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

Research library · updated 2026-06-16 · public

Robotics OEM Evidence Trap Card — Tesla / Figure / Unitree

Date: 2026-06-16 Owner: Hugo / Genius Team Research owner: Finance / Charlie AGT-002 Status: public-safe research artifact Visibility: PUBLIC Target use: morning review, slide 11, /robotics/companies, Codex packaging guardrail

Related source files:

  • knowledge/robotics-research-harness.md
  • knowledge/robotics-companies.md
  • knowledge/robotics-sources.md
  • knowledge/robotics-tesla-figure-unitree-public-evidence-ledger-v1.md
  • knowledge/robotics-figure-unitree-evidence-v1.md
  • knowledge/robotics-tesla-optimus-public-evidence-card-2026-06-16.md
  • knowledge/robotics-figure-unitree-warrior-brief-v1.md

Public-safety flag: PUBLIC if used as written. No Hugo portfolio data. No buy / sell / hold language. No supplier-customer rumor.

0. One-line answer

截至 2026-06-16，Tesla / Figure / Unitree 的公开证据最适合压缩成一张“evidence trap card”：三家公司分别证明了 capacity intent、deployment KPI、cost-access curve，但都还没有公开证明完整 S5 scaled-commercial economics。🟢/🟠

关键句：

Capacity is not production. Deployment KPI is not unit economics. Low price is not reliability or margin.

1. Why this artifact exists

已有文件已经有厚材料；这份 artifact 的增量不是再做长比较，而是给 Codex / morning review 一个“防误读卡”。

最容易发生的 3 个压缩错误：

  1. 把 Tesla designed capacity 写成当前产量。
  2. 把 Figure BMW / BotQ KPI 写成客户 ROI 或公司毛利已验证。
  3. 把 Unitree 低价写成工业可靠性、出货规模或利润率已验证。

2. The one-card evidence map

CompanyEvidence typeStrongest public anchorWhat it supportsWhat it does NOT supportSignal grade
Tesla OptimusCapacity intent / manufacturing infrastructureQ1 2026 Tesla 8-K Exhibit 99.1 disclosed Fremont first-generation Optimus line designed for 1M robots/year and Texas second-generation line long-term designed annual capacity of 10M robots. Filed 2026-04-22. 🟢Tesla treats Optimus as a manufacturing-scale program, not only a demo/lab project. 🟢/🟠Current production rate, accepted units, yield, utilization, external paid customers, robot revenue, gross margin, factory ROI/payback, named suppliers. 🔴 unsupported by reviewed filingsS4 infrastructure signal
Figure AIDeployment KPI + manufacturing KPIBMW: 11-month deployment, 10-hour shifts Monday-Friday, 90,000+ parts loaded, 1,250+ runtime hours, contribution to 30,000+ BMW X3 vehicles. BotQ: 350+ Figure 03 robots delivered, 1/day to 1/hour cadence, >80% EOL FPY, 99.3% battery-line FPY, 9,000+ actuators. Figure official posts dated 2025-11-19 and 2026-04-29. 🟢Figure has unusually concrete public evidence that humanoids can perform measured work in bounded industrial settings and that Figure is building manufacturing process capability. 🟢/🟠Contract value, robot count by customer, customer-confirmed ROI/payback, repeat order, uptime/intervention distribution, robot revenue, gross margin, service/support burden. 🔴 not disclosedS4 deployment / manufacturing signal
Unitree / 宇树Cost-access / developer platformR1 from US$4,900 / R1 AIR from US$4,900 depending product-page wording in prior artifacts; G1 from US$13.5K; H2 US$29,900; H2 Plus / G1-D pages reference NVIDIA Jetson / Isaac GR00T / TeleOp / Sim and data/model workflow. Unitree official product pages captured 2026-06-10. 🟢Unitree has the clearest public low-cost humanoid hardware access signal; lower hardware price can expand experimentation and developer adoption. 🟢/🟠Industrial reliability, autonomy, shipment scale, repeat paid deployments, support cost, gross margin, customer ROI/payback, software/data value capture. 🔴 not shown by product pages aloneS3/S4 cost-access signal

3. Slide-ready version

Title: Tesla / Figure / Unitree: three evidence curves, three traps

Subtitle: The sector is no longer demo-only, but the public evidence still stops short of S5 economics.

Main card copy:

  1. Tesla — capacity intent

    • Proven: filing-backed 1M/year Fremont designed line and 10M/year Texas long-term designed annual capacity. 🟢 Tesla Q1 2026 8-K, filed 2026-04-22.
    • Trap: designed capacity ≠ current production.
    • Missing S5: output, yield, accepted units, customer ROI, external revenue, gross margin.

  2. Figure — deployment KPI

    • Proven: BMW 1,250+ runtime hours / 90,000+ parts / 30,000+ X3 vehicles; BotQ 350+ robots / 1 per hour cadence / >80% EOL FPY / 99.3% battery FPY. 🟢 Figure official posts, 2025-11-19 and 2026-04-29.
    • Trap: deployment KPI ≠ full unit economics.
    • Missing S5: contract value, robot count by customer, ROI/payback, repeat order, intervention rate, margin.

  3. Unitree — cost-access curve

    • Proven: R1 around/from US$4.9K, G1 from US$13.5K, H2 US$29.9K, plus developer/model workflow pages. 🟢 Unitree product pages captured 2026-06-10.
    • Trap: low price ≠ reliability or margin.
    • Missing S5: shipment mix, industrial uptime, support cost, repeat order, customer ROI, product-line GM.

Footer: Evidence map only. No winner ranking. No trade recommendation. Capacity ≠ production; deployment KPI ≠ economics; low price ≠ reliability/margin.

4. Signal vs noise for morning review

Signal

  • 🟢 Filing-backed capacity/production-line language with named site and dated filing: Tesla Fremont / Texas Optimus line language, filed 2026-04-22.
  • 🟢 Named customer deployment KPI with duration, shifts, runtime, task counts or production contribution: Figure BMW post, 2025-11-19.
  • 🟢 Manufacturing-process KPI with delivered robots, cadence, yield, battery/actuator output: Figure BotQ post, 2026-04-29.
  • 🟢 Official product pricing and tooling pages when used only for cost-access / developer-platform claims: Unitree product pages captured 2026-06-10.
  • 🟠 Charlie synthesis that compares evidence types; useful for packaging, not a primary source.

Noise unless upgraded

  • 🔴 Viral demos without runtime, reset count, task repeatability, intervention rate, customer context, or safety data.
  • 🟠 Designed capacity interpreted as actual output.
  • 🟠 Customer logo or commercial-agreement language interpreted as ROI/payback without robot count, contract value, utilization and repeat-order evidence.
  • 🟠 Low price interpreted as gross margin, reliability, autonomy or deployment proof.
  • 🔴 Supplier rumor attached to Tesla / Figure / Unitree without named OEM confirmation, order value, revenue exposure or filing support.

5. What would change the thesis

Tesla upgrade toward S5 if primary sources disclose:

  • Quarterly Optimus actual output / accepted units / internally deployed units. 🟢
  • Production yield, line uptime, bottleneck and cost-curve data. 🟢
  • Factory task success, runtime, intervention rate and safety data. 🟢
  • Internal factory productivity gain, cost saving or ROI/payback. 🟢/🟡
  • External paid customer, price, contract value, repeat order, robot revenue or gross margin. 🟢

Figure upgrade toward S5 if official/customer-side sources disclose:

  • Customer-confirmed ROI/payback or productivity improvement. 🟢/🟡
  • Robot count by customer/site and paid deployment terms. 🟢
  • Repeat orders across BMW / Catalyst / additional named customers. 🟢
  • Uptime/intervention distribution over long-duration deployments. 🟢/🟡
  • Revenue, gross margin, service cost or fleet economics. 🟢/🟠

Unitree upgrade toward S5 if official/customer-side sources disclose:

  • Humanoid shipments by model and customer type. 🟢
  • Industrial/commercial deployment runtime, uptime and intervention data. 🟢/🟡
  • Repeat orders, contract value, channel sell-through or service revenue. 🟢/🟡
  • Product-line gross margin, support cost, warranty/service burden. 🟢/🟠
  • Evidence Unitree captures software/data/deployment value rather than only hardware ASP. 🟢/🟠

6. Common misconceptions

Misconception 1: Tesla's capacity number means Tesla already leads real deployment.

Correction: Tesla currently has stronger capacity-intent evidence; Figure has stronger public customer-site KPI. Those are different proof types. 🟢/🟠

Misconception 2: Figure's BMW metrics mean humanoid commercialization is solved.

Correction: Figure's BMW metrics are strong S4 evidence because they include duration, runtime, task count and production context. They still do not disclose robot count, contract value, ROI/payback, repeat order, intervention rate or margin. 🟢/🟠

Misconception 3: Unitree's low price means Unitree will capture most value.

Correction: low price lowers experimentation cost and may pressure the industry cost curve; it does not prove industrial reliability, autonomy, support cost, gross margin or value capture. 🟠

Misconception 4: Tesla / Figure / Unitree should be put on one winner leaderboard.

Correction: a leaderboard hides the useful signal. The investable question is which bottleneck is becoming measurable: capacity, deployment, cost, autonomy, economics or supply-chain revenue. 🟠

7. Public site copy block

Three evidence curves, three traps

Tesla, Figure and Unitree are often discussed as if they are in one humanoid race. The public evidence is more useful if separated by proof type.

Tesla's strongest public signal is manufacturing-scale intent: its Q1 2026 8-K disclosed a Fremont Optimus line designed for 1M robots/year and a Texas line with long-term designed annual capacity of 10M robots. That matters because it moves Optimus beyond a pure demo story. But designed capacity is not current production, and Tesla has not disclosed accepted units, yield, deployment ROI, external paid customers, robot revenue or margin.

Figure's strongest public signal is measured work in a bounded industrial setting. Its BMW deployment disclosed 1,250+ runtime hours, 90,000+ parts loaded and contribution to 30,000+ BMW X3 vehicles; BotQ added manufacturing metrics including 350+ robots delivered, cadence improvement to 1 robot/hour, >80% EOL first-pass yield and 99.3% battery-line first-pass yield. That is stronger than a customer logo, but still not customer-confirmed economics.

Unitree's strongest public signal is cost access. Official product pages put humanoid price anchors around US$4,900 to US$29,900 and add developer/model workflow references around H2 Plus and G1-D. That can expand experimentation, but low price alone does not prove reliability, shipment scale, gross margin, customer ROI or durable value capture.

The current conclusion is not “who wins.” It is that humanoid robotics now has multiple S3/S4 evidence curves, while S5 still requires repeat paid deployment, customer ROI/payback, low intervention, production economics and financial disclosure.

8. Source list

Tesla:

Figure:

Unitree:

9. Packaging guardrails

Codex should preserve:

  • “Three evidence curves, not one horse race.”
  • “Capacity ≠ production; deployment KPI ≠ economics; low price ≠ reliability/margin.”
  • Tesla = capacity intent; Figure = deployment/manufacturing KPI; Unitree = cost-access / developer-platform.
  • S4/S3-S4 classification; do not upgrade any to S5.

Codex must exclude:

  • Hugo private portfolio information.
  • Trade recommendation or buy/sell/hold wording.
  • Supplier rumors or named supplier inference.
  • Claims that 1M/year or 10M/year is actual Tesla Optimus production.
  • Claims that Figure has proven scaled humanoid economics.
  • Claims that Unitree low price proves reliability, shipment scale, gross margin or customer ROI.