Harmonic Drive Systems / 株式会社ハーモニック・ドライブ・システムズ (6324)

The strain-wave gear near-monopoly — the core humanoid joint reducer, bought through a cyclical earnings trough as the secular volume story arrives.

Investment Research · Photoncap-style deep dive · v1 of "Harmonic Drive Systems" · May 14, 2026

What Harmonic Drive Systems physically does

Every articulated robot joint needs to do one thing: take a small, fast, low-torque electric motor and convert it into slow, powerful, precisely-controlled motion. The component that does this is a reducer (a gearbox), and for compact joints — robot wrists, cobot arms, and now humanoid shoulders, elbows and hands — the dominant architecture is the strain-wave gear, also called a harmonic drive. Harmonic Drive Systems did not just commercialise this device; the company's name is the generic term for it.

A strain-wave gear has three parts and no conventional gear train. A stiff outer ring (the circular spline) has internal teeth. Inside it sits a thin, flexible steel cup (the flexspline) with slightly fewer external teeth. And inside that sits an elliptical bearing-shaped plug (the wave generator) mounted on the motor shaft. As the wave generator spins, it flexes the cup into an ellipse, so the cup's teeth mesh with the ring's teeth only at the two ends of the long axis. Because the cup has, say, two fewer teeth than the ring, one full rotation of the motor advances the output by just two teeth — a reduction ratio of 50:1 or 100:1 in a single, flat, lightweight stage with effectively zero backlash. That combination — high ratio, no backlash, low weight, coaxial input and output — is exactly what a humanoid robot's distal joints require, and almost nothing else delivers it in the same package.

The "binding constraint" framing matters here. A humanoid robot built on the Tesla Optimus reference architecture uses roughly 28 actuators, of which around 14 are rotary; the rotary actuators in the arms, wrists and hands are overwhelmingly strain-wave designs. Harmonic Drive Systems sells the gear itself, the gear-plus-bearing unit, and increasingly the complete actuator (gear, frameless motor, encoder and driver integrated). It does not make the RV cycloidal gears used in heavy industrial-robot base joints — that is Nabtesco's territory — and it does not make linear motion. It is a pure-play on the rotary precision joint, which is both its strength and its concentration risk.

Product roadmap

The core catalogue is the HarmonicDrive component series — the CSF/CSG/SHF families for general robotics and the CPL/CPU "pancake" units for flat joints — plus the CobotPro line launched in 2022 specifically for collaborative-robot integrators who want a drop-in gear-and-bearing module. Above the bare gear, Harmonic Drive Systems sells the FHA and RSF rotary actuators, which integrate an AC servo motor, the strain-wave gear, an encoder and a brake into one housing — the product the company wants humanoid OEMs to buy, because it captures far more bill-of-materials value than the gear alone.

The roadmap event that matters for this theme is the high-torque micro actuator co-developed with MinebeaMitsumi, unveiled for first public demonstration at CES 2026 (January 2026). It is a geared micro actuator measuring 13mm wide, 19.4mm high and 60.4mm long — sized for humanoid robot hands and fingers, the highest-actuator-count, highest-unit-volume part of the humanoid body. This is the clearest signal that Harmonic Drive Systems is positioning for the fingertip-level volume explosion rather than just the shoulder-joint count. Capacity is following: FY2026 capital investment is guided at ¥6,100 million, up 62% year-over-year, explicitly to build humanoid-driven reducer capacity. The medium-term plan, confirmed at the May 2025 results briefing, targets ¥90 billion in sales and ¥15 billion in operating profit by FY03/27 — a roughly 60% revenue increase over the FY2025 trough.

The financial print

Harmonic Drive Systems closed FY2025 (fiscal year ended March 31, 2025; results released May 20, 2025) with net sales of ¥55.65 billion, down 0.3% year-over-year, and — this is the headline — operating profit of ¥1,967 million, down 94.4% from ¥28,577 million the prior year. Operating margin collapsed from roughly 51% to 3.5%. Net income was ¥3.47 billion. This was not a demand failure; it was a savage operating-leverage unwind. The semiconductor-equipment correction hit the high-margin FPD/semi reducer business, the company kept SG&A and headcount up for the coming humanoid ramp, and the result was a near-total profit wipeout on roughly flat revenue. That is the trough the stock is being bought through.

The recovery is already visible in the run-rate. Cumulative 9-month FY2026 (April-December 2025) net sales reached ¥42.18 billion, up 4.5% year-over-year, with management citing surging demand from robot manufacturers in Japan, China and Europe plus a semiconductor-equipment recovery — and noting that orders to robot makers exceeded the earnings-forecast assumption. First-half FY2026 net sales were ¥27.84 billion, up 4.8%. The forward consensus, against the FY03/27 medium-term target of ¥90 billion revenue and ¥15 billion operating profit, implies operating margin rebuilding toward the high teens. At a JPY 6,430 share price and roughly JPY 608 billion market cap, the forward P/E of 85.0 is a trough-earnings multiple, not a steady-state one — the bull case is that FY03/27 normalised EPS makes the multiple look very different. The binary is the FY2026 full-year results, due on or around May 19, 2026 — the print that confirms whether the robot-order momentum is translating to margin.

Customer mix today

Harmonic Drive Systems does not publish a clean customer-name breakout, so the mix is best read by end-market on a reporting basis. Industrial robot manufacturers — the FANUC, Yaskawa, ABB, KUKA tier and their Chinese equivalents — are the largest bucket at roughly 45% of revenue. Semiconductor and flat-panel-display manufacturing equipment is roughly 25%, and it is the most volatile slice: it was the swing factor in the FY2025 profit collapse. Aerospace, optical, medical and automotive together make up the remaining ~30%.

Humanoid-specific robotics is still a small slice — realistically under 5% of FY2025 revenue — but it is the fastest-growing one and the entire reason the multiple is where it is. The structural shift is the story: in 2023 humanoid was a rounding error and the investment case was "GDP-plus industrial automation compounder." By early 2026, with the MinebeaMitsumi micro-actuator co-development, the 62% capex step-up, and management explicitly attributing order upside to robot makers, the mix is tilting. Goldman Sachs Japan's framing — that each humanoid needs dozens of precision actuators and that this could expand Harmonic Drive Systems' addressable market five-to-ten-fold over a decade — is the consensus bull narrative. Treat the "5-10x TAM" as an estimate, not a forecast; what is confirmed is that the order book is inflecting and the company is spending against it.

What's actually happening at the humanoid OEMs

The uncomfortable truth sits at Tesla. Tesla's Optimus is the single largest potential demand pool for compact reducers, and Tesla has not standardised on Harmonic Drive Systems. Per multiple supply-chain reports, Tesla's primary harmonic-reducer supplier for Optimus is China's Suzhou Green Harmonic, which prices 30-40% below the Japanese incumbent and is reportedly targeting roughly 60% of the Optimus reducer market by 2026, with a new Suzhou plant ramping to 500,000 units of annual capacity. Suzhou Green Harmonic alone is cited at roughly 25% global harmonic-drive share. If the highest-volume Western humanoid program is being supplied predominantly by a Chinese low-cost vendor, the bull case for Harmonic Drive Systems cannot rest on Optimus.

Where Harmonic Drive Systems does win is the precision tier: programs where joint accuracy, lifetime and zero-backlash repeatability cannot be compromised, and the higher-mix Western and Japanese humanoid developers (Figure, Agility, Apptronik, and the Japanese cobot ecosystem) plus the entire installed base of industrial-robot OEMs who already qualified Harmonic Drive Systems decades ago. The MinebeaMitsumi micro-actuator partnership is the company's attempt to push down into the fingertip-count volume tier with a cost-engineered product rather than ceding it entirely. The qualification cycles here are long — a humanoid OEM that designs in a reducer keeps it for years — so the 2026-2027 design-win season is the period that sets share for the volume ramp. That is the mechanism to watch.

The competitive threat / Leaderdrive and Suzhou Green Harmonic

The competitive threat is the most concrete bear case on this name and it is not subtle. Chinese strain-wave makers — Leaderdrive (Shenzhen-listed), Suzhou Green Harmonic, and Zhejiang Laifual — have moved from under 5% of units installed in Chinese-assembled robots in 2018 to more than 35% by 2024, per industry data, with Leaderdrive taking the majority of that domestic share and roughly 15% of the global market by 2023. Domestic Chinese harmonic reducers are priced at roughly 40-60% of equivalent Harmonic Drive Systems units. Laifual launched miniature harmonic drives specifically for humanoid robotics in December 2024. This is a textbook oligopoly-erosion-by-cost-competition story, and it is happening in real time.

The counter-argument — and it is a real one, not a hopeful one — is that Chinese reducers still trail on precision-tier metrics: backlash consistency over lifetime, torque ripple, and the reliability data that a Western humanoid OEM needs before committing to a multi-year production design. Harmonic Drive Systems' moat is the qualification base and the metallurgy, not the patent (the core patents expired long ago). But "the Japanese product is better" is exactly what the Japanese machine-tool and the Japanese solar industries said before they lost. The honest read: Harmonic Drive Systems keeps the precision tier and loses the volume tier on price, and the investment question is whether the precision tier grows fast enough in absolute yen to overwhelm the share loss. That is a genuinely open question.

The terminal risk

The terminal risk is not a technology transition away from strain-wave gears — humanoid robots will need compact rotary reducers for the foreseeable future, and the strain-wave architecture is hard to displace for distal joints. The terminal risk is margin, not volume: that the humanoid era arrives, Harmonic Drive Systems' unit volumes genuinely 5-10x, and yet the company earns a commodity-component margin on it because Chinese capacity has set the global clearing price 40-60% below the historical Japanese level. In that world the stock is a low-margin volume compounder dressed up as a monopoly, and the 85x trough multiple was a trap. The secondary terminal risk is architectural: some humanoid designers (and the broader 2026 industry literature) are pushing toward integrated linear actuators and planetary roller screws for high-force hip and knee joints, which would route incremental humanoid actuator dollars to THK and the screw makers rather than to strain-wave. Harmonic Drive Systems' credible defence is the MinebeaMitsumi micro-actuator (cost-engineered, hand-targeted) and the medium-term capex — but the defence is unproven until FY03/27 margins print.

Bull / Gap / Optionality (Photoncap framing)

Bull

1. The single most-levered listed pure-play on humanoid joint count. Every credible humanoid architecture uses compact strain-wave reducers in the upper body, and Harmonic Drive Systems is the brand-defining incumbent. Goldman Sachs Japan's five-to-ten-fold TAM-expansion framing — driven by dozens of precision actuators per robot — is the structural case, and unlike most "robotics exposure" names, there is no segment dilution here: this is what the company does.

2. You are buying the cyclical trough, not the peak. FY2025 operating profit collapsed 94.4% to ¥1,967 million on a semiconductor-equipment correction with SG&A held up for the ramp. The medium-term plan targets ¥90 billion revenue and ¥15 billion operating profit by FY03/27 — roughly 8x the FY2025 operating profit. The 85x forward P/E is a trough multiple; on FY03/27 plan earnings it compresses dramatically.

3. Order momentum is already inflecting. 9-month FY2026 sales rose 4.5% to ¥42.18 billion with management explicitly stating orders to robot makers exceeded forecast assumptions, across Japan, China and Europe. The semiconductor-equipment business is also recovering. The recovery is in the run-rate, not just the slideware.

4. The MinebeaMitsumi micro-actuator is a credible volume-tier play. The jointly-developed 13×19.4×60.4mm geared micro actuator, demonstrated at CES January 2026, targets humanoid hands and fingers — the highest unit-count part of the body. Co-developing with MinebeaMitsumi's manufacturing scale is the right answer to the cost problem, and it shows management understands the threat.

5. Capex is being spent against the thesis. FY2026 capital investment guided up 62% year-over-year to ¥6,100 million, explicitly for reducer capacity. Management is not merely talking the humanoid story; it is committing balance sheet to it ahead of the volume.

Gap

1. Chinese cost competition is eroding the oligopoly in real time. Domestic Chinese harmonic makers went from under 5% of Chinese-robot installs in 2018 to over 35% by 2024; Leaderdrive holds roughly 15% global share; Chinese units price 40-60% below Harmonic Drive Systems. This is not a future risk — it is a現在-tense share loss, and it caps the achievable margin.

2. Tesla Optimus — the largest single demand pool — is not a Harmonic Drive Systems program. Suzhou Green Harmonic is Optimus's primary harmonic-reducer supplier, targeting ~60% of that market by 2026 at 30-40% below Japanese pricing, with a 500,000-unit Suzhou plant ramping. The biggest Western humanoid volume story is largely closed to this name.

3. The valuation requires the recovery to be flawless. At 85x forward earnings on a trough base, any slippage in the FY03/27 ¥15 billion operating-profit target — a semiconductor double-dip, a slower humanoid design-win season, yen strength compressing export margins — leaves the multiple badly exposed. There is no valuation cushion.

4. The architecture could route humanoid dollars elsewhere. The 2026 industry shift toward integrated linear actuators and planetary roller screws for high-force hip/knee joints would direct incremental humanoid actuator spend to THK and screw makers, not to strain-wave. Harmonic Drive Systems owns the upper-body joint, but the upper body may not be where the BOM dollars concentrate.

Optionality

The trade

Harmonic Drive Systems is the cleanest listed expression of the humanoid-joint thesis and simultaneously the name most exposed to the Chinese cost attack — which is precisely why it is a Bucket B, not Bucket A. Initiate at JPY 6,110-6,750 (current JPY 6,430 ±5%, RSI 69.1 is momentum in an early-cycle theme, not a sell signal), size at 1.5% of risk capital given the 85x trough multiple leaves no error margin, and stop at JPY 5,300 (below the 50-day moving-average structure and the recent consolidation floor). The defining binary is the FY2026 full-year print around May 19, 2026 — the number that shows whether robot-order momentum is converting to margin against the ¥15 billion FY03/27 target. If the thesis is right but you want a less single-point-of-failure expression, MinebeaMitsumi (6479) gives you the same humanoid micro-actuator optionality bolted onto a ¥1.6 trillion diversified bearings-and-motors base trading at 18x — a far cheaper way to own the same partnership, with the joint-layer concentration risk diversified away. Conviction: 7 / 10.

Sources referenced inline throughout: Harmonic Drive Systems FY2025 results briefing (May 20, 2025) and 9M FY2026 disclosure; Goldman Sachs Japan robotics framing; CES 2026 MinebeaMitsumi co-development announcement; Tesla Optimus supply-chain reporting (Suzhou Green Harmonic); industry data on Leaderdrive / Laifual Chinese strain-wave share. Reference v1 of this template format: _Watchlist/hanmi-photoncap-style.md.

\newpage

6479 — MinebeaMitsumi / ミネベアミツミ株式会社 · BUY (Tier-1) · Conv 7/10 · Bucket B

ticker: 6479 name: MinebeaMitsumi / ミネベアミツミ株式会社 theme: Robotics bucket: B conviction: 7 entryzonelo: 3779 entryzonehi: 4177 currentprice: 3978 pricedate: 2026-05-14 positionsizepct: 1.5 stoploss: 3350 thesisoneline: The miniature-ball-bearing world leader plus a co-developed humanoid micro-actuator with Harmonic Drive — humanoid optionality bolted onto a ¥1.6T diversified base at just 18x. catalystnext: FY2026 (FYE Mar 2026) full-year results catalystdate: 2026-05-13 rsi: 84.4 vs50ma: 35.3 forwardpe: 18.5 themecycleposition: early customermixsummary: Precision components (bearings) ~core margin engine; electronic devices & components and semiconductors the balance; humanoid still tiny but the micro-actuator is the entry vehicle. terminalriskoneline: The eight-business conglomerate structure diluting any humanoid upside, while Chinese miniature-bearing makers commoditise the legacy bearing base. bulldriverscount: 5 gapriskscount: 4 optionalitycount: 5 lastearningsdate: 2026-05-13 nextearnings_date: 2026-08-07