Physical AI’s $40 Trillion Question: How Elon Musk Confirms the Biggest Winners
One Nvidia press release moved a small power-chip maker more than 180% in a single session, and that’s where the real trade lies
Listen to the audio version of this article (generated by AI).
On May 21, 2025, Nvidia Corp. (NVDA) named a partner for its new 800-volt data center architecture.
For Nvidia, this was housekeeping. A supplier announcement. As such, NVDA stock hardly budged.
For Navitas Semiconductor Corp. (NVTS), a small power-chip maker, it was the most important day in the company’s history. Shares spiked more than 180% in a single session.
Two completely different outcomes for the same press release.
We call that the magnification effect. And we think it is one of the more useful ideas an investor can weild in a technology boom.
The next boom has already started, and it isn’t chatbots. It’s Physical AI… AI with a body. And the person generating more Physical AI demand than anyone else on Earth is Elon Musk.
The opportunity gap for you is that Musk cannot possibly build all of it himself…
Why the Supplier Usually Beats the Star
Everybody remembers the PC makers of the 1980s and ’90s. IBM. Compaq. Dell. Each a fine business.
But the company that powered the revolution was Intel Corp. (INTC), whose chips became the brains inside nearly every PC on Earth. A $1,000 investment in Intel at the start of 1980 would have grown into roughly $252,200 — a total return of about 25,210%.
Then came the internet, and with it the flameouts. Pets.com. Webvan. The real money was in the plumbing.
Cisco Systems Inc. (CSCO) made the routers and switches that became the backbone of the internet. It went public in February 1990 at $18 a share. By its March 2000 peak, Cisco had delivered total returns of just over 100,600% and had briefly become the most valuable company in the world at a $546 billion market cap.
And most recently, digital AI. While investors chased the AI app of the week, the company that captured the lion’s share of the gains was the one selling GPUs. Over the ten years through late August 2026, Nvidia returned roughly 33,280%.
Intel for computing. Cisco for the internet. Nvidia for digital AI.
Three booms, three household names that got the magazine covers, and three suppliers that got the returns.
Physical AI Isn’t Coming. It’s Already Here
For the past few years, AI lived on a screen. It wrote emails and answered questions.
Physical AI is different. It’s AI that can see, move, and work in the world… We’re talking robots, robotaxis, and machines on a factory floor.
Nvidia CEO Jensen Huang has called this the next great shift:
“The next wave of AI is physical AI. AI that understands the laws of physics. AI that can work among us. Everything is going to be robotic.”
The long-range numbers Wall Street is attaching to that idea are staggering.
In The Humanoid Economy, Morgan Stanley’s Adam Jonas and Sheng Zhong project the humanoid market is “likely to reach $5 trillion by 2050,” built on more than 1 billion humanoids in use, about 930 million of them in industrial and commercial roles. Citi’s GPS team goes further in The Rise of AI Robots, forecasting “648m units and a $7 trillion humanoid market by 2050.” Huang himself has repeatedly framed humanoid robots and labor automation as a $40 trillion total addressable market… possibly, in his words, “the largest industry of all.”
We’d treat all of that as scenario work, not a forecast. Estimates that land in 2050 vary this widely for a reason, and none of them are tradeable.
What is tradable is the near-term ramp.
In its March 2026 research note Physical AI, part 2: Humanoid robots, Bank of America Global Research projects humanoid shipments jumping from 20,000 units in 2025 to 90,000 in 2026… That is a more-than-fourfold leap in a single year, on the way to 10 million units by 2035.
That is the part that creates purchase orders. So the question becomes: who is writing them?
Elon Musk Is Physical AI’s Demand Engine
More than anyone else on the planet, Musk is.
Tesla Inc. (TSLA) has called its Gen 3 Optimus its “first design meant for mass production,” with production targeted to begin before the end of 2026 and a converted Fremont line designed for capacity of up to 1 million robots a year. Musk has floated an aspirational 10 million units a year at Gigafactory Texas.
Space Exploration Technologies Corp. (SPCX) completed the largest IPO in history in June 2026 and is scaling Starlink toward tens of millions of subscribers while pouring billions into orbital compute. Its most recent quarter showed AI-segment revenue up 247%.
xAI is racing to build some of the largest data centers on the planet… hungry for chips, power, memory, and connectivity.
Every one of those is a demand engine. And building a machine like Optimus is genuinely hard: Musk has said roughly 10,000 of its parts are new, and much of the supply chain had to be created from scratch.
Somebody has to make those parts. Somebody has to supply the actuators, the harmonic drives, the rare-earth magnets, the sensors, the power chips, the connectivity.
That “somebody” is where we want your attention.
The Magnification Effect, in Numbers
The math behind the magnification effect is simple.
Tesla does around $100 billion in annual revenue. A few-hundred-million-dollar parts order is a rounding error at that scale. But to a supplier doing $300 million in sales, that same order could double the business overnight.
Same contract. Wildly different impact on the stock.
We’ve watched it happen in real time. When Nvidia disclosed in a July 18, 2024 SEC filing that it owned about 10% of Serve Robotics Inc. (SERV), Serve shares soared 187% in a single day. Nvidia barely moved. Ten months later, Navitas did the same thing on the 800-volt announcement.
Now imagine that dynamic playing out across Musk’s entire Physical AI supply chain over the next several years.
Four Things Physical AI Cannot Exist Without
When we map Musk’s Physical AI push, we keep landing on the same four things it cannot exist without: the data that trains it, the compute that runs it, the connectivity that moves it, and the machines that carry it into the physical world. Musk is spending aggressively to lock up all four… But locking up a layer isn’t the same as making everything inside it.
On the data side, robots need training footage the internet simply doesn’t contain — how a hand grips a part, adjusts when it slips, seats a connector. Nvidia’s Isaac and Cosmos platforms exist to simulate that world and generate that data. Teaching machines about physical space is becoming its own industry.
On the compute side, training and running these models takes memory, power delivery, and thermal management. That’s where suppliers such as Micron Technology Inc. (MU) and Monolithic Power Systems Inc. (MPWR) live.
On connectivity, tens of thousands of accelerators are useless if they can’t talk to each other fast enough. Nvidia has committed billions to lock up optical supply from firms including Lumentum Holdings Inc. (LITE) and Coherent Corp. (COHR). A company only does that when it’s worried about getting enough.
And on the machine side, there’s the body itself. Machine vision from Cognex Corp. (CGNX). Test and automation from Teradyne Inc. (TER). Precision motion and signal chains from Analog Devices Inc. (ADI). Factory integration from Rockwell Automation Inc. (ROK).
None of those are recommendations. They’re an illustration of the method: follow the layer, find the bottleneck, then ask who solves it… and how much of their revenue it would move.
We think there’s a fuller map of this than we can fit in one essay, and we’re not the only ones who think so.
Where This Trade Can Break
Small suppliers can be violently volatile, and the ones tied to a single giant customer carry real concentration risk. A design change or a second-source decision can take the story away as fast as a press release created it.
Musk’s timelines are also famously aggressive. He acknowledged on the Q4 2025 earnings call that the Optimus program is still “primarily for learning.” Treat the biggest unit numbers as targets, not promises… And assume any supplier priced today for mass production in 2027 has room to fall if that slips to 2029.
The magnification effect cuts both ways, too.
A stock that gains 180% on one announcement can give most of it back when the next quarter shows the order was smaller than the market priced in.
The Stars Get the Headlines, the Suppliers Get the Gains
We’ve spent months mapping which companies sit across those four layers… Not the trillion-dollar names everyone already owns, but the potential Intels, Ciscos, and Nvidias of the Physical AI age.
That’s the research we’re presenting at a free workshop on Wednesday, September 9, at 8 p.m. Eastern, alongside our colleagues Louis Navellier and Eric Fry. We’ll walk through Musk’s empire layer by layer, show you where we think the critical bottlenecks are forming, and give away the name and ticker of one company from the research for free.
You can reserve a seat right here.
Because the pattern is as old as Intel, as proven as Cisco, and as powerful as Nvidia: when an enormous amount of money starts moving through a brand-new supply chain, the biggest returns tend to go to the companies getting paid on the way through.
