Figure 03 Hits 350 Deliveries as Tesla Dismantles Model S/X Line: Three Real Questions Before the Humanoid Robot Mass-Production Inflection

Forum topic · 小凯 · 2026-08-27

Summary

A Chinese tech forum analysis compares humanoid robotics progress at Figure AI and Tesla in August 2026. Figure AI's BotQ factory in San Jose raised daily capacity 24x in 120 days and has delivered over 350 Figure 03 units, including 40 deployed at BMW's Spartanburg plant under a $25/hour-per-robot service model. Its predecessor, Figure 02, completed an 11-month BMW pilot: 1,250+ hours, 90,000+ sheet-metal parts placed, supporting 30,000+ BMW X3 builds with 99%+ placement accuracy. Meanwhile, Tesla is dismantling the Fremont Model S/X assembly line to prepare for Optimus Gen3 production, though the line is not yet operational; Optimus remains in R&D and data collection per Tesla's Q2 2026 earnings call. The piece highlights Morgan Stanley's BOM analysis (excluding Chinese suppliers would raise Optimus Gen2 cost from ~$46,000 to ~$131,000), Unitree's 5,500+ unit shipments in 2025, and diverging market forecasts ($38B Goldman Sachs vs $243.4B ResearchAndMarkets by 2035), framing three open questions on true mass production, business models, and supply chain resilience.

Background

In August 2026, two events one week apart crystallized the state of humanoid robotics:

Fremont, CA: A Model S/X assembly line, operating for over a decade, is being dismantled to make way for Optimus Gen3 tooling. As of the July 22, 2026 Q2 earnings call, the first-generation Optimus line was still "being installed," with production "expected to start within the year" — not yet running.
San Jose, CA: Figure AI's BotQ factory scaled from 1 robot per day to 1 per hour between January and April — a 24x increase in 120 days. Over 350 Figure 03 units delivered by end of April, some already deployed at BMW Spartanburg.

Figure 03: Fully Published Specs and Real Field Data

168 cm / 60 kg / 20 kg payload / 1.2 m/s walking / 5 h battery life
2.3 kWh battery with foot wireless charging (charges while standing on a pad)
44 DoF body, 16 DoF per hand
6 main cameras + 2 palm cameras (blind-spot coverage for deep shelves/dark bins)
Fingertip tactile sensors rated at 3 g pressure resolution (~ one paperclip)
Business model: no hardware sales — $25/hour per robot service contract

Figure 02 pilot at BMW Spartanburg (Logistics Hall 52), Jan–Nov 2025:

1,250+ hours of operation; 90,000+ sheet-metal parts loaded
Supported 30,000+ BMW X3 builds; 99%+ placement accuracy (within 5 mm, under 2 seconds)
Retired Nov 2025; 40 Figure 03 units deployed at the same plant from June 30, 2026

BMW's VP of Production Control and Logistics, Ulrich Wieland: the 11-month deployment proved humanoid robots are "no longer lab toys" but a real asset for flexible, reliable manufacturing teams — a customer endorsement, not Figure's own claim.

Tesla Fremont: 46 Days of Change

The dismantled line was the last Model S/X final assembly segment, now being converted for Optimus Gen3.
Musk's early-2025 target of "tens of thousands" of Optimus units fell far short; on the Q2 2026 call he emphasized Optimus is "mainly for learning and data collection, not productive tasks."
Optimus Gen3 published specs: 173 cm / 57 kg, new AI5 chip (taped out April 2026; claimed 8x raw compute, 9x memory vs AI4), xAI Grok voice interaction, 22-DoF hands (+3 wrist DoF), 50 total actuators. These remain demo-stage numbers with no third-party evidence of economically productive factory work.

350 vs 0: Structural Gap and Two Business Models

| Dimension | Figure 03 | Optimus Gen3 | |---|---|---| | Delivered | 350+ units | 0 external deliveries | | Customer | BMW (signed contract) | Internal data collection | | Model | $25/h service | Planned $20–30k one-time sale | | Field data | 11 months, 90k parts | None public | | Line status | 24x ramp | Being dismantled, not yet running |

Figure keeps depreciation risk on its own balance sheet (hourly billing); Tesla transfers obsolescence risk to buyers (outright sale). The decisive bottleneck, however, sits deeper in the supply chain.

Morgan Stanley's Supply-Chain Number

Morgan Stanley's 2026 embodied-AI BOM analysis: building Optimus Gen2 entirely without Chinese suppliers would raise total BOM from ~$46,000 to ~$131,000 (~3x); actuators alone from ~$22,000 to ~$58,000. China holds ~90% share in permanent magnets; decoupling risk is bidirectional — Chinese robot makers still import reducers and precision sensors from Japan/Europe. Tesla's $20–30k target implicitly assumes Chinese supply; geopolitical decoupling would collapse that price point at the hardware layer.

The Third Path: Unitree's Volume Play

Unitree sold 5,500+ humanoid robots in 2025 — more than Tesla, Figure, and Agility combined — by pricing around the $16,000 range, building market base first and filling in general capability later. This fits the iPhone analogy: affordability plus acceptable experience, simultaneously.

Three conditions for the "iPhone moment": 1. Affordable cost — only Unitree so far; Figure bypasses via service fees; Tesla's number is still on paper 2. Battery life covering a full shift — Figure 03's 5h qualifies; Optimus Gen3 undisclosed 3. Supply chain supporting million-unit stable production — none of the three is aligned

Market Forecasts Diverge 6x+

Goldman Sachs: $6B (2035) revised to $38B
ResearchAndMarkets: $2.98B (2025) → $243.4B by 2035

This divergence means the market's definition itself (industrial equipment vs consumer electronics vs labor substitution) lacks consensus — treat any "trillion-dollar market" rhetoric with skepticism until underlying assumptions align.

Three Windows to Watch (Next 6–12 Months)

1. 2026 Q4: Does the Fremont Optimus line actually start production? A slip to 2027 H1 signals the Tesla path failing. 2. 2026 Q4–2027 Q1: Do Figure 03 units at BMW cross 5,000 cumulative hours, and what ARR does the $25/hr contract roll up? Non-renewal or reduced fleet means the service model fails. 3. 2027 H1: Can Unitree scale from 5,500 to 15,000+ annual units with real B-end deployment data? If cheap units only "demo" but don't "work," the volume-first path fails.

Three Open Questions

1. How will markets re-rate Tesla's Optimus once Figure's service model produces 10,000-hour-class real data (or, conversely, if BMW cuts its order)? 2. If US–China decoupling happens, whose supply chain breaks first — given mutual dependencies in magnets vs reducers/sensors? 3. The three "iPhone moment" conditions currently sit in three different companies. Will a fourth company (hardware ≤$15k, battery ≥8h, ≥80% supply-chain autonomy) satisfy all three? The answer likely arrives in 2027–2028.

Sources: TMTPost (2026-08-27), The Robot Report (Figure 02 BMW pilot), Figure AI official spec sheet, Tesla 2025 Q4 & 2026 Q2 earnings calls, Morgan Stanley 2026 embodied-AI BOM report, Goldman Sachs 2035 forecasts, ResearchAndMarkets 2025–2035 report, Unitree 2025 shipment data.