AGI Investment Tracks · Track 7

Manufacturing & Industrial Automation

Analyst: manufacturing-automation · Date: 2026-05-26 · For: Ravi (AGI-pilled lens)
TL;DR. Reasoning from "AGI is happening," the binding constraint shifts from cognition to the physical install base that converts designs into hardware. Pure automation/PLC names (Rockwell, Siemens, ABB) already have the AI-data-center bid; the truly under-priced layer is the motion-control supply chain that gates humanoid robot scaling — precision reducers, ball screws, linear motion, and high-spec actuators where Japanese/Taiwanese/German incumbents have 5-to-10-year manufacturing know-how moats. Industrial AI demand also re-rates the test-and-measurement layer (KLA/Cognex/Keyence) and specialty alloys (Carpenter, Howmet, RBC Bearings). My top three highest-conviction picks: Nabtesco (6268.T), THK (6481.T), and Keyence (6861.T), with HIWIN (2049.TW) and Carpenter Technology (CRS) rounding out the five.

1. Thesis

If AGI lands inside 2-3 years and recursive self-improvement is already starting, the next binding constraint is the world: physical things that have to be built — humanoid robots, fabs, reactors, satellites, EVs, data centers. A superintelligence can design the optimal factory in minutes; it cannot conjure the install base of CNC machine tools, harmonic reducers, RV reducers, ball screws, linear guides, encoders, vision systems, and motion controllers that the design depends on. Those install bases were laid down over 30-50 years by Japanese and German specialists who have no real challenger in the West, and demand for them is now reflexively levered to robotics, AI-data-center mechanical content, and re-shored chip fabs.

The cleanest way to express this is not the famous AI-bid names (Rockwell, Siemens, Eaton, Celestica) which have already substantially re-rated. It is the 2nd-derivative motion-control suppliers whose units-per-humanoid bill of materials is high (Nabtesco ~25-30 actuators × $$$$ per robot, THK linear motion in every humanoid leg/arm, HIWIN ball screws), plus quality-control sensing (Cognex, Keyence) which is the only way you can run an unmanned factory. Specialty alloys (Carpenter, Howmet, RBC Bearings) are the unsexy commodity layer that AI capex can't substitute around.

2. The Bottlenecks (Quantitative)

BottleneckData pointTagImplication
Eaton data-center backlog 228 GW — "12 years of backlog at 2025 build rates"; data-center orders +240% YoY in Q1 2026; total electrical backlog +48% YoY; book-to-bill 1.2 Fact Demand is far ahead of US electrical-equipment manufacturing capacity. Same is true for switchgear, transformers, and cable.
KLA wafer-fab equipment outlook WFE spending "$140 B+" in 2026, advanced-packaging market growth raised from ~20% to >30%; CFO Higgins: "I can't recall this level of [2027] visibility… in the early-May timeframe" in a 13-year tenure Fact Chip-equipment lead times rising. WFE up-cycle has at least 2 more years of visibility — drags every motion/optics/metrology supplier with it.
Humanoid actuator content ~25-30 actuators per humanoid (Schaeffler); $685 M (~¥5 B RMB) reported Tesla order to Sanhua Intelligent Controls implying ~180k Optimus units Estimate / unconfirmed Even at 100 k humanoids in 2026, that's 2.5-3 M precision actuators. At 1 M units by 2028 (Tesla target), 25-30 M. Today's combined harmonic-reducer global capacity is ~2 M units/yr (industry estimate).
Cognex Q1 2026 Revenue +24.3% YoY, gross margin 71%, op margin 22.3%, two new AI vision systems driving the beat Fact Vision-as-a-service for unmanned factories is real and pricing power is intact. The pure recovery cycle is now compounded by AI factory upgrades.
Siemens FY2026 guidance raise Orders +10% YoY to €21.4 B; revenue +4%; CFO now targets upper half of 6-8% revenue growth, EPS up 20¢, "driven by AI and data center demand" Fact Already priced in to a meaningful extent — but signal that automation order book is reaccelerating after 2024-25 destocking.
RBC Bearings Q4 FY2026 Revenue +18.3% to $518 M; A&D segment +41.2%; adj. EBITDA margin 32.6%; pointed to "deep backlogs" in commercial aviation/defense Fact Precision bearings are choke-point components for everything that spins — robots, jet engines, EV motors, fab tools. Pricing power is durable.
THK FY profit guidance Lifted full-year profit forecast +115% after Q1 2026 (per Japan-Earnings) Fact Structural reform + early robot/EV motion-control demand. Has been a "value trap" for 3+ years; that may be ending.
Nabtesco share-price reaction +136.5% in 12 months, market cap ¥662 B, despite revenue −3.9% TTM Fact Market already starting to assign humanoid optionality to Nabtesco's precision-reducer franchise. Re-rating mid-stream — not late.
Fanuc market-cap recovery +88% in 12 months, +35% YTD 2026 to ~$48 B Fact Classic auto-industrial cyclical is being re-rated as AI/robot pure-play. Less mispriced than 6 months ago.
Keyence operating margin FY2026 revenue ¥1.17 T (+10.4%), net margin ~38%, market cap ¥18.7 T, +25.8% TTM Fact One of the highest-quality industrial businesses on Earth. 42x P/E is rich but is the only durable compounder in factory automation with sensor / vision moat.

3. Supply / Demand Gap Model

Rough cuts of the supply gap for AGI-era physical scale. Tag each row by confidence — most of these are estimates assembled from multiple sources.

ComponentCurrent global supply2030 demand if humanoids/AI build outGapBeneficiary
Harmonic reducers (small/precision) ~1-1.5 M units/yr Est. 20-40 M units/yr if humanoids ramp to ~1 M/yr (each needs ~25 reducers, but only ~10 are harmonic-style) Spec. ~10-25× supply scale-up needed Harmonic Drive Systems (6324.T), Leaderdrive (688017.SS), Schaeffler
RV reducers (heavy joints) Nabtesco ~60% global share Est. Roughly 5-10× current for industrial + service humanoids combined Spec. 3-7× supply needed; high barrier — Nabtesco has been only credible supplier for 30+ yrs Nabtesco (6268.T) is the prime monopoly
Precision ball screws & linear guides THK + HIWIN + NSK ≈ 70% of premium global supply Est. Each humanoid uses ~6-10 linear motion elements; chip fabs use thousands per tool 3-5× scale-up, multi-year tooling lead time THK (6481.T), HIWIN (2049.TW), NSK (6471.T)
Machine vision systems Keyence + Cognex + Basler + Datalogic + Teledyne dominate Inspection per chip, per robot joint, per part — order-of-magnitude content growth Mostly throughput/sensor-supply limited, not capital limited Keyence (6861.T), Cognex (CGNX)
WFE / metrology / OSAT KLA 7.5× nearest competitor in process control $140 B+ WFE in 2026, 2027 higher; advanced packaging > 30% growth Optical sub-components are bottleneck KLA, Cognex (peripheral), Keyence (OEM)
Specialty alloys (nickel/titanium/cobalt) Carpenter, Haynes, ATI, Howmet — small, vertically-integrated Robot gears, reactor pressure vessels, aero engines, hypersonics Constrained by melt capacity, multi-year permits Carpenter (CRS), Howmet (HWM), Haynes (HAYN), ATI
EMS / contract assembly Foxconn, Flex, Jabil, Celestica, Sanmina AI server racks, robotics PCB assembly, network hardware Capacity utilizations high; capex catching up Celestica (priced in), Flex, Jabil, Sanmina
Industrial electrical content / switchgear / transformers Eaton, Schneider, ABB, Siemens, Hubbell, nVent — multi-year backlogs 228 GW DC backlog + grid build-out 10+ yr at current build rates per Eaton Eaton (ETN), Schneider, Hubbell (HUBB), nVent (NVT)

Note: gap estimates are deliberately wide. They convey direction (multiples-of-supply needed), not precise yardage.

4. Investable Public Companies

Ticker Name Mkt cap Valuation (rough) Why bottleneck Risks Priced in?
6268.T Nabtesco ¥662 B (~$4.2 B) ~42× fwd EPS est ~60% global share in RV reducers; only credible supplier for large-joint humanoids; 30+ yr know-how moat China (Leaderdrive, Laifual, Nidec Drive) ramping; revenue declining near-term; humanoid timing Partly (+136% TTM) — but units-per-robot math still has room
6481.T THK ~¥620 B est ~15-18× EPS after upgrade est #1 / #2 in linear motion guides + ball screws globally; chip fab + robot + EV exposure Historically value-trap; China LGB price competition Re-rating in motion (+115% profit guide) — mid-cycle
2049.TW HIWIN ~NT$110 B (~$3.5 B) est Cyclically depressed multiple Taiwanese ball screw / linear guide leader, lowest-cost premium supplier; humanoid demos at Computex 2026 Cyclicality, Chinese price competition, Taiwan geopolitical risk Not yet — has lagged THK/Nabtesco
6861.T Keyence ¥18.7 T (~$120 B) 42× P/E, 37× fwd fact Direct-sales sensor/vision monopoly with 38% net margins; AI-factory upgrade tailwind Expensive; Japan rates risk; succession Yes, but is the only "Buffett-style compounder" in the space
6324.T Harmonic Drive Systems ~¥350 B est Cyclical; recovering Co-inventor of strain-wave (harmonic) reducer, the standard for small robot joints Chinese harmonic suppliers (Leaderdrive) winning low-end; tech commoditizing Partly
CGNX Cognex ~$8 B est ~35-40× fwd est AI vision recovery + new product cycle; 71% gross margin Auto OEM exposure; tariff risk; insider selling Partly — Q1 +24% revenue has re-rated it
6471.T NSK ~¥350 B est Mid-teens P/E Bearings + motion; humanoid joint bearings exposure Auto cyclicality dominates No — basically auto-priced
ROLL (RBC) RBC Bearings ~$11 B est ~30× fwd est Precision bearings for A&D (+41% YoY) + industrial; deep backlog Industrial slowdown could mute pace; valuation rich A&D bid partly in; AGI/robot bid not
CRS Carpenter Technology ~$13 B est ~20-25× fwd est Specialty alloys: nickel-superalloy and Ti for aero engines, hypersonics, reactors, robot gearing; multi-year sold-out backlog Aerospace cycle peak risk; ESG/permit risk Re-rated but earnings keep beating; arguably still undervalued
HWM Howmet Aerospace ~$80 B est ~40× fwd est Turbine blades + airframe Ti structures — capacity-constrained Priced for perfection; commercial aero cyclicality Mostly yes
SHA.DE Schaeffler ~€5 B est ~10× P/E est All-in-one humanoid actuator launched at CES 2026; Neura, Humanoid Inc. deals — explicit play to be Tier-1 robotics supplier Auto exposure drags consolidated multiple; execution risk on robotics Not at all — buried under auto narrative
ETN Eaton ~$140 B est ~28× fwd est 228 GW DC backlog; data-center orders +240% Q1; NVIDIA 800V DC collab; Boyd Thermal liquid cooling Mostly priced in; price-cost lag pressure Largely yes
ROK Rockwell Automation ~$40 B est ~28× fwd est US PLC/SCADA install-base monopoly in re-shored factories End-market softness in process; OT-cyber liabilities Mostly yes
SIE.DE Siemens ~€180 B est ~20× fwd est Orders +10% to €21.4 B; raised AI/DC outlook; Digital Industries software stack Conglomerate discount; cyclical units Mostly yes
CLS Celestica ~$25 B est ~25× fwd est AI hyperscaler EMS — runaway 2024-2026 winner Customer concentration; multiple already +10x Fully — momentum trade
JBL Jabil ~$25 B est ~20× fwd est Quietly winning AI server, custom silicon, optical — less crowded than CLS Customer concentration; auto/EV drag Partly
FLEX Flex ~$20 B est ~16× fwd est Power & data-center modular hardware; cheap optionality Mix exposure; energy capex risk Partly
KEYS Keysight ~$30 B est ~28× fwd est Test & measurement for chips, networking, optical — every AI rack tested Cyclical; defense exposure mixed Partly
ANSS / ADSK / PTC / DSY.PA Industrial CAD / digital twin $30-50 B each est 30-40× fwd est Software backbone of AI-designed factories AI/agent disruption of CAD seats; ANSS Synopsys deal closed Mostly yes; AI substitution risk underrated
GWW / FAST Grainger / Fastenal ~$50 B / $50 B est 25-30× fwd est MRO picks-and-shovels supply for everything that gets built/repaired Disintermediation by Amazon Industrial; tariffs Mostly yes

5. Pre-IPO / Private to Watch

6. Commodity & Physical Plays (link to materials track)

7. Talent / Know-how Concentration

The honest read: the people who know how to grind a 5-meter precision ball screw to 1-µm tolerance or vacuum-melt a 30-ton nickel-superalloy ingot are concentrated in a few small cities. AGI cannot retrain them overnight; even if AGI can simulate the metallurgy, somebody has to operate the furnace.

8. Top Picks — Ranked by Conviction

#1 — Nabtesco (6268.T) HIGH conviction

Why: Closest thing to a monopoly in RV precision reducers for industrial-grade humanoid joints. If even 30% of the optimistic humanoid ramp (≥1 M units/yr by 2028-29) is real, Nabtesco's volumes go up 3-5× on a near-fixed cost base. Already +136% in 12 months but TTM revenue still negative — the market is paying for an option, not earnings. Earnings catch-up is the next leg.

Position sizing logic: Standalone is fine; pair-trade against Harmonic Drive Systems if you want to hedge harmonic-vs-RV mix risk.

Bear case: Humanoids slip 3-5 years, Chinese RV reducers commoditize the category before scale arrives. In that world Nabtesco trades at ~25× normalized = ~30% downside.

#2 — THK (6481.T) HIGH conviction

Why: Just lifted full-year profit guidance 115% on a single quarter — a sign the multi-year value-trap pattern is breaking. Three independent demand vectors compounding: (1) chip-fab linear motion (KLA et al.), (2) humanoid leg/arm linear motion, (3) EV servo content. Cheaper than Nabtesco on every multiple. Less narrative attention because it's not a pure humanoid play.

Bear case: Auto-cycle weakness drags consolidated print; reform stalls. ~25% downside.

#3 — Keyence (6861.T) HIGH conviction (lower IRR, lower variance)

Why: The single highest-quality industrial business globally — 38% net margins, direct sales force, ~30% ROIC for two decades. AI-factory upgrade cycle (per Cognex Q1) is now hitting Keyence directly. Owning Keyence over 5 years has been the right answer at almost any entry price in the last 15 years, and that's likely still true if AGI compresses physical-buildout timelines.

Bear case: 42× P/E expects continued double-digit growth. JP rates lift, multiple compresses. ~25-30% downside.

#4 — HIWIN (2049.TW) MED-HIGH conviction

Why: The "value version" of THK. Same drivers; smaller; cyclically depressed; Computex 2026 robotics push gives it explicit humanoid narrative. Asymmetric — if Asia robotics ramp is even half-real, this re-rates hard. Taiwan geopolitical risk is real but partially diversifiable across the portfolio.

#5 — Carpenter Technology (CRS) MED conviction

Why: Sold-out backlog in nickel-superalloys and titanium that nobody can substitute for. Aerospace engine + hypersonics + reactor demand all hitting at once. Has re-rated but earnings keep beating. Less directly "AGI-themed" than the motion-control names, but a clean way to own the unsubstitutable-materials layer.

Honorable mentions (next tier)

9. What Would Change My Mind

10. Sources

All accessed 2026-05-26. Primary sources marked as Fact; analyst summaries and aggregator pages as Estimate.


This page is research for personal portfolio construction. Not investment advice. Estimates marked est or spec are explicitly back-of-envelope. Every "humanoid TAM" number in this market is suspect; treat order-of-magnitude as the useful signal, not point estimates.