Should you substitute a domestic harmonic drive
The question buyers ask most often, and the one marketing material distorts most reliably. The answer is scenario-dependent: if the joint is replaceable under warranty and downtime is cheap, substitute. On a continuously running line where downtime costs far more than the part, domestic units are more expensive once you divide price by life. Here is the arithmetic.
Judge on cost per hour of life, not on unit price. Domestic harmonic drives are quoted 30-50% below Japanese equivalents, but the published life figures are roughly one third of Harmonic Drive Systems' benchmark. Divide one by the other and domestic units come out at around 1.5-2x the cost per operating hour - unless swapping the part is cheap or needs no downtime at all. So the real test is not "is domestic good enough", it is "what does one replacement on this joint cost you, and how long does the machine sit still". C
On this page
Last verified: 2026-10-09. The life and accuracy comparison is taken from Chinese industry research at second hand, and the sources disagree on basis (conflicts listed in the body). Grade C - demand the measured report for your own specification before signing.
1. Three numbers on the table
Start with capacity and share, because they decide whether you have a choice at all. Leaderdrive's 2025 annual-report figures: revenue of RMB 571 million, of which RMB 476 million from harmonic reducers and metal parts (+46.42% YoY), and 425,200 harmonic reducer units sold (+72.48% YoY). JPMorgan's March 2026 estimate puts its share of the Chinese market at roughly 30-40%. B
On the other side, Harmonic Drive Systems remains the reference for both global share and technical benchmarks, but its capacity expansion has been conservative - published figures for 2026 capacity are around 3.47 million units, which means lead time and capacity are themselves risk items when set against incremental humanoid demand. C
For HDS global share we found bases ranging from around 58% (2024) to around 80%, while a third study switches to a different metric entirely - "domestic substitution rising from under 5% to 20-25%". Three figures that do not reconcile means the counting basis differs (units versus value, robotics-only versus all applications). So we adopt none of them as a conclusion. We use them for one thing only: share is moving, the direction is clear, the magnitude is not.
2. The arithmetic: price divided by life
Take mid-range values and run it once: Japanese unit at a relative price of 100 and a life of 12,000 hours; domestic unit at 40% off, so 60, with a life of 4,000 hours.
| Option | Unit price (relative) | Published life | Cost per operating hour |
|---|---|---|---|
| Japanese | 100 | 12,000 h | 0.0083 /h |
| Domestic | 60 | 4,000 h | 0.0150 /h |
Basis: worked example using relative prices and midpoints of published life ranges. C The result is highly sensitive to both the discount rate and the life figure - rerun it with your own quotation and your own measured life.
The result is counter-intuitive: per hour of life, the domestic unit costs roughly 1.8x more. That is not an argument that domestic parts are bad. It is an argument that "cheaper" only holds when replacement is free and downtime is painless. The moment a swap means stripping half an arm and stopping a line, three replacements eat the entire 40% price advantage and then some. C
Not the part cost - the downtime. Three replacements means three stoppages, and most purchasing comparison tables have no row for the cost of a stopped line. Add it and the conclusion frequently flips.
3. Three scenarios, three answers
| Your scenario | Recommendation | Why |
|---|---|---|
| R&D, prototypes, small batches | Go domestic outright | Fast iteration, life never fully consumed, downtime irrelevant - and the saving buys another two configuration attempts |
| Volume machines with serviceable joints (cobots, most humanoid joints) |
Dual source, domestic-led | Joints swappable under warranty and replacement labour is bounded, so the price advantage reaches the bottom line |
| Continuous lines that cannot stop (semiconductor, medical, space-grade) |
Japanese, or domestic only on non-critical axes | Downtime dwarfs the price gap, and the accuracy and life databases are part of the entry ticket in these sectors |
Scenario split and recommendations are our own judgement. C Grounded in the life-versus-price arithmetic above and in the qualification conventions of each application domain.
4. Export control: the variable that can change the answer
This has to be checked at the selection stage, not at shipment. In May 2026 the US Bureau of Industry and Security published a proposed rule to add high-precision harmonic reducers meeting specific performance thresholds to the Commerce Control List under a new ECCN 2A001.b.5. Thresholds reported at second hand: backlash of 1 arc-minute or less, torque density of 12 N·m/kg or more, and angular repeatability within ±5 arc-seconds. Docket BIS-2026-0012, comment period closing 15 May 2026. B
As of this page's verification date we have not confirmed a final rule text. A proposal creates no licensing obligation - obligations begin only when a final rule is published in the Federal Register. But the thresholds are public, which means the high-precision tier on your selection sheet may sit inside them. What to do: check the current status of BIS-2026-0012 at regulations.gov; obtain a written classification opinion from the supplier for the specific model; and where US-origin technology or re-export is involved, build compliance review into the contract. This page offers no workaround of any kind.
5. How to do it: four steps of dual-source validation
- Do not swap the whole batch - keep a control group. Fit some machines with domestic units and some with Japanese, run a full service interval, then compare. Do not decide from datasheets.
- Ask for the life curve, not the life number. "Rated life X hours" is not comparable across vendors, because the definition is not. What you want is the failure distribution under your load spectrum.
- Put replacement labour into the comparison sheet. How long a joint takes to strip and whether it needs recalibration decides total cost more often than unit price does.
- Lock lead time, not price. Once humanoid demand arrives, capacity and delivery are the terms most likely to be breached.
6. When this judgement fails
The verdict above does not hold in three cases. Change the answer if any apply:
- Your load spectrum is shock-dominated rather than steady. Flexspline fatigue is highly sensitive to shock loading, so rated life shortens markedly under it - the life gap widens further and the domestic case gets harder to close.
- You have a measured life report for the specific model that beats the ranges above. This page uses industry ranges, not any single vendor's model data. A vendor's own measured report outranks this page.
- The end customer has nominated the supplier. In semiconductor, medical and space applications, qualification is frequently dictated by the end customer's approved vendor list. Selection freedom does not exist, and cost is not the deciding variable.
7. Sources and evidence grades
Grades: A official primary · B authoritative secondary, traced to the origin · C our own derivation, not citable as fact.
- Leaderdrive 2025 annual-report figures: revenue RMB 571 million, of which RMB 476 million from harmonic reducers and metal parts (+46.42% YoY); 425,200 harmonic reducer units sold (+72.48% YoY). B Retrieved via an encyclopaedia entry at second hand; the annual report itself was not consulted directly. Verify against the exchange disclosure before citing or contracting.
- JPMorgan, March 2026: Leaderdrive's share of the Chinese harmonic reducer market estimated at 30-40%. B
- Price gap of 30-50%; domestic life of 3,000-5,000 hours; HDS life of 10,000-15,000 hours; HDS 2026 capacity of 3.47 million units; transmission accuracy (domestic around 1 arc-minute versus under 30 arc-seconds for Japanese). C All from Chinese industry research at second hand, with inconsistent bases across sources. Not usable as acceptance criteria.
- BIS proposed ECCN
2A001.b.5and thresholds (backlash ≤1 arc-minute, torque density ≥12 N·m/kg, repeatability within ±5 arc-seconds), docketBIS-2026-0012. B Via trade media at second hand; final status not verified on the primary Federal Register page. Check regulations.gov yourself. - The cost-per-hour arithmetic, the three scenario recommendations, and the dual-source validation steps. C All our own derivation; assumptions stated in the body.
Related
Last updated 2026-10-09. Revised when the BIS proposal status, vendor annual results or capacity figures change.