NVIDIA owns the brain. Who builds the rest?

Hi {{first_name|Investor}}

Part 3, the final issue of the NVIDIA earnings series. We covered Vera, the CPU story, then the sovereign and enterprise customer almost nobody models. This issue is the one I find most interesting for a long-horizon portfolio: physical AI, and the supply chain that rides behind it.

For years NVIDIA folded robotics and automotive into a much larger number. This quarter it carved out a new Edge Computing platform and, for the first time, disclosed a direct figure for physical AI.

Edge Computing came in at $6.4 billion for the quarter, up 29% year over year. Inside it, NVIDIA ($NVDA) called out physical AI specifically: more than $9 billion in revenue over the last twelve months. That is not a projection. It is trailing revenue already on the books.

Then Jensen Huang attached a timeline. He said NVIDIA is practically the only company serving physical AI today, that the roughly $100 trillion of global industry untouched by IT over the last 30 years is about to be reshaped by AI, and that within the next five years physical AI and robotics will grow incredibly fast. He pointed to a hard commitment as proof the demand is real: NVIDIA said its Uber partnership will put robotaxi fleets across nearly 30 cities and four continents by 2028.

The headline is NVIDIA. The research question is everything around it. NVIDIA owns the brain of the robot, the GPU and now Vera, at premium margin. A robot operating in the real world also needs senses to perceive, a nervous system rugged enough to run in the field, and a body that moves and grips. Those layers ride the same demand curve NVIDIA just put a clock on, often from a far smaller base and a far smaller valuation.

Owning the rest of the robot only works if you know what you own, and a recent McKinsey teardown of the humanoid supply chain draws the line that matters: bottleneck versus demand-pull. A bottleneck part has little adjacent capacity to borrow from. It does not ride a bigger neighboring industry, so supply can stay tight and pricing power can persist as volumes rise. A demand-pull part borrows scale from an adjacent industry like EVs or consumer electronics, so it grows with the theme but typically competes harder for margin along the way. Both can make money, in very different ways.

That distinction is the stakes. Own a demand-pull name and you are betting on execution in a market that scales around you. Own the one real bottleneck and you are betting on pricing power that compounds as robot volumes rise. Same theme, two different return profiles, and sorting the basket along that line is the part that pays.

The rest of this issue is for Insider members: the five-name basket sorted by exactly that line, my top pick and why, and the upstream chokepoint above it. Upgrade to keep reading. Insider also gets you the full archive and every week's names.