈。
Huang 说,物理 AI 业务已经每年产生约100亿美元的收入,并有望达到约1000亿美元,而且他最近把丰田、发那科以及一批日本最大的工业巨头加入了 Nvidia 的机器人联盟。
Nvidia CEO Jensen Huang calls this coming wave " physical AI" and says the "chat GPT" moment for robots has already happened. The long-term prize is estimated at around $40 trillion.
This is not just talk. Huang says that the physical AI business is already generating about $10 billion a year with a trajectory of about $100 billion, and he recently added Toyota, Fanuc and a host of Japan’s biggest industrial giants to Nvidia’s robotics alliance.
但我想向你抛出一些同样重要的事实,因为这里的悲观视角确实很重要。如今,这些机器人大脑的业务对 Nvidia 来说太小了。物理人工智能占其收入不到3%。而近90%仍然来自向人工智能数据中心出售芯片。
But I want to throw at you some equally important factual facts , because the pessimistic viewpoint here is really important. Today, the work of these robot brains is too small for Nvidia.
Physical artificial intelligence accounts for less than 3% of its revenue. And nearly 90% still comes from selling chips to artificial intelligence data centers.
所以,让我直说吧。你现在买入 Nvidia 股票,买的不是一家机器人公司。你买的是一家数据中心公司,附带一张机器人世界的彩票。而且这张彩票并不便宜。Nvidia 已经是世界上最昂贵的公司之一,其价格反映的是一个几乎完美契合的未来。
So, let me be frank. When you buy Nvidia shares now, you are not buying a robotics company. You are buying a data center company with a lottery ticket in the robotics world attached.
And it's not a cheap ticket. Nvidia is already one of the most expensive companies in the world, and its price reflects a future that is almost perfectly aligned.
唯一自己造大脑的巨头,也就是“特斯拉”,完全不需要 Nvidia。中国的许多机器人制造商也是如此,面对美国的出口限制,它们越来越多地转向自己的国产芯片。如果更多大玩家认定大脑太重要、不能外包,并开始设计自己的芯片,Nvidia 对这个市场的掌控可能会比乐观者预期的更快松动。
The only giant that builds its own brain, which is "Tesla," does not need Nvidia at all. The same is true for many robot manufacturers in China who are increasingly turning to their own domestic chips in light of US export restrictions.
If more big players decide that brains are too important to be hired and start designing their own chips , Nvidia’s grip on this market could weaken faster than optimists expect.
那个40万亿美元的数字是对遥远未来的预测,而不是今年的实际收入。那么,你究竟应该怎样看待这件事?说实话,归根结底就是一个问题。你认为军火商总是赢家吗?如果人形机器人普及开来,哪怕只达到 Huang 预测的一小部分,无论哪家公司在机器人领域胜出,Nvidia 都会卖出大脑。
那区区3%会成为它下一轮大增长的引擎。但如果物理人工智能在未来五年里仍然只是一个科研项目,那你今天就是在为一个关于明天的故事支付上万亿美元。
That figure of $40 trillion is a projection for the distant future, not actual income for this year. So , how should you really think about this ? Honestly, it boils down to one question.
Do you think the arms dealer always wins? If humanoid robots become widespread, even to a fraction of what Huang predicts, Nvidia will sell brains regardless of which company excels in robotics.
That tiny 3% becomes the engine of its next great growth . But if physical artificial intelligence remains just a scientific project for another five years, you are paying a trillion dollars today for a story about tomorrow.
那么,你真正应该留意什么?Nvidia 将于8月26日公布财报。这一次,不要只盯着大家都在看的数据中心数字。听听他们怎么说机器人和物理人工智能;那才是机器人大脑板块是否终于开始复苏的真正指标。
还要关注“Jetson Thor”实际出现在工厂车间里正在工作的机器人内部的速度有多快。
So, what should you really be watching out for ? Nvidia will announce its earnings on August 26 . This time, don't just stare at the data center numbers that everyone is watching.
Listen to what they say about robots and physical artificial intelligence; that is the true indicator of whether the robot brains sector is finally starting to recover. And watch how quickly " Jetson Thor" actually appears inside working robots on the factory floor.
主导这一领域的是 Nvidia。巧妙之处在于,Nvidia 其实根本不是在造机器人。它把“大脑”卖给所有造机器人的人:装在机器人内部的芯片、用来训练它的软件,以及一个完全模拟的世界——机器人在真正动起来之前,先在那里被训练数百万次。
The company that dominates this segment is Nvidia. The clever thing is that Nvidia is n't actually trying to build robots at all. She sells brains to anyone who does it . The chip that lives inside the robot, the software that trains it, and a fully simulated world where robots are trained millions of times before they actually move.
这和 Nvidia 在 AI 革命中采取的做法完全一样。那个在淘金热中卖镐的军火商,如今把注意力转向了物质世界。
It is exactly the same approach that Nvidia took in the AI revolution. The arms dealer who sold pickaxes during the gold rush is now turning his attention to the material world.
观察点
这个频道对 $NVDA 的历次判断
Benzinga 在这只股票上共 11 条判断,这里只列相邻的几条。