Figure AI vs Unitree — deployment proof vs cost-curve proof brief v1

Date: 2026-06-15 Owner: Hugo / Genius Team Agent: Finance / Charlie AGT-002 Status: SYNTHESIS_CANDIDATE Visibility: PUBLIC Output target: site / slide brief Related source packets:

knowledge/robotics-figure-unitree-evidence-v1.md
knowledge/robotics-tesla-figure-unitree-comparison-v1.md
knowledge/robotics-why-now-hardware-cost-catalyst-v1.md
knowledge/robotics-customer-side-deployment-funnel-v1.md

0. One-line answer

截至 2026-06-15，Figure AI 与 Unitree / 宇树最适合被包装成一组“不同证据类型”的比较：Figure 的最强公开证据是 BMW 产线部署 KPI 与 BotQ 制造爬坡，Unitree 的最强公开证据是 US$4,900–29,900 的 humanoid 价格锚、NVIDIA / Isaac 工作流对接、以及数据/训练工具链。Figure 更像 S4 deployment / manufacturing proof；Unitree 更像 S3/S4 cost-access / platform proof；二者都还不是 S5 scaled-commercial-economics proof。🟢 Figure official posts and Unitree official product pages captured in prior packets, accessed 2026-06-10; 🟠 Charlie classification as of 2026-06-15.

1. Why this brief matters

Slide / site 上如果把 Figure 与 Unitree 写成“谁领先”，容易误导。更高质量的框架是：

Figure 在回答：humanoid 是否已经能在真实工业场景中持续完成可计量任务？
Unitree 在回答：humanoid 硬件是否已经便宜到足以扩大实验和开发者生态？
投资研究真正要追踪的是：deployment proof、manufacturing proof、cost proof、AI/data proof、commercial-economics proof 五条证据链是否收敛。

Public-safe headline:

Figure shows the strongest public deployment metrics; Unitree shows the strongest public cost-access signal. Neither has yet shown the full economics chain.

2. Evidence table for Codex / Warrior

Dimension	Figure AI	Unitree / 宇树	Current read	Source grade	Signal
Named customer / site	BMW Group Plant Spartanburg; Catalyst Brands Reno distribution/logistics start site.	Reviewed official product pages emphasize products/prices/tooling rather than named customer KPI.	Figure has stronger customer-site evidence.	🟢 company official posts/pages	Figure S4; Unitree S3
Runtime / task KPI	BMW: 11-month Figure 02 deployment, 10-hour shifts Monday-Friday, 90,000+ parts loaded, 1,250+ runtime hours, contribution to 30,000+ BMW X3 vehicles.	No comparable customer-site runtime / task-output KPI in reviewed official product pages.	Figure provides the better deployment proof-quality anchor.	🟢 Figure official BMW post dated 2025-11-19	S4, not S5
Manufacturing evidence	BotQ: 350+ Figure 03 robots delivered; production cadence from 1/day to 1/hour; >80% EOL first-pass yield; 99.3% battery-line FPY; 500+ battery packs; 9,000+ actuators across 10+ SKUs.	Product breadth is visible across R1/G1/H2/H2 Plus/G1-D, but reviewed pages do not disclose factory cadence/yield.	Figure has stronger disclosed manufacturing-process evidence.	🟢 Figure official BotQ post dated 2026-04-29; 🟢 Unitree pages	Figure S4; Unitree S3
Price / cost access	No public customer pricing or contract economics in reviewed materials.	R1 AIR from US$4,900 / R1 from US$5,900; G1 from US$13.5K; H2 US$29,900, tax/shipping excluded.	Unitree has the strongest humanoid hardware price anchor.	🟢 Unitree official product pages accessed 2026-06-10	S3/S4 cost-access signal
AI / platform surface	Helix 02 official claim: full-body autonomy; highlighted 4-minute continuous dishwasher/kitchen task with no resets / no human intervention.	H2 Plus references NVIDIA Jetson T5000, FP4 2,070 TFLOPS, Isaac GR00T, Isaac TeleOp, Isaac Sim; G1-D references data acquisition, processing, labeling, model training and inference tools.	Figure is task-demo/deployment-adjacent; Unitree is developer-platform/tooling-adjacent.	🟢 company official posts/pages; 🟠 commercialization implication	S3/S4
Commercial economics	Missing: customer-confirmed ROI, contract value, robot count at BMW/Catalyst, repeat-order terms, long-run uptime/intervention distribution, gross margin.	Missing: humanoid shipment mix, industrial uptime, repeat orders, gross margin, warranty/support burden, software/service revenue.	Neither has S5 proof.	🟢 absence in reviewed primary materials; 🟠 classification	Not S5

3. Slide-ready framing

Card 1 — Figure: strongest public deployment proof

As of reviewed sources through 2026-06-15, Figure’s BMW data is among the strongest public humanoid deployment signals: 11 months, 10-hour weekday shifts, 90,000+ parts, 1,250+ runtime hours, and 30,000+ X3 vehicles touched. That makes it stronger than a demo reel, but not yet a commercialization proof because contract value, robot count, uptime/intervention distribution, payback, repeat order, and margin are not public. 🟢/🟠

Card 2 — Unitree: strongest public cost-access proof

Unitree’s evidence is different: R1 AIR from US$4,900, G1 from US$13.5K, and H2 at US$29,900 create public price anchors for humanoid experimentation. H2 Plus and G1-D also show an attempt to package compute, teleoperation, simulation, data acquisition, model training, and inference around the robot body. This can widen the experiment base, but it does not prove industrial reliability or economics. 🟢/🟠

Card 3 — What neither has proven

Neither Figure nor Unitree has yet closed the S5 chain: accepted units by customer/site, multi-period uptime, intervention rate, repeat-order economics, customer ROI/payback, robot revenue, gross margin, service/support burden, and warranty quality. Until those appear, the right label is “stronger S4 evidence,” not “scaled commercial proof.” 🟠

4. Signal vs noise

Signal

Figure BMW deployment KPI: S4 material signal because it combines named customer, real production site, duration, runtime, and task output. 🟢 as of 2025-11-19 post.
Figure BotQ manufacturing KPI: S4 material signal because it discloses delivered robot count, production cadence, yields, batteries, actuators, and functional verification tests. 🟢 as of 2026-04-29 post.
Figure Catalyst agreement: S3/S4 because it is described as a commercial agreement with a named starting site, but lacks robot count/value/economics. 🟢 as of 2026-05-26 post.
Unitree low-price humanoids: S3/S4 because public price anchors can change experiment access and cost-curve expectations. 🟢 accessed 2026-06-10.
Unitree H2 Plus / G1-D tooling: S3 because compute/data/model toolchain evidence matters, but adoption and economics remain undisclosed. 🟢 accessed 2026-06-10.

Noise unless upgraded

Viral humanoid videos without duration, repetitions, intervention data, site, or economics. 🔴/🟠
Customer logos without robot count, contract value, task KPI, or repeat deployment. 🟢/🟠
Price/spec pages treated as deployment proof. 🟢/🟠
Production cadence treated as profitable utilized fleet. 🟢/🟠
TAM charts treated as investable timing. 🟠

5. What would change the classification

Figure upgrade path to S5

Upgrade if BMW, Catalyst, or another named customer confirms robot count, paid economics, repeat order, task ROI/payback, uptime/intervention distribution, and deployment replication across sites. 🟢

Downgrade if BMW remains a showcase, Catalyst metrics stay undisclosed, or BotQ output rises without field utilization / customer deployments. 🟠

Unitree upgrade path to S5

Upgrade if Unitree discloses humanoid shipments by product line, named industrial deployments, uptime/intervention, customer repeat orders, humanoid gross margin, warranty/support cost, or software/service revenue. 🟢

Downgrade if low-cost humanoids remain mainly demo/research hardware, or if value migrates to NVIDIA/open models/integrators while Unitree captures commoditized hardware margin. 🟠

6. Common misconception corrections

Misconception: “Figure has BMW metrics, so humanoids are solved.” Correction: BMW is a strong S4 proof point, but S5 needs customer economics and repeatability. 🟢/🟠
Misconception: “Unitree is cheap, so it wins.” Correction: low price can expand adoption or compress hardware value capture; it does not prove autonomy, support burden, or margin. 🟠
Misconception: “Manufacturing output equals commercialization.” Correction: 350+ robots and 1/hour cadence matter, but utilized profitable fleet evidence is separate. 🟢/🟠
Misconception: “NVIDIA / Isaac alignment means Unitree owns the AI layer.” Correction: alignment may strengthen the platform surface while shifting value to compute, models, data, or deployment services. 🟢/🟠

7. Think Deeper questions

If Figure gets repeat orders but keeps economics undisclosed, what minimum operating KPI should qualify as S5-adjacent?
If Unitree pushes humanoids below US$5K–30K, does the profit pool migrate away from OEM hardware toward AI/data/deployment layers?
Which is more thesis-changing for the sector: Figure customer-verified repeat deployment or Unitree verified humanoid shipment scale?
If low-cost humanoids accelerate embodied-AI data collection, who captures the data advantage: OEM, model provider, integrator, customer, or compute platform?

8. Public-safe site draft section

Figure vs Unitree: deployment proof vs cost-curve proof

Figure and Unitree are useful because they show two different ways humanoid robotics can become investable evidence.

Figure’s public evidence is stronger on deployment. Its BMW deployment ran for 11 months, included 10-hour weekday shifts, loaded 90,000+ parts, accumulated 1,250+ runtime hours, and contributed to 30,000+ BMW X3 vehicles. Its BotQ post also disclosed 350+ Figure 03 robots delivered and a production cadence improvement from 1 robot/day to 1 robot/hour.

Unitree’s public evidence is stronger on cost access. Its official pages list R1 AIR from US$4,900, G1 from US$13.5K, and H2 at US$29,900. H2 Plus and G1-D also connect the robot body to compute, teleoperation, simulation, data acquisition, model training, and inference workflows.

The lesson is not “Figure wins” or “Unitree wins.” The lesson is that humanoid robotics now has multiple measurable evidence paths. Figure is closer to deployment proof; Unitree is closer to cost-curve proof. Neither has yet disclosed the full S5 economics chain: repeat orders, uptime, intervention rate, customer payback, revenue, gross margin, support cost, and warranty quality.

9. Source list

Figure AI:

Figure, “F.02 Contributed to the Production of 30,000 Cars at BMW”, 2025-11-19, accessed 2026-06-10: https://www.figure.ai/news/production-at-bmw 🟢
Figure, “Figure Signs Agreement with Catalyst Brands to Scale Humanoid Operations”, 2026-05-26, accessed 2026-06-10: https://www.figure.ai/news/figure-signs-agreement-with-catalyst-brands 🟢
Figure, “Ramping Figure 03 Production”, 2026-04-29, accessed 2026-06-10: https://www.figure.ai/news/ramping-figure-03-production 🟢
Figure, “Introducing Helix 02: Full-Body Autonomy”, 2026-01-27, accessed 2026-06-10: https://www.figure.ai/news/helix-02 🟢

Unitree / 宇树:

Unitree R1 product page, accessed 2026-06-10: https://www.unitree.com/R1 🟢
Unitree G1 product page, accessed 2026-06-10: https://www.unitree.com/g1/ 🟢
Unitree H2 product page, accessed 2026-06-10: https://www.unitree.com/H2 🟢
Unitree H2 Plus product page, accessed 2026-06-10: https://www.unitree.com/H2plus 🟢
Unitree G1-D product page, accessed 2026-06-10: https://www.unitree.com/G1-D 🟢

Internal synthesis references:

knowledge/robotics-figure-unitree-evidence-v1.md 🟠
knowledge/robotics-tesla-figure-unitree-comparison-v1.md 🟠
knowledge/robotics-customer-side-deployment-funnel-v1.md 🟠
knowledge/robotics-why-now-hardware-cost-catalyst-v1.md 🟠

10. Public safety

Public-safe: yes, if used as an evidence map.

Do not include Hugo private portfolio data, private channel checks, trade rationale, buy/sell/hold language, paid-report excerpts, or unverified supplier/customer rumors. Do not imply Figure has proven scaled humanoid economics. Do not imply Unitree low price proves reliability, autonomy, gross margin, customer ROI, or shipment scale.