If you’re planning to buy the “robot supply chain” to ride the humanoid wave — get one thing straight first: the highest-value, highest-pricing-power components in the middle of this chain are almost entirely un-buyable on US markets (the two ends — upstream compute, downstream scenes — you can buy; it’s the richest middle parts you can’t).

That’s not sentiment. It’s structure. Rank the nine layers of the humanoid value chain by “thickness of value,” and you get a smile curve:

• Thick upstream: compute, models, simulation — moats from technology, ecosystem, network effects (NVIDIA at the core).
• Thick downstream: real-world deployment, data flywheels, outcome-based pricing — moats from proprietary scenes, customers, switching costs.
• Thin middle: body assembly, standard hardware — squeezed from three sides (upstream, downstream, and the Chinese supply chain), commoditized.

The problem lives in that middle.

The two richest parts — exactly where US markets have zero exposure

The bulk of a humanoid’s cost is in actuators (joints), and the two components inside those actuators with the highest value and deepest technical moats are high-end harmonic reducers and planetary roller screws.

Who controls them? High-end harmonic reducers belong to Japan’s Harmonic Drive Systems and Nabtesco — decades-old oligopolists. On the standard tier, China’s Leader Harmonious Drive already holds 60%+ domestic and 35%+ global share. Screws are similar — carved up by Japan, Europe, and China.

US-market exposure to these two richest links is basically zero. To participate, you’d have to buy Japanese or Chinese equities, or an ETF.

So what can you buy on US markets in the middle? Only a heavily-diluted “second tier” — frameless motors, torque sensing, precision bearings (the Allient / Novanta / RBC Bearings crowd). They do have humanoid content, but it’s a tiny fraction of their own revenue, and none of them breaks out humanoid revenue separately — you’re buying “real content, badly diluted, with the premium already priced in by the sell side.”

“Can’t the US just build harmonic reducers?” — Sure. But not the moat.

Someone will push back: it’s just precision gears; can’t the US make them?

They can. But they can’t build the moat. The barrier here is decades of accumulated metallurgy, precision grinding, yield learning curves, and hard-to-document manufacturing know-how — and this kind of barrier is “time-incompressible.” Harmonic Drive and Nabtesco took 40 years to grind yields and cost to where they are; a new entrant can’t buy those 40 years with money.

So US markets face a double bind in the middle: enter now, and the richest links are un-buyable; wait for commoditization, and it’s even more pointless — China is dragging these links into price parity at national scale, and by the time you get in, the profit is gone.

The conclusion is blunt: value sits at the two ends of the smile curve. Allocate to the ends, avoid the middle. Buy the certainty of “shovel sellers” upstream and the moats of “scene + data flywheel” downstream. The noisy middle — body assembly — is, for a US investor, visible but untouchable.

Next time someone pitches you a “humanoid supply-chain stock,” ask one thing: which part of the smile curve is it on?

Which “robot supply-chain” names have you owned? In hindsight, were they at the ends or the middle?

(Independent industry and educational research, not investment advice. Companies named are illustrative examples, not recommendations. Data is from public sources and may change. The author and affiliates may hold positions in securities mentioned.)

— Adapted from Embodied Intelligence Investing, Ch. 16 (Value Distribution) & Deep Dive I (Supplier Map)

#EmbodiedAI #Humanoids #SupplyChain