Investment ResearchTopic research, evidence curves, and cycle signals
HomeSign inEN

Research library · updated 2026-06-12 · public

Humanoid OEM make-vs-buy public compression brief v1

Date: 2026-06-12 Owner: Hugo / Genius Team Agent: Finance / Charlie AGT-002 Status: public-safe slide/site compression brief Visibility: PUBLIC Source discipline: 🟢 primary / 🟡 secondary / 🟠 estimate / 🔴 guess Public-safety: safe if kept as evidence map; no trade recommendation; no Hugo portfolio data. Related source files:

  • knowledge/robotics-oem-make-vs-buy-v1.md
  • knowledge/robotics-tesla-optimus-primary-claim-map-v1.md
  • knowledge/robotics-figure-ai-deep-dive-v1.md
  • knowledge/robotics-unitree-deep-dive-v1.md
  • knowledge/robotics-leaderdrive-case-study-slide-evidence-v3.md

0. Why this brief exists

Codex already has deep evidence packets. The packaging risk is compression error: turning “OEM self-designed modules + supplier product fit” into “specific supplier is confirmed for Tesla / Figure / Unitree.”

This brief gives one public-safe slide/site block that links OEM make-vs-buy behavior to supplier economics without making unsupported supplier-customer claims.

1. One-line answer

截至 2026-06-12，公开证据更支持这个判断：humanoid OEM 不是简单采购标准化“整套关节模组”，而是在自定义 / 自研关键模块的同时采购 individual components、piece parts、materials、electronics、machining、testing equipment 等外部供应链能力。Figure 的 primary evidence 最强；Tesla 的 production-line / designed-capacity 证据强但 BOM/supplier 未披露；Unitree 的 price/access 证据强但供应商拆分和工业部署经济性仍缺。🟢/🟠

Public-safe wording:

Humanoid 供应链不是一张“赢家供应商清单”。头部 OEM 正在定义自己的关键模块，供应商仍然重要，但价值捕获要看 customer confirmation、volume、value content、ASP、margin 和 repeatability。

Do not say:

Tesla / Figure / Unitree 已确认会大量采购某个 A 股供应商的 reducer / actuator。

Reason: reviewed public sources do not confirm named supplier relationship or humanoid-specific revenue/order economics for Tesla, Figure, Unitree and 绿的谐波 / Leaderdrive. 🟢/🟠 as of 2026-06-12.

2. Slide-ready evidence table

Evidence curveWhat public evidence supportsWhat it does not supportSignal grade
Figure AI make-vs-buyFigure says Figure 03 critical module builds including actuators, batteries, sensors, structures, electronics are vertically integrated and designed completely in-house; suppliers support individual components. 🟢 Figure official, 2025-10-09Does not identify reducer / motor / battery-cell / sensor / PCB suppliers; does not prove external complete actuator-module purchases. 🟢/🟠S4 module-architecture signal
Figure BotQ manufacturingFigure says BotQ line initially capable of up to 12,000 humanoids/year and goal of 100,000 robots over four years; BotQ post says core assembly includes actuators, hands, batteries, final assembly, while outside vendors support piece-part manufacturing when needed. 🟢 Figure official, 2025-03-15 and 2025-10-09Capacity/goal language is not current production, revenue, margin, or customer ROI. 🟢/🟠S4 manufacturing-path signal
Tesla OptimusTesla Q1 2026 update disclosed Fremont first-generation Optimus line designed for 1M robots/year and Texas second-generation line with long-term designed annual capacity of 10M robots. 🟢 Tesla Form 8-K Exhibit 99.1, filed 2026-04-22Does not disclose BOM, reducer / actuator suppliers, actual output, yield, robot revenue, robot gross margin, external paid customers, or customer payback. 🟢/🟠S4 capacity-intent signal
Unitree / 宇树Official product pages list R1 from US$4,900, G1 from US$13.5K, H2 US$29,900; H2 Plus / G1-D reference NVIDIA Jetson / Isaac / TeleOp / Sim / data toolchain. 🟢 accessed 2026-06-10Does not disclose official BOM supplier split, humanoid shipment volume, industrial runtime KPI, gross margin, service cost, or repeat-order economics. 🟢/🟠S3 cost/access + platform signal
绿的谐波 / Leaderdrive2025 annual report disclosed revenue RMB 570.714m (+47.31%), net profit RMB 124.367m (+121.42%), harmonic reducer sales 425,158 units (+72.48%), and “工业及具身智能机器人零部件” revenue RMB 422.528m with 34.88% gross margin. 🟢 disclosed 2026-04-23Does not disclose named Tesla / Figure / Unitree customer, humanoid-specific revenue, humanoid order volume, per-robot value content, or humanoid-specific margin. 🟢/🟠S4 filing-backed supplier signal

3. What this means for supplier research

Signal

  • OEM official language that defines which modules are designed in-house and which parts are externally sourced. Example: Figure critical modules vs individual components. 🟢
  • Supplier filings showing revenue, unit sales, gross margin, and product lines relevant to embodied-intelligence robot components. Example: 绿的谐波 2025 annual report. 🟢
  • Named supplier / customer confirmation with order amount, volume, contract scope, or production part number. 🟢
  • Teardown / certification / customs evidence with identifiable part numbers and production hardware linkage. 🟡/🟢 depending source quality.

Noise / insufficient evidence

  • “Optimus supplier list” without Tesla, supplier filing, or identifiable teardown evidence. 🔴
  • “Every humanoid needs N harmonic reducers” without architecture, model generation, and supplier validation. 🟠/🔴
  • Stock-price move used as order evidence. 🔴; 绿的谐波 2026-06-09 abnormal-trading announcement said no undisclosed major matter after 2026-06-04 to 2026-06-08 share price +36.00% vs SSE Composite -3.05% and automation-equipment index +2.13%. 🟢
  • Supplier product fit treated as confirmed customer revenue. 🟠/🔴

4. Public-safe slide block

Title: Humanoid supply chain: product fit is not customer proof

Subtitle: OEMs are self-defining critical modules; suppliers still matter, but value capture must be verified.

Three boxes:

  1. OEM architecture signal

    • Figure says critical modules are in-house designed; Tesla discloses production-line ambition but not BOM; Unitree discloses low-cost platform specs, not supplier economics. 🟢/🟠

  2. Supplier validation signal

    • 绿的谐波 has filing-backed revenue / shipment / margin evidence in robot-relevant component categories, stronger than concept-stock language. 🟢

  3. Missing before S5

    • Named customer, order volume, contract value, humanoid-specific revenue, per-robot value content, margin durability, repeat orders, and customer ROI/payback. 🟢/🟠

Footer:

Evidence map only. No supplier-customer inference. No trade recommendation. Product fit ≠ confirmed revenue.

5. Public-safe site draft section

为什么 humanoid 供应链不能直接写成“赢家名单”

人形机器人供应链的核心问题不是“哪个零件听起来相关”，而是“谁在实际量产硬件中捕获价值”。截至 2026-06-12，公开证据显示两条线同时成立：一方面，Figure 等 OEM 正在把 actuators、hands、batteries、electronics、final assembly 等关键模块高度内化；另一方面，外部供应商仍可能在 precision parts、materials、motors、electronics、machining、testing equipment、piece-part manufacturing 等层面提供价值。🟢/🟠

这会改变投资研究的顺序：先看 architecture fit，再看 customer confirmation，然后才看 volume、value content、ASP、gross margin 和 repeatability。绿的谐波是一个很好的教学案例：报表里的 robot-relevant revenue、harmonic reducer unit sales 和 margin 是强信号；但它还不是 Tesla / Figure / Unitree 等特定 humanoid OEM 的公开确认供应商，也没有披露 humanoid-specific revenue 或 order economics。🟢/🟠

6. What would change our mind

Upgrade toward S5 if:

  1. Tesla, Figure, Unitree, UBTECH, AgiBot, or another leading OEM publicly names a reducer / actuator / motor / sensor supplier with volume, production hardware, or contract scope. 🟢
  2. A supplier filing discloses humanoid-specific revenue, backlog, customer concentration, or order amount. 🟢
  3. Production teardown / certification / customs evidence identifies actual production components and supplier names, not just prototype speculation. 🟡/🟢
  4. Supplier robot-related revenue scales while gross margin stays stable or improves, rather than being offset by ASP pressure or working-capital stress. 🟢/🟠
  5. OEM language shifts from in-house critical module design to externally purchased module platforms. 🟢

Downgrade if:

  1. OEMs continue internalizing actuators / hands / electronics and outsource only low-margin piece parts. 🟢/🟠
  2. Supplier revenue growth comes with material gross-margin compression, inventory buildup, or receivables stress. 🟢/🟠
  3. Customer exposure remains anonymous and unquantified across multiple reporting periods. 🟢
  4. Robot architectures reduce harmonic reducer content through linear actuators, quasi-direct-drive, tendon/cable systems, or proprietary integrated modules. 🟠

7. Common misconceptions

Misconception 1: “只要 humanoid 放量，谐波减速器供应商一定线性受益。”

Correction: volume matters only after architecture fit, attach rate, customer confirmation, ASP, and margin durability are verified. 🟠

Misconception 2: “Figure 自研关键模块，所以供应商不重要。”

Correction: Figure’s own posts distinguish critical module in-house design from suppliers for individual components / piece parts. Supplier value may move down-stack rather than disappear. 🟢/🟠

Misconception 3: “Tesla 产能规划大，所以可以倒推出供应商订单。”

Correction: Tesla disclosed designed capacity / production-line intent, not BOM, suppliers, actual output, or order commitments. 🟢/🟠

Misconception 4: “绿的谐波报表强，所以已经是已验证 humanoid winner。”

Correction: filing-backed growth is meaningful, but named humanoid OEM, humanoid-specific revenue, contract value, order volume, and margin remain undisclosed. 🟢/🟠

8. Think Deeper questions

  1. 如果 OEM 把 actuator module 设计内化，外部供应商的价值会从完整模组迁移到哪些 piece parts / equipment / materials？
  2. 哪个指标最能区分“产品适配”与“客户确认”：named customer、production part number、order value、revenue split，还是 gross margin stability？
  3. 如果 humanoid 量上升但 supplier margin 下降，应该被理解为规模化成功还是议价能力下降？
  4. Tesla/Figure/Unitree 的 make-vs-buy 差异，会让不同国家供应链的价值捕获路径出现结构性分化吗？

9. Source list

Primary / official sources:

Internal synthesis sources:

  • knowledge/robotics-oem-make-vs-buy-v1.md 🟠 synthesis, dated 2026-06-12.
  • knowledge/robotics-leaderdrive-case-study-slide-evidence-v3.md 🟠 synthesis, dated 2026-06-11.
  • knowledge/robotics-public-claim-bank-v1.md 🟠 synthesis, dated 2026-06-11.

10. Public-safety check

Safe to package if:

  • No Hugo private portfolio data.
  • No buy / sell / hold language.
  • No supplier-customer inference without primary confirmation.
  • No claim that designed capacity equals actual production.
  • No claim that product fit equals revenue.
  • All material claims preserve source grades and as-of dates.