Humanoid Robot BOM DD — Full Supply Chain (JP + US + Global)

Date: 2026-05-16 | Time: 00:15 JST | Type: Robotics-specific DD (fired via new `research_robotics.sh` script) | Theme: Robotics | Method: T-Glass per-component backward trace from humanoid BOM

Trigger

User-supplied humanoid robot BOM image with full cost breakdown (~$55k/unit, typical Tesla Optimus / Figure / Apptronik class). Special focus: harmonic drive systems and relevant JP companies, plus US/global competitors.

The BOM (reproduced)

Total component count per humanoid:

• 14 harmonic reducers (6 shoulder + 2 elbow + 6 waist/pelvis)
• 12 planetary reducers (hands)
• 6 ball screws + 8 planetary roller screws (linear actuators in limbs)
• 28+ frameless torque motors (rotary + linear)
• 12 coreless motors (hands)
• 60+ bearings (cross roller + angular contact + 4-point contact + ball)
• 50+ encoders
• 30+ force sensors (torque + 1D + 6D)

T-Glass per-component supplier matrix

Layer 1: Harmonic reducers (14 per humanoid) — HIGHEST FOCUS

T-Glass insight (most important finding): Harmonic drive supply is a JP-CN duopoly, not a JP monopoly. Tesla Optimus mass production (Q3 2026) is the volume tier where Green Harmonic wins on cost; high-spec Western humanoids (Figure, Apptronik, 1X, Boston Dynamics) is where HDS retains design wins.

Layer 2: Planetary reducers (12 per humanoid — hands)

Layer 3: Ball screws (6 per humanoid) + Planetary roller screws (8 per humanoid)

Layer 4: Frameless torque motors (28+ per humanoid)

Layer 5: Coreless motors (12 per humanoid — hands)

Layer 6: Bearings (60+ per humanoid)

Layer 7: Encoders (50+ per humanoid)

Layer 8: Force sensors (30+ per humanoid)

Layer 9: Head (FSD + camera + IMU)

Layer 10: Battery (2.3 kWh @ 52V)

Updates to robotics_companies.json

Conviction changes (3 stocks UPGRADED)

NEW companies added (3)

#### 6471 NSK Ltd. — MEDIUM conviction

• Layer: Bearings + ball screws (BOM direct)
• Mkt cap: ¥593B | PE TTM 26.0 / Fwd 20.1 | Div yield 2.68% | +83% 52w
• Key: Top-3 globally in industrial bearings; cheapest bearings-play valuation in the basket
• FY26: Revenue +14.4%, earnings +114.8% (strong operational leverage)

#### 6516 Sanyo Denki — MEDIUM conviction

• Layer: Frameless / AC servo motors
• Mkt cap: ¥266B | PE TTM 30.7 / Fwd 23.3 | Div yield 2.23% | +124% 52w
• Key: Smaller cap = more torque per design win; FY26 earnings +53.7%
• Caveat: Cooling segment (Sanyo Cooler) creates AI data center cycle correlation

#### 6592 Mabuchi Motor — MEDIUM conviction

• Layer: Coreless / small motors for hand actuators
• Mkt cap: ¥388B | PE TTM 13.6 / Fwd 21.2 | Div yield 3.53% | +48% 52w
• Key: Cheapest valuation in entire humanoid basket (PE 14×) + highest yield (3.5%). Fwd PE > TTM = consensus expects margin normalization
• Caveat: May sit a step below Maxon/Faulhaber in premium tier; auto exposure adds cycle risk

Anti-Pattern Check (per research_philosophy.md, MANDATORY for robotics)

AP01 — Peak Earnings Trap

• HDS (6324): PE 378× TTM on trough earnings (-53.7% YoY NI) and +85% 52w = textbook AP01 flag. Even after upgrade to MEDIUM, position size should reflect this. Conviction-capped.
• Nabtesco (6268): +128% 52w heated but earnings +55% YoY supports the move. Not flagged.
• THK (6481): +99% 52w, OI -9.3% in latest quarter. Cost-pressure cycle ongoing. Manageable.
• Sanyo Denki (6516): +124% 52w but earnings +53.7% supports.
• Mabuchi (6592): +48% only, PE 14× — cheapest, no AP01 concern.
• Bearings (NSK / JTEKT / NTN): mid-cycle on auto + machinery. No AP01.

AP02 — Capacity Hangover (CRITICAL FOR HUMANOID)

FLAGGED — clear evidence of China supply-side flood:

• Suzhou Green Harmonic opening 500K units/year factory in 2026. That's roughly 35K humanoid robots/year of harmonic reducer supply from one Chinese factory alone (assuming 14 reducers per humanoid).
• LeaderDrive + Beijing CTKM + Zhuhai Hi-Rel: multiple Chinese harmonic reducer entrants
• Sanhua got $685M Optimus order (different components — thermal management, valves)

This is the biggest risk to the entire JP humanoid component thesis. The China cost-tier is real and Tesla is dual-sourcing explicitly.

Mitigation: focus JP exposure on non-Tesla humanoid customers (Figure, Apptronik, 1X, Boston Dynamics) where high-spec / Western-aligned sourcing protects JP position.

AP03 — Concept Stock Trap (also CRITICAL for robotics buzz)

Companies in the JP humanoid basket where the humanoid % of revenue is small relative to the multiple they trade at:

• HDS (6324): PE 378× on entire business — most of HDS revenue is still industrial robotics (Fanuc + Yaskawa anchor), not humanoids. The humanoid-fueled re-rating is partly justified, partly concept-stock.
• THK (6481): humanoid is currently <5% of revenue per industry estimates. Most of P&L is machine tools + auto. The +99% 52w move is partly multiple expansion on humanoid concept.
• Sanyo Denki: humanoid would be <3% of FY26 servo segment revenue. +124% 52w is partly AI cooling + partly humanoid concept.

Mitigation: weight position sizes to the FUNDAMENTAL humanoid revenue contribution today, not the projected 2030 TAM.

Cycle position (per stock)

Trade construction — humanoid BOM exposure

Total HIGH+MEDIUM new-money allocation to humanoid BOM direct hits: ~46%. The remaining 54% covers automation anchors + diversification.

Risks

1. China cost-tier wins volume (BIGGEST risk): Tesla Optimus + Chinese humanoid makers (Unitree, Fourier, etc.) drive volume to Suzhou Green Harmonic / Sanhua / Chinese motor makers. JP names retain premium / Western tier but the unit-economic upside compresses.

2. AP01 multiple compression: HDS PE 378× cannot persist. Either FY27-28 earnings catch up (10× FY26 trough) or the multiple compresses. Either way, the volatility is asymmetric.

3. Humanoid mass production timing: Tesla Optimus Q3 2026 mass production targeted but execution risk real. If slipped to 2027+, the JP component pop fades.

4. JPY strength: JP exporters with USD revenue compress on yen rally. Most of the BOM JP basket is auto + industrial + global trade exposed.

5. Tariffs: US-China tariffs could push Western humanoid OEMs back toward JP supply (positive). Or US could try to subsidize US suppliers (negative for JP).

Sources

• BOM image (user-provided): typical Tesla Optimus / Figure / Apptronik class humanoid breakdown ~$55k
• Harmonic Drive global market share: https://optimusk.blog/blog/tesla-optimus-suppliers/
• Suzhou Green Harmonic Tesla Optimus dual-source: https://eu.36kr.com/en/p/3780414717129481
• Suzhou Green Harmonic factory: https://nextfinancial.substack.com/p/one-million-robots-per-year-the-38
• Sanhua $685M Optimus order: https://eu.36kr.com/en/p/3510288514980998
• Morgan Stanley Humanoid 100: https://advisor.morganstanley.com/john.howard/documents/field/j/jo/john-howard/The_Humanoid_100_-_Mapping_the_Humanoid_Robot_Value_Chain.pdf
• TrendForce Japan components dominance: https://www.trendforce.com/presscenter/news/20251209-12825.html
• KraneShares Humanoid 2026: https://kraneshares.com/humanoid-robotics-in-2026-the-race-from-pilot-to-platform/
• iREX 2025 HDS humanoid demo: https://www.roboticstomorrow.com/news/2026/04/08/global-robotics-industry-converges-on-japan-for-humanoids-summit-tokyo-2026/26378
• StockAnalysis price/PE per ticker: https://stockanalysis.com/quote/tyo/{ticker}/
• J-Quants v2 live prices: api.jquants.com/v2/equities/bars/daily

Outstanding follow-ups

1. Verify named humanoid design wins — Tesla Optimus / Figure AI / Apptronik / 1X / Boston Dynamics supplier disclosures should be cross-checked at each company's quarterly earnings or any teardown reports

2. Watch Q3 2026 — Tesla Optimus mass production start = single biggest robotics catalyst this year

3. iREX 2027 — biennial robotics expo; next round of humanoid OEMs will showcase JP component design wins

4. Add Yaskawa to detailed update — currently MEDIUM in robotics theme; humanoid catalyst should re-evaluate

5. Nidec (6594) currently AVOID in robotics theme due to restructuring — should be re-evaluated since their motor business is directly humanoid-relevant. Follow-up evolve round needed.

ERRATA — Cross-validation update (2026-05-16, post-publication)

Per user request to "double confirm the PE that you wrote", cross-validated all 6 humanoid stock PE figures against Yahoo Finance JP (company guidance) + Buffett-Code where available. Two material findings:

Finding 1 — Systematic StockAnalysis-vs-guidance gap

StockAnalysis Forward PE values are analyst consensus (more bullish), while Yahoo Finance JP Forward PE values are company guidance (conservative). The original DD used the consensus values without flagging this. Corrected:

The pattern is the same one previously documented for Fujikura / Furukawa Electric in the light theme — StockAnalysis lags or uses consensus while company guidance can differ materially. Going forward, every DD should cite BOTH where they diverge.

Finding 2 — MAJOR missed catalyst: NSK + NTN merger announced 2026-05-12

This is the biggest miss of the original DD. On 2026-05-12 (3 days before the DD was published), NSK Ltd. (6471.T) and NTN Corporation (6472.T) announced a basic agreement to merge by October 2027 into a joint holding company. The combined entity will be:

• World's largest bearing maker at ~24% global share, surpassing SKF
• NSK president + NTN chairman of joint HC
• Final agreement target: within 6 months from MoU (i.e. by ~2026-11-12)
• TSE Prime technical listing: October 2027

Market reaction:

• NTN: +10.18% on 5/13, then -9.93% on 5/15 (uncertainty trade)
• NSK: -12.2% over 3 trading days
• Sources: Nikkei Asia, NSK Americas official announcement

Implications for the humanoid BOM thesis:

• The combined NSK+NTN will be the dominant supplier for the 60+ bearings every humanoid needs (cross roller + angular contact + 4-point contact + ball bearings)
• Better pricing power and capacity rationalization should support margins through the China commodity-tier pressure
• Short-term volatility = potential entry opportunity if the integration thesis is right
• Antitrust review across Japan / EU / US / China is the main execution risk

Action taken

1. Added NTN Corporation (6472.T) to robotics_companies.json (MEDIUM conviction, Fwd PE 16.28×, 3.16% div yield)

2. Rewrote NSK (6471.T) thesis around the merger

3. Updated all 6 forward_pe values + added forward_pe_consensus field where they differ

4. Added 6 errata entries to robotics_companies.json

5. Added recent_updates[] entry of type errata_correction

6. THK (6481.T) Q1 2026 OP +364.4% YoY noted — even stronger than original DD credited

Trade construction update

The original DD allocation table for bearings should now read:

HDS allocation unchanged (5-7%) — the AP01 flag is even stronger on guidance Fwd PE 149× than the original 114×.

Lesson recorded: Single-source PE citations are unreliable. Both consensus (StockAnalysis) and company guidance (Yahoo JP / Buffett-Code) values should be quoted for any catalyst-driven DD. Major corporate-action events (mergers, splits, spin-offs) need an explicit pre-publish check against TDnet / EDINET disclosures from the prior 30 days.