The Humanoid-Robot Supplier Basket:
Why We're Not Buying (Yet)

Harmonic Drive invented the strain-wave gear — the compact joint reducer robots use — and the market has crowned it the "humanoid robot play" of Japan. The problem is that every physical indicator points the other way. When demand is truly straining supply, lead times stretch and factories max out: in the 2021–22 robot boom its lead times hit 10 months. Today they are 1–2 months, and utilization is about 70% — the company shipped roughly 460–580k reducers last year against ~1.85 million of annual capacity. That is slack, not shortage.

The humanoid story also doesn't survive contact with teardowns. Physical bill-of-materials work on Figure's and Tesla's robots found zero stamped Harmonic Drive part numbers; on Tesla's Optimus the company is at best a second source behind a Chinese supplier, and management itself has denied the near-term humanoid revenue the market is imagining. Meanwhile in China — where the robot volume actually is — its reducer sales fell 37% last year, because local maker Leaderdrive sells at 30–62% below its prices.

So the earnings ramp baked into the share price (¥16.7 → ¥96.8 per share by 2028 in analyst numbers) requires a demand wave the company's own factory floor says hasn't arrived, in a market where the wave — if it comes — lands on cheaper Chinese cost curves first. If the humanoid shortage ever does reach Japan's premium tier, you will see it first in lead times and utilization. Today both say no.

THK is a fine industrial company — the world leader in linear guides and ball screws, the components that let machine tools move precisely. The trouble is what the stock price assumes on top of that business. Working backwards from its valuation, about half the market cap is paying for a future in humanoid-robot actuation.

Two facts collide with that premium. First, the component that actually matters in humanoid limbs is the planetary roller screw — a heavier-duty cousin of the ball screw — and THK doesn't have a production PRS product; our sweep of primary sources found no named humanoid program where THK is qualified. Second, even if it built one, the Chinese suppliers already qualified at robot makers are running at roughly 4% of the capacity they've built — meaning any demand surge gets absorbed by cheap idle capacity long before it reaches Japanese pricing.

You can like THK's core business and still recognize that the robot premium in the price is paying for a product that doesn't exist, in a market segment that's already oversupplied. The full research write-up completes tonight; the direction is already clear from the company's own product line and the capacity data.

Leaderdrive is China's leading harmonic-reducer maker and the one company in this basket with a genuine, physical shortage: monthly output has been running at 131–150% of its official nameplate capacity, humanoid orders reportedly exceed everything it can build, and its next big factory doesn't arrive until 2028 (the precision grinding machines it needs are export-controlled and back-ordered).

Here's the uncomfortable arithmetic. A factory can only ship capacity × price. Capacity next year is at most ~800–960k units even with heroic debottlenecking. Price is the killer: despite being sold out, Leaderdrive's average selling price fell 15% last year, because competitor Laifual profitably sells similar reducers at less than half Leaderdrive's price and has spare capacity. Sold out × flat-at-best pricing caps next year's revenue around ¥0.9–1.1B — below the ¥1.3B analysts already assume. Even our aggressive 2028 scenario produces about ¥2.3 of earnings per share; the stock trades at ¥379.5.

What would change our mind (worth watching, genuinely): its realized prices turning up — that would mean the shortage finally outran the whole Chinese cost curve; a filed capacity addition for its higher-value integrated modules; or sustained output above 80k/month. Until one of those appears, this is a great business at a price that requires three miracles at once. (Also practical: STAR Market shares are effectively closed to foreign retail investors anyway.)

Ouster makes solid 3D lidar and has real, diverse customers — warehouse robots, traffic intersections, mining trucks, defense drones. None of that is the issue. The issue is market structure: Hesai and RoboSense ship millions of units a year from China at list prices as low as $200–1,000, while Ouster's blended price is above $5,000 and drifting down every quarter. Its own contract manufacturer's line could build 100k+ units a year; Ouster shipped 25k last year.

The company needs roughly $286M of annual revenue to stop burning cash at current margins; analyst consensus for this year is about $220M. So the math requires either much faster volume (which, in this competitive structure, is usually bought with price cuts) or prices holding (which three years of data contradict). Robots getting cheap lidar is great for robots — it's the robot makers who capture that value, not the sensor supplier being commoditized.

Hengli is a genuinely good industrial company — it broke the Japanese/German lock on excavator hydraulics and is applying the same playbook to precision screws. But its core business remains construction-machinery hydraulics (deeply cyclical, tied to Chinese construction), and the roller-screw option that excites the market faces the same physics as everyone else's: humanoid-qualified Chinese screw makers are running at a few percent of built capacity, and the price-setter sells at ¥499 a screw. When your premium product's market clears at a tenth of the price you'd need, volume — not price — is the only prize, and Hengli's valuation already assumes a large one. Final numbers land tonight; the mechanism verdict is already in.

Best Precision is a legacy turbocharger-parts maker (a real but slowly-shrinking business as cars go electric) with an announced pivot to planetary roller screws for humanoids. Announced is the key word: our research found ~250 sample screws delivered, screw revenue of about ¥223k — that's thousands of dollars, not millions — and finished-screw capacity around 25k sets a year against an industry that has already built 2M+ of nameplate. Every yuan of its robot valuation is paying for execution that hasn't happened, in a segment already oversupplied before it arrives. It could execute brilliantly; the price asks you to assume it in advance.

This is the most interesting of the five, and the one where we're genuinely waiting on tonight's math. Shuanglin's reverse-engineered screws at ¥499 are what collapsed the market's pricing — meaning it isn't the victim of the cost-curve war, it's the aggressor. There's a real scenario where humanoid volume arrives, Shuanglin takes outsized share as the cost leader, and scale economics turn thin unit margins into meaningful profit — the way BYD won in electric cars during a price war.

What has to be true: actual binding orders from robot programs (today's evidence is partly modeled, not filed), margins that survive at ¥499 (the gross spread over cost is narrow), and a valuation that doesn't already assume victory. That last check is what tonight's deterministic pass computes. Until then: watchlist, not wallet.

Wuzhou makes bearing components and screw blanks — the semi-finished steel that becomes a roller screw after precision grinding. The humanoid story attached to it assumes tightness at the top of the market (genuine: Swiss makers Rollvis and GSA quote months of lead time at $1,350–2,700 a screw) flows down to everyone in the chain. Our mechanism test found it doesn't: the domestic qualified tier has slack, and blanks are the least differentiated step — several suppliers can make them, so scarcity pricing never accrues there. A fine business; a story stretched one link too far up the chain.

Not every humanoid uses roller screws — Unitree's robots use planetary QDD joints with zero screws and zero harmonic reducers, and ZD Leader supplies that architecture. We included it deliberately as the "what if the other design wins" hedge. The mechanism test came back the same: tightness in rotary actuation clears through cheaper qualified domestic suppliers, and ZD Leader's precision-reducer business runs ~21% gross margins — the signature of a commoditized component, not a bottleneck owner. If the QDD architecture wins, robot makers win cheaper joints; the joint maker doesn't win pricing power.

A strain gage is a foil element that measures force by how much a metal part flexes — it's how a robot hand knows how hard it's gripping. VPG has quietly shipped preproduction force sensors (~$600k so far, ~$5M expected this year) to unnamed humanoid maker(s), with roughly $400–500 of content per robot. The go-to-production decision is expected in the second half of this fiscal year — a rare, dated, binary catalyst.

The upside math is violent because the share count is tiny: if humanoids reach ~650k units in 2028 and VPG keeps ~20% of the sensing, that's roughly +$1.74 of extra EPS on a $3.07 consensus — a +57% earnings surprise; the aggressive case roughly doubles that. And unlike the reducer/screw markets, VPG's niche has real protection: its former parent is contractually barred from competing, and cheap alternatives aren't precise enough for the calibration-locked sockets it serves.

Three reasons this is an option, not a conviction buy: the stock already re-rated on its AI test-equipment boom (you're paying ~37× 2028 consensus before any robot revenue); VPG historically monetizes through volume and mix, never price increases; and Minebea Mitsumi mass-produces 2 million cheaper force sensors a month — if robot makers decide "good enough" beats "perfect," the option expires worthless. Playbook: small starter at most, add only when the production decision actually lands, exit on a slip into next year or a competitor win.