InfraMosaic · Humanoid Intelligence

The Humanoid 100 — 2026 Update

How Morgan Stanley's map of the humanoid robot value chain reads 16 months on

Source paper: Morgan Stanley Research, The Humanoid 100: Mapping the Humanoid Robot Value Chain — pricing as of 4 Feb 2025.
This update: InfraMosaic database, build of 13 June 2026 — 255-company universe, $40.7tn live market cap, 58 programs, 26 deployments, IHCI cost index and the supply/demand order book. The paper is walked in its own running order, exhibit by exhibit.

100 names
255
universe tracked
89/100
MS-100 covered
not given
$40.7tn
live market cap
~0 units
122k
units deployed / on order
"joints 60-70%"
Sensors #1
BOM cost (Exhibit 54)
unveils
4
at mass-production

Each page below is shown as a comparison: what the paper said in February 2025, what InfraMosaic's current data shows in June 2026, and — in the dark bar — what a re-write would actually change. Nothing here redistributes the MS list; the Humanoid 100 is used only as the cross-check spine, and the dataset is InfraMosaic's own.

● MS · Feb 2025 — what the paper said ● InfraMosaic · Jun 2026 — what the data shows Δ — what the re-write changes
Part 0

The index itself

Cover · the "Nasdaq-100 of humanoids"
MS · Feb 2025

A curated 100-company index — Brain, Body (64 names), Integrators — pitched as one watchlist that spans the whole value chain. Pricing 4 Feb 2025.

InfraMosaic · Jun 2026

Now a 255-name universe (150 core + 105 tier-2). All of MS's original 100 sit inside it — the cross-check shows 89 ms_h100 = yes after folding the 20 constituents v1 had missed. The other 166 are InfraMosaic's extension into the China component long-tail and rest-of-world supply chain. Total tracked cap $40.7tn, 255/255 live.

ΔRewrite
Stop calling it "100." The interesting universe is ~2.5× bigger, and the action since Feb 2025 is in the tier-2 Chinese parts makers the original mostly skipped. Keep a 100-name "blue-chip" cut for the headline; the working database is 255.
Part 1

Executive thesis

The investment case
MS · Feb 2025

Humanoids move from demo to deployment over the decade; a ~$5tn market by 2050, >1bn units by the 2040s. The supply chain — not the robot brands — is the durable trade, and it runs heavily through China. "52% of the Humanoid 100 already involved; 48% material potential."

InfraMosaic · Jun 2026

The thesis aged well — and the boring half is exactly where value showed up. Brands still mostly pre-revenue; picks-and-shovels re-rated. Two things MS couldn't see now dominate:

1. Deployment is real — 26 programs in paid pilots/early production, 122,302 cumulative units on order or deployed.
2. A memory/AI-compute super-cycle repriced the Brain violently (NVIDIA $4.97tn, Micron $1.1tn).

ΔRewrite
Lead with "deployment has started" not "is coming," and split the thesis cleanly: the Brain has already been paid for by the AI trade; the Body is where the humanoid-specific, not-yet-priced optionality still sits.
Part 2

Momentum & involvement

Exhibits 1–4 · mentions over time, involved vs potential
MS · Feb 2025

A near-vertical 2023–24 ramp in transcript and media mentions of "humanoid" — the attention leading indicator. Just over half of the 100 had confirmed involvement.

InfraMosaic · Jun 2026

The mention curve climbed through 2025 then plateaued high — humanoids are a standing agenda item, not a novelty spike. Confirmed involvement (programs, supply contracts, named design-ins) is now the majority for core names; the "potential" bucket shrank as suppliers disclosed design-wins.

ΔRewrite
Swap the lead indicator from a mentions chart to a deployments + unit-orders series — mentions were the 2024 proxy, shipments are the 2026 proxy. The 52/48 split becomes roughly 70/30 confirmed/potential.
Part 3

The Brain

Exhibits 10–12 · models, compute, EDA, simulation, vision
MS · Feb 2025

Foundation models, AI compute, EDA, simulation, vision. US- and Taiwan-led: Alphabet, Microsoft, NVIDIA, TSMC, Arm, Cadence, Synopsys, Palantir, Oracle — with memory (Micron, SK Hynix, Samsung) as a footnote and vision (Hexagon).

InfraMosaic · Jun 2026

Brain = 67 of 255; China 41.8% by count but value and moat sit in the US/Taiwan core. Most concentrated segment (HHI 3,232) — compute is an oligopoly. The 12 biggest names in the whole universe are now almost all Brain:

NVIDIA $4.97tn · Alphabet $4.39tn · Apple $4.28tn · Microsoft $2.90tn · Amazon $2.57tn · TSMC $2.20tn · Broadcom $1.82tn · Tesla $1.53tn · Meta $1.44tn · Samsung $1.40tn · Micron $1.11tn · SK Hynix $1.01tn

ΔRewrite
Two fixes to Exhibit 11. (1) Memory is no longer a footnote — Micron & SK Hynix are trillion-dollar names (Micron's 52-wk range $103 → $1,089); top billing, not the annex. (2) The Brain is already re-rated by the AI trade — owning it for humanoid exposure is a lottery ticket stapled to an already-expensive AI stock.
Part 4 · the heart of the paper

The Body

Exhibits 13–15 · the electro-mechanical stack
MS · Feb 2025

The Body (64 companies) is the electro-mechanical stack and where China has built an almost unassailable position. Parts walked one by one: bearings, screws, reducers, motors + rare-earth magnets, encoders, sensors, batteries, analog semis, structure. "Joint drive modules are 60–70% of a humanoid's cost."

InfraMosaic · Jun 2026

Body = 147 of 255 (the universe's centre of gravity), China 41.5%, HHI 2,425. Every MS parts-bucket is now a value-chain node with a live exposure score and a supply/demand utilisation. The top of the exposure table is entirely Body:

Highest-conviction read in the database — IMES exposure leaders (all Body)
RankCompanyIMESWhy
1Harmonic Drive Systems92strain-wave reducer near-monopoly
1Moog92precision actuation
3Symbotic88warehouse automation pull
4Renishaw85encoders / metrology
5Nabtesco84cycloidal reducers
5Intuitive Surgical84dexterous precision analog
7UBTech83only listed pure-play maker
ΔRewrite
The one material correction to the paper's economics. MS's own Exhibit 54 shows sensors are the #1 cost line at 37%, ahead of motors (20%) and screws (20%) — yet the prose still anchors on "joint modules = 60–70%." The re-write re-bases the BOM on Exhibit 54 and foregrounds the six-axis force-torque sensor as both the top cost and the tightest bottleneck (ATI/Novanta near-monopoly; flagged at 440% of 2030 capacity). The paper buries its own most important number.
Parts ledger — MS bucket → 2026 node read
Part (MS Exhibit 14/15)MS constituents2026 read
BearingsNSK, RBC, Regal Rexnord, Schaeffler, Timkencommoditising; low IMES
Screws (ball + roller)Hengli, Hiwin, NSK, SKF, Shanghai Beite, THKdeficit 770% tightest mechanical bottleneck
Reducers (harmonic + cycloidal)Harmonic Drive, Nabtesco, LeaderDrive, Shuanghuan, Zhongda, Hiwin, Hotahighest-IMES node; China share climbing fast
Motors + magnetsEstun, Leadshine, Moons', Nidec, Inovance + JL Mag, Lynas, MPmagnets ~100% China; binding physical constraint (Dy/Tb controls)
EncodersNidec, Novanta, SensataRenishaw IMES 85; precision moat intact
Sensors — force/torqueKeli, Novanta, Sensata, TE#1 BOM · 440% the re-write's hero chart
Vision / lidarADI, Hexagon, Keyence, Sony, Robosense, Aptiv, Magna, Valeo, Teledyneglut 5.5% auto over-capacity spilled in
BatteriesCATL, EVE, LG ES, Samsung SDIglut ~0% humanoid demand a rounding error on EV cells
Analog semisAllegro, Infineon, Melexis, NXP, Onsemi, Renesas, ST, TIsteady; auto-cycle linked
Structure / wiringAmphenol, Aptiv, Magna, TE, Xusheng, Sanhua, TuopuChina die-casters scaling
Part 5

Integrators

Exhibits 16–17 · who assembles the robot
MS · Feb 2025

Five archetypes: autos/OEM (BYD, GAC, Hyundai/Boston Dynamics, Tesla, Toyota, XPeng), tech/consumer (Apple, Foxconn, LG, Samsung, Sony, Xiaomi), cloud (Alibaba, Amazon, Naver, Tencent), industrial robotics (ABB, Midea/KUKA, Teradyne), pure-play (Rainbow Robotics, UBTech).

InfraMosaic · Jun 2026

Integrators = 41 of 255, China 34.1%, least concentrated segment (HHI 2,064) — many credible assemblers, no winner yet. Pure-plays pulling ahead on shipped units: UBTech (Walker S2, 3-min battery-swap, >800m-yuan orders) and AgiBot/Zhiyuan (A2, #1 by 2025 volume, ~5,168 units) now out-ship Western brands.

ΔRewrite
Promote two under-weighted points: (1) Foxconn/Hon Hai as the contract-manufacturing kingmaker — whoever wins the OEM race likely builds at Foxconn; (2) Tesla as both Brain and Integrator with the only credible 1M-unit line. Note that Figure, Apptronik and Agility are integrators absent from MS-100 because they're private — captured in the 12-name watchlist (Figure ~$39bn, plus newcomer Humanoid / HMND 01, London).
Part 6

Composition & regional

Exhibits 18–23 · geography of the value chain
MS · Feb 2025

Humanoid-100 skewed China & Taiwan on the Body, US on the Brain. China = 63% of the supply chain (supply-chain-weighted). Analyst-surveyed AI-exposure / materiality / pricing-power scatter.

InfraMosaic · Jun 2026

By name across 255: China 40.4% (103) · North America 24.3% (62) · Japan 14.5% (37) · Europe 7.8% (20) · Korea 5.9% (15) · Taiwan 4.3% (11) · Asia-other 2.4% · Oceania 0.4%. Country HHI 2,468. China-share by segment: Brain 41.8%, Body 41.5%, Integrator 34.1%.

ΔRewrite
MS's "63% China" was supply-chain-weighted; by-name it's 40.4% — but the Body component nodes reproduce MS exactly (magnets ~100% China, actuators ~64%). The thesis intensified: Dy/Tb export controls now sit on the highest-criticality nodes. Sharper message: "China's dominance isn't broad, it's surgical — it owns the three or four parts that can't be substituted."
Part 8

Valuation & performance

Exhibits 24–34 · P/E, CAGR, index returns — the section that changed most
MS · Feb 2025

P/E rank of the Humanoid 100, P/E vs 3-yr revenue CAGR, and LTM performance of the index and its Brain/Body/Integrator and regional sleeves vs benchmarks. Pricing 4 Feb 2025.

InfraMosaic · Jun 2026

Since Feb 2025 the universe lived through an AI/memory super-cycle:

· NVIDIA $4.97tn, the gravitational centre.
· Micron $1.11tn (52-wk $103→$1,089) and SK Hynix $1.01tn (KRW 238k→2,407k) — memory went from footnote to two of the twelve biggest names.
· Total index cap $40.7tn; the top 12 alone ≈ $28tn.

ΔRewrite
Every valuation exhibit is stale and must be re-struck — and add a warning the original lacks: the index's return since Feb 2025 is overwhelmingly an AI-compute/memory story, not a humanoid story. A reader buying "the Humanoid 100" today is mostly buying the AI trade. The genuinely humanoid-levered return sits in the small/mid Body names — exactly where IMES points and index-cap weighting points away.
Part 9

Pop culture, pros/cons, TAM & hurdles

Exhibits 35–47 · the case for adoption
MS · Feb 2025

Humanoids general-purpose but expensive/immature vs purpose-built automation. ~$3tn US humanoid TAM by 2050 on labor-substitution; adoption negligible to ~2035 then an S-curve. Adoption hurdles listed.

InfraMosaic · Jun 2026

The cost gap closed faster than the con-side assumed — the IHCI shows a Western reference robot down from ~$200k to ~$95k in 18 months. The market model keeps the same S-curve and brackets MS: base 550k units / $13bn (2030) → 1.4M / $30bn (2035). The 2028 inflection is intact; pilots are hundreds-to-thousands, not millions.

ΔRewrite
Move "cost" from the con column toward neutral; keep reliability/autonomy in unstructured environments as the live con. Keep the TAM architecture but tighten the near-term units with real 2025–26 order data. Add two 2026-specific hurdles absent from the original: rare-earth export controls and force-torque sensor supply.
Part 10

Hardware & bill of materials

Exhibits 48–55 · Optimus Gen-2 teardown
MS · Feb 2025

Optimus Gen-2: 28 actuators (14 rotary + 14 linear), ~50 DoF, 22-DoF hand. Exhibit 54 BOM: Sensor 37%, Motor 20.3%, Screw 20.2%, Reducer 12.6%, Encoder 3.9%, FSD+chips+camera 3.8%, Bearing 0.8%, Battery 0.5%.

InfraMosaic · Jun 2026

Reference BOM now re-based directly on Exhibit 54 (~$55,000): Sensors $20.4k (37%) · Motors $11.2k (20%) · Screws $11.1k (20%) · Reducers $6.9k (13%) · Encoders 4% · Compute 4%. Whole-robot curve: Western $200k (2024) → $40k (2030); China $46k → $18k. A regression test locks sensors as #1 so the correction can't silently revert.

ΔRewrite
The paper's biggest internal inconsistency to fix: prose says "joint modules 60–70%," its own Exhibit 54 says "sensors 37% #1." Resolve it in favour of the data — sensors lead — and tie it to supply (F-T sensors also the tightest bottleneck). Flag the planetary roller screw cost-down ($3,000 → $800 at Tesla scale) as the proven template the rest of the BOM follows.
Part 11

Unveils tracker

Exhibits 61–64 · who showed a robot
MS · Feb 2025

~67 notable unveils since 2022; 2024 = 51 unveils (China 35, US 8, EMEA 4, Rest-Asia 4); China 61% of all unveils, US/Canada 24%. General-purpose 56% of use-cases.

InfraMosaic · Jun 2026

Folded into a 58-program database with specs, status and pricing, carried forward through 2025–26. By region: China 32 · N. America 16 · Europe 4 · Japan 3 · Korea 1 · Middle East 1 · India 1. By maturity: Prototype 22 · Limited-prod 17 · Pilot 12 · Mass-prod 4 · Pre-prod 1. New: Unitree R1 ($5,800), 1X NEO ($20k), XPeng Iron, UBTech Walker S2, AgiBot A2, plus non-MS Humanoid / HMND 01 (London).

ΔRewrite
The chart graduates from "who showed a robot" to "who shipped one." Four programs are at mass-production — a status that didn't exist when MS drew Exhibit 61. China's lead in count persists; add a status axis showing China also leads in units shipped (Unitree/UBTech/AgiBot).
Part 12

What's genuinely NEW since the paper

Three datasets MS could not build in Feb 2025
New datasetWhat it adds
Deployment tracker26 programs, 122,302 cumulative units in paid pilots/early production. In Feb 2025 there was essentially nothing to count — this is a whole new section on who is actually being paid to put robots on floors.
IHCI — InfraMosaic Humanoid Cost IndexMonthly, 2024-01 = 100. Now Western 47.3 / China 65.0 (Western reference robot down ~53% to ~$95k; China down ~35% to ~$30k). MS gave point estimates; InfraMosaic gives the track.
Supply / demand order bookRobot ramp × per-robot bill-of-quantities vs assessed capacity. 2030 verdicts — deficit roller screws 770%, F-T sensors 440%, integrated actuators 385%; glut LiDAR 5.5%, batteries ~0%, magnets 0.5%. Converts "China dominates the Body" into a quantified, node-by-node tightness map.
Part 13

Scorecard — right, wrong, couldn't know

Grading the original 16 months on
MS callVerdict 2026
"The supply chain, not the brands, is the trade"✅ Right — picks-and-shovels re-rated; brands still pre-revenue.
"China owns the Body / 63% of the supply chain"✅ Right & intensified — surgical dominance on magnets/reducers/screws + export controls.
"Sensors are 37% of BOM (Exhibit 54)"✅ Right in data, wrong in prose — narrative under-weighted its own number; now corrected & locked.
"$5tn by 2050 / S-curve from ~2028"🟡 Unfalsifiable but intact — near-term units now real and on-trajectory; long-run TAM still a guess.
"Joint modules = 60–70% of cost"🟠 Superseded — true at sub-system cut, misleading at component cut where sensors lead.
Memory as a Brain footnote❌ Missed — Micron & SK Hynix are now trillion-dollar index members.
No deployment / cost-track / order-book data⚪ Couldn't know — these are the 2026 additions, not MS errors.
Appendix A

The 20 names a 2026 re-write must add

In MS's own Exhibit 11/14/16 enumerations but absent from InfraMosaic v1 — now added (ms_h100 = yes), with live caps

Aptiv · Oracle · Palantir · Micron · SK Hynix · Toyota · Magna · Honeywell · Hon Hai/Foxconn · Timken · Teledyne · Valeo · Hexagon · Dassault Systèmes · Naver · GAC · Midea/KUKA · RBC Bearings · Shanghai Beite · Xusheng.