Tesla / Figure / Unitree public evidence ledger v1

Date: 2026-06-13 Owner: Hugo / Genius Team Research owner: Finance / Charlie AGT-002 Status: public-safe research artifact Visibility: PUBLIC Target use: slide 11, /robotics/companies, /robotics/signals

Related files:

• knowledge/robotics-research-harness.md
• knowledge/robotics-tesla-figure-unitree-comparison-v1.md
• knowledge/robotics-tesla-optimus-filing-evidence-audit-v1.md
• knowledge/robotics-figure-ai-deep-dive-v1.md
• knowledge/robotics-unitree-deep-dive-v1.md
• knowledge/robotics-figure-unitree-morning-review-evidence-audit-v1.md

Public-safety: yes. No private portfolio data. No buy / sell / hold language. No external publishing action.

0. One-line answer

截至 2026-06-13，Tesla / Figure / Unitree 最适合被公开讲成“三条证据曲线”，不是一个 winner ranking：

• Tesla Optimus: S4 manufacturing-infrastructure / capacity-intent evidence。SEC filings 支持 Fremont pilot line、production-line installation、Gen 3 mass-production design、Fremont 1M/year designed line、Texas 10M/year long-term designed line；但不支持 actual output、yield、deployment KPI、external paid customer、robot revenue 或 margin。🟢/🟠 as of Tesla Form 8-K Exhibit 99.1 sequence filed 2025-04-22 to 2026-04-22.
• Figure AI: S4 deployment KPI + manufacturing KPI evidence。官方披露 BMW 11-month deployment、10-hour shifts Monday-Friday、90,000+ parts loaded、1,250+ runtime hours、30,000+ BMW X3 vehicles contribution；BotQ 披露 350+ Figure 03 delivered、1/day to 1/hour cadence、>80% EOL first-pass yield、99.3% battery-line first-pass yield、9,000+ actuators。仍缺 customer-confirmed ROI/payback、contract value、robot count by customer、intervention rate、gross margin。🟢/🟠 as of Figure official posts dated 2025-11-19, 2026-04-29, 2026-05-26.
• Unitree / 宇树: S3/S4 hardware cost/access + developer-platform evidence。官方页面给出 R1 AIR from US$4,900 / R1 from US$5,900、G1 from US$13.5K、H2 US$29,900、H2 Plus NVIDIA Jetson / Isaac GR00T / TeleOp / Sim、G1-D data acquisition and model workflow；但不支持 industrial uptime、shipment scale、gross margin、repeat orders、customer ROI。🟢/🟠 as of Unitree official product pages captured 2026-06-10.

1. Why this artifact exists

已有文件分别覆盖 Tesla、Figure、Unitree。本文件的增量是给 Codex 一个“claim ledger”：每家公司有哪些 public-safe claim、对应证据等级、绝对不能误读成什么。

核心压缩规则：

Capacity ≠ production; deployment KPI ≠ full economics; low price ≠ reliability or margin.

2. Evidence ledger

3. What each curve actually answers

Tesla answers: can humanoids enter mega-scale manufacturing planning?

Signal:

• Quantified designed-capacity anchors make Tesla stronger than demo-only newsflow. 🟢
• Named manufacturing sites make the evidence trackable: Fremont and Texas. 🟢
• Tesla's own caveat says installed capacity is not current production rate and production depends on equipment uptime, component supply, factory-upgrade downtime, regulatory considerations and other factors. 🟢/🟠

Not answered:

• Are Optimus units accepted, reliable, safe, low-intervention, and economically useful in factories? 🔴 unsupported as of reviewed filings.
• Are external customers willing to pay? 🔴 unsupported as of reviewed filings.
• Do suppliers capture durable value? 🔴 unsupported without named customer / order / revenue evidence.

Figure answers: can a humanoid produce measurable work in a bounded industrial setting?

Signal:

• BMW deployment metrics are more concrete than customer-logo evidence because they include shifts, parts, runtime hours and vehicle-production contribution. 🟢
• BotQ manufacturing metrics are more concrete than “factory planned” language because they include delivered robots, cadence and yield references. 🟢
• Catalyst adds another named commercial deployment surface, though economics are not disclosed. 🟢/🟠

Not answered:

• Was BMW deployment profitable for the customer or Figure? 🔴 not disclosed.
• How many robots were deployed by customer and how much human intervention was required? 🔴 not disclosed.
• Are the tasks repeatable across many customers, or bespoke to one cell / site / workflow? 🟠 open question.

Unitree answers: can humanoid hardware become cheap and accessible enough to expand the experiment base?

Signal:

• Sub-US$5K to US$30K official price anchors are meaningful because they change developer/research/pilot access cost. 🟢/🟠
• Product breadth and G1-D / H2 Plus tooling language make Unitree relevant to the developer-platform and data-collection layer. 🟢/🟠

Not answered:

• Are the robots durable enough for paid industrial work? 🔴 not shown by product pages alone.
• What is Unitree's humanoid shipment scale, customer mix, gross margin, service cost, or repeat-order behavior? 🔴 not disclosed in reviewed official pages.
• Does low-cost hardware capture the profit pool, or does value migrate to software/data/deployment services? 🟠 open framework question.

4. Slide-ready compression

Suggested title: Tesla / Figure / Unitree: three evidence curves, not one horse race

Three cards:

1. Tesla — capacity ambition

   • Filing-backed line installation and 1M/year Fremont / 10M/year Texas designed-capacity anchors. 🟢
   • Missing: actual output, yield, runtime, intervention, ROI, external revenue. 🟢/🟠

2. Figure — deployment + manufacturing KPI

   • BMW runtime/parts/shifts and BotQ cadence/yield/robot-delivery metrics. 🟢
   • Missing: customer-confirmed economics, robot count by customer, intervention rate, gross margin. 🟢/🟠

3. Unitree — cost/access platform

   • R1 from US$4,900; G1 from US$13.5K; H2 US$29,900; G1-D / H2 Plus developer-model workflow. 🟢
   • Missing: deployment reliability, shipments, repeat orders, margin, ROI. 🟢/🟠

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

5. Signal vs noise rules

Signal:

• 🟢 Filing-backed production-line / capacity language with dated site and capacity anchor.
• 🟢 Official customer-site deployment KPI with hours, parts, shifts, task contribution or runtime.
• 🟢 Official manufacturing KPI with robot count, cycle time, yield, actuator/battery output.
• 🟢 Official price / product / developer-tooling pages when used only for cost/access claims.
• 🟠 Synthesis that compares evidence types, as long as each claim points back to a primary anchor.

Noise unless upgraded:

• 🔴 Viral demo videos without task KPI, runtime, intervention or safety data.
• 🔴 Supplier rumors tied to Tesla / Figure / Unitree without named customer, order/revenue, teardown or filing-backed exposure.
• 🟠 Customer logo without economics, volume, runtime or repeat-order detail.
• 🟠 Low price interpreted as high margin or reliable industrial deployment.
• 🟠 Designed capacity interpreted as achieved production.

6. S5 upgrade map

Tesla upgrades toward S5 if primary evidence discloses at least 3 of these:

• Quarterly Optimus output / accepted units / internal deployed units. 🟢
• Production yield, tool uptime, bottleneck and cost curve. 🟢
• Runtime hours, uptime, task success, intervention rate, safety data. 🟢
• Factory productivity improvement, cost saving, ROI/payback. 🟢/🟡
• External paid customer, robot count, contract value, pricing, repeat order. 🟢
• Robot revenue, gross margin, warranty/service cost or segment disclosure. 🟢

Figure upgrades toward S5 if primary/customer-side evidence discloses:

• Customer-confirmed ROI/payback or productivity gain. 🟢/🟡
• Robot count by customer/site and paid deployment terms. 🟢
• Repeat orders across multiple customers/workflows. 🟢
• Uptime/intervention/safety metrics over long duration. 🟢/🟡
• Revenue, gross margin, service cost or fleet economics. 🟢/🟠

Unitree upgrades toward S5 if official/customer evidence discloses:

• Humanoid unit shipments by product line and customer mix. 🟢
• Industrial or commercial deployment runtime / uptime / intervention data. 🟢/🟡
• Repeat orders, contract value, channel sell-through or service revenue. 🟢/🟡
• Product-line gross margin, support cost and warranty burden. 🟢/🟠
• Evidence that Unitree captures software/data/deployment value, not only hardware ASP. 🟢/🟠

7. Common misconceptions

Misconception 1: “Tesla has the biggest capacity target, so Tesla leads deployment.”

Correction: Tesla leads in disclosed capacity ambition; Figure currently has stronger public customer-site KPI. Capacity ambition and deployment KPI answer different questions. 🟢/🟠

Misconception 2: “Figure's BMW data means humanoid commercialization is solved.”

Correction: Figure's BMW metrics are meaningful S4 evidence, but S5 still needs customer-confirmed economics, robot count, repeat orders, intervention rate and margin. 🟢/🟠

Misconception 3: “Unitree's price proves it will win.”

Correction: low price lowers experimentation cost; it does not prove reliability, autonomy, margin or customer ROI. It may also shift value away from hardware toward software/data/deployment. 🟠

Misconception 4: “The three companies should be scored on one leaderboard.”

Correction: a single winner score hides the real question. The sector is currently a set of bottleneck tests: capacity, deployment, cost, autonomy, economics. 🟠

8. Think Deeper questions

1. Which evidence changes the sector thesis first: Tesla output/yield, Figure repeat-customer ROI, or Unitree shipment scale?
2. If low-cost humanoid bodies become widely available before autonomy is solved, where does value migrate: OEMs, model/data platforms, integrators, services, or customers?
3. If Tesla keeps Optimus internal for a long time, should investors treat factory productivity/payback as equivalent to robot revenue?
4. If Figure has the clearest deployment KPI but remains private, which public-market proxies actually capture that signal without overfitting supplier rumors?
5. If Unitree compresses hardware ASP, does that expand the addressable market faster than it compresses hardware margins?

9. Public site draft section

Three evidence curves, not one horse race

Tesla, Figure and Unitree are often discussed as if they are in one humanoid race. A better way to read the evidence is to separate the bottlenecks they are testing.

Tesla is testing whether a vertically integrated AI/manufacturing company can move humanoids into mega-scale production planning. Its public evidence is strongest on manufacturing infrastructure and designed-capacity intent: Fremont 1M/year and Texas 10M/year long-term capacity language. That is meaningful, but still not proof of current production or customer economics.

Figure is testing whether humanoids can do measurable work in bounded industrial environments. Its BMW and BotQ disclosures provide the strongest public deployment/manufacturing KPIs in this three-company set, including runtime hours, parts loaded, shifts, production cadence and yield references. The missing piece is customer-confirmed economics and repeatability across customers.

Unitree is testing whether humanoid hardware can become cheap enough to expand the experiment base. Its public price anchors are unusually explicit, from R1 AIR at US$4,900 to H2 at US$29,900, with developer/model workflow pages around G1-D and H2 Plus. The missing piece is commercial deployment reliability, shipment scale, margin and customer ROI.

The current conclusion is not “who wins.” It is that robotics evidence has improved from demo-heavy narrative into S3/S4 KPI tracking, but the sector still lacks complete S5 scaled-commercial-economics proof.

10. Source list

Tesla:

• Tesla Q1 2025 Update, Form 8-K Exhibit 99.1, filed 2025-04-22: https://www.sec.gov/Archives/edgar/data/1318605/000162828025018851/exhbit991.htm 🟢
• Tesla Q3 2025 Update, Form 8-K Exhibit 99.1, filed 2025-10-22: https://www.sec.gov/Archives/edgar/data/1318605/000162828025045861/exhibit991.htm 🟢
• Tesla Q4 2025 Update, Form 8-K Exhibit 99.1, filed 2026-01-28: https://www.sec.gov/Archives/edgar/data/1318605/000162828026003837/exhibit991.htm 🟢
• Tesla Q1 2026 Update, Form 8-K Exhibit 99.1, filed 2026-04-22: https://www.sec.gov/Archives/edgar/data/1318605/000162828026026551/exhibit991.htm 🟢
• Internal verification base: robotics-tesla-optimus-filing-evidence-audit-v1.md, reviewed 2026-06-13. 🟠 synthesis / audit note

Figure:

• Figure, “F.02 Contributed to the Production of 30,000 Cars at BMW”, 2025-11-19: https://www.figure.ai/news/production-at-bmw 🟢
• Figure, “Ramping Figure 03 Production”, 2026-04-29: https://www.figure.ai/news/ramping-figure-03-production 🟢
• Figure, “Figure Signs Agreement with Catalyst Brands to Scale Humanoid Operations”, 2026-05-26: https://www.figure.ai/news/figure-signs-agreement-with-catalyst-brands 🟢
• Figure, “Introducing Helix 02: Full-Body Autonomy”, 2026-01-27: https://www.figure.ai/news/helix-02 🟢
• Internal verification base: robotics-figure-ai-deep-dive-v1.md, reviewed 2026-06-13. 🟠 synthesis / audit note

Unitree:

• Unitree R1 product page, captured in internal source-backed artifact 2026-06-10: https://www.unitree.com/R1 🟢
• Unitree G1 product page, captured in internal source-backed artifact 2026-06-10: https://www.unitree.com/g1/ 🟢
• Unitree H2 product page, captured in internal source-backed artifact 2026-06-10: https://www.unitree.com/H2 🟢
• Unitree H2 Plus product page, captured in internal source-backed artifact 2026-06-10: https://www.unitree.com/H2plus 🟢
• Unitree G1-D product page, captured in internal source-backed artifact 2026-06-10: https://www.unitree.com/G1-D 🟢
• Internal verification base: robotics-unitree-deep-dive-v1.md, reviewed 2026-06-13. 🟠 synthesis / audit note

11. Public-safe flag

Public-safe: yes.

Exclude:

• Hugo private portfolio data.
• TSLA / private company / supplier buy-sell-hold language.
• Any claim that Tesla designed capacity equals current production.
• Any claim that Figure deployment KPI proves full economics.
• Any claim that Unitree low price proves reliability, gross margin, shipment scale or customer ROI.
• Any unsupported supplier-customer inference.