Nvidia 是人形机器人领域的最佳股票，因为它拥有他人赖以建立业务的平台，说话者将其作为 25% 的核心持仓；按约 17 倍远期市盈率计算，它比标普 500 的 19 倍远期市盈率更便宜。

And after Nvidia’s recent keynote, I’m sure the AI revolution is far from over . So, let me explain what just happened and how I invest in this field.

If you only look at Nvidia's financial data, it's a data center company. In the last quarter, Nvidia generated $96.2 billion in revenue, of which $89 billion came from data centers.

This represents more than 92%. Everything else, from computers and gaming consoles to robots and cars, accounted for less than 8%.

Last week, Nvidia CEO Jensen Huang gave a keynote address at the GTC conference in Berlin, and most of the headlines were about data centers. But the parts I found most exciting were almost everything else.

In his keynote address, Jensen said that a 1-gigawatt AI factory would cost around $50 billion, while Nvidia's blog puts the price at a higher level, around $60 billion.

Only a portion of this money goes to Nvidia. The remainder is paid for things like land, buildings, energy, and cooling. This is the problem that the new DSX pod system is trying to solve.

Each unit contains seven vaults of Vera Robin chips. This means 504 graphics processing units and about 25 exaflops of AI computing, all liquid-cooled and connected by more than 10 miles of cables.

It usually takes a dedicated data center crew months to unpack, install, connect, and test it on-site. However, DSX units can be built and tested at the factory and then shipped ready to be plugged and played right out.

Nvidia says this reduces deployment times from months to weeks, and the first DSX units are scheduled to ship early next year.

Nvidia wants to move that work away from the factory floor and into an early stage of the robot's lifecycle. Jensen says that each robot actually needs three computers to operate.

One is in the data center to train its artificial intelligence, the second is to operate a "digital twin," which is a physically accurate simulation of its environment where it can learn and practice skills, and the third is inside the robot itself.

But with AI RAN, it runs as software on Nvidia's accelerated computers, so the same machine can handle our calls and AI workloads simultaneously . This month, Nvidia, T-Mobile and Nokia completed the world's first AI RAN call on a live network at a commercial cellular site in Washington.

When I was at CES last January, Nvidia released the “Alpaca Mayo” model, a family of open-source AI models that watch video from car cameras, decide where to go, and explain the reasons for its decisions while driving.

Automakers use "Alpaca Mayo" to train smaller prototypes of the self-driving cars that actually operate in their vehicles.

After riding in a Mercedes powered by the " Alpaca Mayo" system at the Nvidia GTC conference and seeing it for myself, I agree with Jensen when he says that the "ChatGPT" moment for self-driving cars has arrived.

At that time, Nvidia and Uber also announced plans to launch self-driving taxis powered by Nvidia's Autonomous Driving System and Uber's passenger transport network, starting in Los Angeles and San Francisco in the first half of 2027, and expanding to offer driverless services in 28 cities around the world by 2028.

BYD, Geely, Isuzu, and Nissan are all working on manufacturing vehicles that are ready to operate as self-driving taxis using the Hyperion platform. It is Nvidia’s complete package that includes sensors, computers and safety software, which automakers can build on to make cars drive themselves.

Volvo, Mercedes and Jaguar Land Rover also joined to be compatible with the "Hyperion" platform. This represents about another 3% of all new cars sold last year. As Jensen said, "Self-driving taxis are the first large-scale commercial application of robotics and physical artificial intelligence."

Nvidia is betting that almost every car manufacturer will want to rely on a common platform rather than having to develop and test its own.

During the GTC conference in Berlin , Jensen said that humanoid robots will likely be the most widespread type of robot, and this makes sense , because the modern world was built for humans and by human hands, so humanoid robots will find it easier to work in it.

To accelerate the pace of this sector, Nvidia designed a reference robot in collaboration with a German company called Agile Robots, which manufactures and sells it. It is called "Agile 1SN ".

It is about 4.5 feet long , has 31 independent joints, and is powered by the Jetson Thor computer and Nvidia's Groot robotic AI models. Jetson Thor is a small computer that is placed inside the robot, and is built on the same design as the Blackwell chip used in Nvidia's data center graphics processing units .

In this way , the robot can run its artificial intelligence locally instead of needing the internet and waiting for it to connect to the cloud. "Groot" is Nvidia's family of open-source prototypes specifically designed for humanoid robots.

Think of it as a pre-trained robot brain that companies can then adjust to suit their robots and their specific tasks. The Agile 1 SN will ship early next year assembled and ready to go , allowing researchers to save months or even years of building their own robots from scratch.

I think the same thing will happen in the field of robotics, where the biggest winner may not be the one who makes the robots himself. Companies like Uber, Airbnb and Nvidia are winning by owning the platform that everyone connects to.

That's why I think Nvidia is actually the best stock for humanoid robots. Most of the companies working on it are still private, and several large companies are already building on Nvidia's Jetson Thor platform, including Boston Dynamics, Figure, and Agility Robotics. It is also being evaluated by 1X and OpenAI.

Jetson Thor delivers up to 7.5 times the AI computing power of Nvidia’s previous robotics chip, the Jetson Orin, while being 3.5 times more energy efficient, and the development kit starts at just $3,500 .

This means that the barrier to entry to the platform is very low.

Nvidia's stock itself trades at about 17 times forward earnings, which is cheaper than the S&P 500 index as a whole. More than 92% of Nvidia's revenue still comes from data centers today, and that's where most of the money is,

For future markets, I would prefer to have Uber as the network that self-driving taxis rely on , and Nvidia as the platform that robots will use for simulation, training, and inference in real time.

If I had to divide my money between them, I would keep Nvidia as my primary focus at 25%, because it owns the ecosystems that everyone else builds their businesses on.

Shares of Nokia, Uber, and Nvidia are all trading below the S&P 500's forward price-to-earnings ratio of 19.

Five of these six companies will also be reporting earnings in the next few weeks. Each of these companies also has a direct connection to Nvidia, and possesses a layer of infrastructure that everyone relies on.

So it doesn't matter which AI lab trains the best model, whose robots come off the assembly line first, or which car manufacturer builds the best self- driving taxi. All of these companies are still making profits, making them a great way to get rich without needing luck.