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

Research library · updated 2026-06-14 · public

Robotics Mainline Q0-Q5 Evidence Matrix v1

Date: 2026-06-14 Owner: Hugo / Genius Team Agent: Finance / Charlie AGT-002 Status: SYNTHESIS_CANDIDATE Visibility: PUBLIC, if used without private portfolio information Target use: research map update, Hugo review, later site-first synthesis candidate

Source discipline:

  • 🟢 primary / official / filing
  • 🟡 credible secondary
  • 🟠 Charlie synthesis or calculation from cited evidence
  • 🔴 unsupported / guess

Public-safety:

  • No Hugo private portfolio data.
  • No paid-report excerpts.
  • No unverified supplier rumors.
  • No buy / sell / hold language.
  • Evidence map only; not investment, legal, or tax advice.

0. One-line answer

按新版 robotics-research-map 主线，当前 Q0-Q5 可以压缩为一个证据矩阵：robotics 已从 demo narrative 进入可研究的 S3/S4 证据阶段，但还没有跨层进入 S5 回报周期；最重要的研究任务不是继续堆公司，而是把 timing、value layer、Tesla capacity curve、Figure deployment/manufacturing curve、Leaderdrive supplier analogy、China/US deployment-vs-frontier split 放在同一张 upgrade/downgrade map 上。🟠 synthesis from 🟢/🟡 sources reviewed 2026-06-14.

1. Mainline evidence matrix

QuestionWorking answerBest current evidenceWhat it changesMissing before stronger conclusionStatus
Q0. Cycle stage / timing行业已经从 S2/S3 demo-heavy 进入 S3/S4 measurable-evidence phase；仍未进入 broad S5 economics phase。Figure BMW 1,250+ runtime hours / 90,000+ parts / 30,000+ X3 vehicles 🟢; Figure BotQ 350+ Figure 03 / 1 per hour / >80% EOL FPY 🟢; Tesla Fremont 1M/year and Texas 10M/year designed lines 🟢; Leaderdrive 2025 revenue RMB 570.714m +47.31% and harmonic reducer sales 425,158 +72.48% 🟢.Timing question should be framed as “construction cycle vs return cycle,” not “is robotics real?”Accepted units, paid deployment value, repeat orders, intervention rate, customer ROI/payback, robot revenue, gross margin, service burden.SYNTHESIS_CANDIDATE
Q1. Which layer?“Robotics” is not one market; evidence sits across OEM body, AI/model/data, components, deployment/integration, customers, and public-market proxies.NVIDIA Isaac GR00T provides open data/pipelines, open foundation model, simulation, middleware, CUDA-X runtime, Jetson Thor 🟢; Unitree price/toolchain evidence in prior files 🟢; Leaderdrive filing-backed component revenue 🟢; BMW/Figure customer workflow evidence 🟢.Prevents false winner ranking; value-capture analysis must be layer-specific.Layer-by-layer revenue/margin capture, supplier attach, software/data monetization, integrator/service economics, customer productivity proof.SYNTHESIS_CANDIDATE
Q2. Tesla OptimusTesla has the strongest filing-backed capacity-intent curve; not S5 because output/economics are missing.Tesla Q1 2026 Form 8-K Exhibit 99.1: “infrastructure and AI software” for Robotaxi and future robotics; Optimus progress ahead of mass production; Fremont first-generation line and Texas second-generation long-term designed capacity in prior Tesla artifact 🟢.Upgrades Tesla from demo-only to S4 manufacturing-infrastructure evidence.Actual Optimus output, yield, utilization, internal deployment KPI, intervention rate, productivity/payback, external customers, robot revenue/margin.RESEARCH_ONLY / existing deep pack already sufficient
Q3. FigureFigure has the cleanest public customer-site deployment KPI plus unusually concrete manufacturing KPI; still missing S5 economics.BMW: 11-month deployment, 10-hour shifts Monday-Friday, 90,000+ parts, 1,250+ runtime hours, 30,000+ X3 vehicles, target zero interventions per shift 🟢. BotQ: 350+ Figure 03, 24x throughput improvement under 120 days, 150+ workstations, 50+ inspection points, >80% EOL FPY, 99.3% battery FPY, 9,000+ actuators 🟢.Figure is the best “deployment + manufacturing KPI” curve, not necessarily the broad winner.Robot count by customer, contract value, ROI/payback, intervention distribution actually achieved, repeat order, revenue, gross margin, support cost.SYNTHESIS_CANDIDATE / existing timeline already added
Q4. Leaderdrive / CATL-NVIDIA analogyLeaderdrive is a strong S4 supplier signal, but CATL/NVIDIA analogy remains unproven.2025 revenue RMB 570.714m +47.31%; net profit RMB 124.367m +121.42%; “工业及具身智能机器人零部件” revenue RMB 422.528m, 74.0% of revenue by calculation, GM 34.88%; harmonic reducer sales 425,158 +72.48% 🟢/🟠.Moves Leaderdrive from concept exposure to filing-backed supplier validation.Named humanoid OEM design win, humanoid-only revenue/backlog, single-robot value content, architecture durability, margin durability, platform/module expansion.SYNTHESIS_CANDIDATE
Q5. China vs US pathChina leads industrial robot deployment scale and domestic supplier flywheel; US leads frontier humanoid/AI/model/mega-round concentration. Neither proves S5 humanoid economics yet.IFR China 2024 installations 295,000, 54% of global demand, 2.027m operational stock, domestic supplier share 57% vs 47% in 2023 🟢; A3 North America 2024 orders 31,311 robots, US$1.963bn, +0.5% units and +0.1% value 🟢; Figure US$675m Series B at US$2.6bn valuation in prior file 🟢.Puts geography after evidence/layer analysis; China-vs-US should be synthesis, not first premise.Comparable 2023-2026 funding dataset by country/category, China humanoid deployment outcomes, US multi-customer paid deployments beyond BMW/Figure, customer economics.SYNTHESIS_CANDIDATE

2. Q0: cycle-stage and timing answer

Working answer

The cycle is no longer pure demo watching. It has entered an S3/S4 “construction evidence” period: public sources now disclose capacity plans, production-line metrics, customer-site operating KPIs, component revenue/shipments, model/data stack releases, and China deployment infrastructure. 🟢/🟠

But it is not yet broad S5 “return evidence.” The missing link is economics: customer payback, repeat orders, low intervention, vendor margin, service burden, and financial materiality. 🟠

Why this matters for timing

  • Signal: measurable S4 evidence justifies a research system and watchlist discipline. 🟠
  • Noise: viral demos, TAM slides, designed capacity, customer logos, or share-price spikes alone should not be treated as return-cycle proof. 🟠
  • Upgrade condition: at least one major humanoid path discloses repeat paid deployments with runtime/intervention, ROI/payback, and vendor economics. 🟠
  • Downgrade condition: capacity build-out rises while deployment KPI, sell-through, or economics remain absent for multiple quarters. 🟠

3. Q1: layer map / value-capture lens

Working answer

Robotics must be analyzed as a stack:

  1. OEM body / full robot platform.
  2. AI / model / data / simulation layer.
  3. Components: reducers, actuators, batteries, sensors, compute.
  4. Deployment/integration/RaaS/services.
  5. Customer productivity capture.
  6. Public-market expression.

The evidence currently lands unevenly: Tesla and Figure are OEM evidence curves; NVIDIA/GR00T is model/data/simulation infrastructure; Leaderdrive is component evidence; BMW/Figure is customer-workflow evidence; Unitree is cost-access/developer-platform evidence. 🟢/🟠

Direct artifact implication

A public field guide should avoid saying “robotics winner.” It should ask: which layer captures value if the sector moves from construction cycle to return cycle? 🟠

4. Q2: Tesla Optimus evidence curve

Working answer

Tesla Optimus is now an S4 capacity/manufacturing-infrastructure curve because the evidence is filing-backed and quantified by site/design capacity. It is not yet S5 because no reviewed public source discloses accepted units, actual production rate, utilization, intervention rate, customer economics, robot revenue, or robot margin. 🟢/🟠

Key sources

  • Tesla Q1 2026 Form 8-K Exhibit 99.1: progress on infrastructure and AI software for Robotaxi and future robotics, and Optimus progress ahead of mass production. 🟢
  • Existing robotics-tesla-optimus-public-evidence-pack-v2.md: Fremont first-generation line designed for 1M robots/year and Texas second-generation line designed for long-term annual capacity of 10M robots. 🟢/🟠

Next question

If Tesla first uses Optimus internally, should the decisive S5 evidence be “robot revenue,” or should it be “factory productivity/payback attributable to Optimus”? 🟠

5. Q3: Figure evidence curve

Working answer

Figure’s public edge is not only a robot demo; it is the combination of named customer-site KPI and manufacturing-process KPI. BMW gives the strongest current public deployment anchor; BotQ gives production/yield/cadence anchors. Still, neither discloses customer economics or vendor financials. 🟢/🟠

Quantified anchors

  • BMW deployment: 11 months, 10-hour shifts Monday-Friday, 90,000+ parts loaded, 1,250+ runtime hours, 30,000+ X3 vehicles, estimated 1.2m+ robot steps / 200+ miles. 🟢 Figure, 2025-11-19.
  • BMW task KPI definitions: 84-second total cycle time, 37-second load time, target >99% placement success per shift, zero interventions per shift goal, 5mm placement tolerance in 2 seconds. 🟢 Figure, 2025-11-19.
  • Derived BMW operating density: 90,000 parts / 1,250 hours = 72 parts per runtime hour. 🟠 Charlie calculation.
  • BotQ: 350+ Figure 03 delivered; production from 1/day to 1/hour = 24x throughput improvement under 120 days; 150+ workstations; 50+ inspection points; >80% EOL FPY; 99.3% battery-line FPY; 500+ battery packs; 9,000+ actuators across 10+ SKUs; 80+ verification tests per robot. 🟢 Figure, 2026-04-29.

Next question

Does Figure’s bottleneck now move from “can it build robots?” to “can it deploy fleets with low intervention and customer ROI?” 🟠

6. Q4: Leaderdrive analogy test

Working answer

Leaderdrive is a real supplier evidence case, but “CATL/NVIDIA of robotics” is still a hypothesis. CATL/NVIDIA analogies require platform power, customer lock-in, durable high-value content, margin durability, and visible design wins. Current public evidence gives revenue/volume/margin, not named humanoid customer or platform lock-in. 🟢/🟠

Quantified anchors

  • 2025 revenue RMB 570.714m, +47.31%; net profit RMB 124.367m, +121.42%. 🟢 2025 annual report via existing Leaderdrive artifacts.
  • 2025 “工业及具身智能机器人零部件” revenue RMB 422.528m, +52.61%, gross margin 34.88%; this equals 74.0% of company revenue by calculation. 🟢/🟠
  • 2025 harmonic reducer production 433,655 units, +72.27%; sales 425,158 units, +72.48%; inventory 18,238 units, -6.34%. 🟢
  • Margin caution: existing artifact notes Leaderdrive GM declined from 41.14% in 2023 to 36.91% in 2025, a -4.23pp change. 🟢/🟠

Next question

If humanoid OEMs vertically integrate joints/actuators, will Leaderdrive capture module/platform value or remain a component supplier with cyclical/margin pressure? 🟠

7. Q5: China vs US path comparison

Working answer

China/US should be analyzed after Q0-Q4 because it is a synthesis of cycle stage, layer, company evidence, and supplier/value-capture evidence. Current evidence says China has a larger industrial deployment and domestic supplier flywheel; the US has concentrated frontier humanoid and AI/model capital. Neither side has public S5 humanoid economics yet. 🟢/🟠

Quantified anchors

  • China 2024 industrial robot installations: 295,000 units, +7%, 54% of global demand; operational stock 2.027m robots; domestic supplier share 57% vs 47% in 2023. 🟢 IFR, 2025-09-25.
  • North America 2024 robot orders: 31,311 robots valued at US$1.963bn; +0.5% units and +0.1% revenue vs 2023. 🟢 A3, 2025-02-17.
  • China vs US installation reference: 295,000 China installations versus 34,200 US installations in prior IFR-derived artifact = 8.63x. 🟢/🟠
  • USCC: no general-purpose fully autonomous humanoid robot is commercially viable today; US and China lead development; China uses subsidies, tax breaks, zones, state-backed centers, local pilots, and public-private collaboration. 🟢 USCC, 2024-10-10.

Next question

Does China’s deployment flywheel convert into humanoid economics faster than US frontier AI/model stacks convert into deployable fleets? The answer requires 2026-2027 deployment outcome tracking, not geography rhetoric. 🟠

8. Signal / noise rules for the mainline

Signal

  • S4 manufacturing/capacity KPI with named site, line, cadence, yield, or output. 🟢/🟠
  • S4 deployment KPI with named customer, task, runtime, throughput, intervention, and safety context. 🟢/🟠
  • Filing-backed supplier revenue, shipment, inventory, and margin evidence. 🟢
  • Model/data stack evidence that changes training, simulation, deployment, or data flywheel. 🟢/🟠
  • Geography evidence tied to installations, operational stock, domestic share, funding, and named deployments. 🟢/🟡

Noise unless upgraded

  • Viral demos without runtime/intervention/task-repeatability. 🔴/🟠
  • Designed capacity treated as actual production. 🔴
  • Customer logo treated as customer ROI. 🔴/🟠
  • Low price treated as reliability or margin. 🔴/🟠
  • Supplier product fit treated as named OEM exposure. 🔴
  • Share-price movement treated as order evidence. 🔴

9. Current public-safe synthesis candidate

Robotics is not yet a broad S5 commercial-economics story, but the research unit has changed. Tesla makes the manufacturing-infrastructure question measurable; Figure makes customer-site and production-process KPI measurable; Leaderdrive makes supplier revenue/volume/margin measurable; IFR/A3/USCC make the China/US deployment-vs-frontier split measurable; NVIDIA/GR00T and related model-stack sources make the AI/data layer measurable.

That is enough for a public field-guide spine, but not enough for a stock-picking conclusion. The public narrative should be: “S4 evidence is now measurable; S5 economics remain the open gate.” 🟠

10. Research map update payload

Suggested updates by section:

Q0

  • Working answer: explicitly add “construction cycle vs return cycle.”
  • Direct artifact: add this file.
  • Missing evidence: add accepted units / repeat paid deployments / customer ROI / vendor margin as the S5 gate.
  • Next question: which S5 disclosure matters first for timing?

Q1

  • Working answer: add six-layer value-capture stack.
  • Direct artifact: add this file.
  • Missing evidence: layer-by-layer revenue/margin and public-market expression.
  • Next question: which layer captures economics if OEM bodies commoditize?

Q2

  • Working answer: keep Tesla at S4 capacity-intent.
  • Direct artifact: add this file as synthesis bridge only, not replacement for Tesla deep pack.
  • Missing evidence: actual output, yield, utilization, deployment KPI, factory ROI.
  • Next question: internal productivity proof vs external revenue proof.

Q3

  • Working answer: Figure = deployment KPI + manufacturing KPI; still pre-S5 economics.
  • Direct artifact: add this file as synthesis bridge.
  • Missing evidence: robot count, contract value, ROI/payback, intervention distribution, repeat order.
  • Next question: is deployment engineering now the bottleneck?

Q4

  • Working answer: Leaderdrive passes S4 supplier validation; CATL/NVIDIA analogy unproven.
  • Direct artifact: add this file.
  • Missing evidence: named humanoid design win, humanoid-only revenue/backlog, architecture durability, margin durability.
  • Next question: component supplier vs module/platform supplier.

Q5

  • Working answer: China = deployment/supplier flywheel; US = frontier humanoid/model/mega-round concentration; neither has S5 humanoid economics.
  • Direct artifact: add this file.
  • Missing evidence: comparable 2023-2026 funding/deployment dataset; China humanoid deployment outcomes; US multi-customer economics.
  • Next question: which path converts first into repeat paid deployment?

11. Source list

Primary / official:

Internal synthesis / prior artifacts:

  • workspace/finance/knowledge/robotics-research-map.md 🟠
  • workspace/finance/knowledge/robotics-s5-evidence-gap-tracker-v1.md 🟠
  • workspace/finance/knowledge/robotics-four-deep-questions-us-china-tesla-figure-leaderdrive-v1.md 🟠
  • workspace/finance/knowledge/robotics-figure-history-product-deployment-timeline-v1.md 🟠
  • workspace/finance/knowledge/robotics-tesla-optimus-public-evidence-pack-v2.md 🟠

12. Risks / exclusions

  • Do not publish as a trade recommendation.
  • Do not imply S4 evidence equals S5 economics.
  • Do not include Hugo private portfolio weights, purchase prices, trade rationale, paid-report excerpts, private channel checks, or unverified supplier rumors.
  • Do not turn China vs US into a winner ranking.
  • Do not use Leaderdrive as a confirmed Tesla/Figure/Unitree supplier without primary confirmation.
  • Do not imply Tesla designed capacity is current output or Figure deployment KPI is customer ROI.