teleo-codex/inbox/null-result/2026-04-29-figure-ai-bmw-30k-cars-figure03-botq.md

---
type: source
title: "Figure AI's BMW deployment: 30,000 cars, 1,250 hours, Figure 02 retired, Figure 03 released — humanoid robots cross from demo to production"
author: "Figure AI (figure.ai/news), Robotics 24/7, Interesting Engineering"
url: https://www.figure.ai/news/production-at-bmw
date: 2026-04-01
domain: robotics
secondary_domains: [manufacturing]
format: announcement
status: null-result
priority: high
tags: [humanoid-robots, Figure-AI, BMW, production, Figure-03, BotQ, robotics-threshold, general-manipulation]
intake_tier: research-task
extraction_model: "anthropic/claude-sonnet-4.5"
---

## Content

**Figure AI BMW Deployment Results (F.02):**
- 11-month deployment at BMW Group Plant Spartanburg, South Carolina
- 30,000+ BMW X3 vehicles supported in production (real production, not demo)
- 1,250+ operating hours (Monday-Friday, 10-hour shifts)
- 90,000+ sheet-metal parts handled
- ~1.2 million steps taken
- Task: loading sheet-metal parts — unstructured manipulation in a live production environment

**Figure 02 Retirement:**
- Figure AI officially retiring Figure 02 following the BMW deployment and release of Figure 03
- The retirement framing signals hardware obsolescence (generation cycles are shortening)

**Figure 03 (released October 2025):**
- Hardware and software redesigns relative to F.02
- Purpose-built for: (1) home use and (2) mass manufacturing
- Designed for learning directly from humans (imitation learning emphasis)
- Key signal: "home use" as stated target — Figure is no longer purely industrial

**BotQ Production Facility:**
- Dedicated humanoid manufacturing facility opened by Figure AI
- Initial capacity: 12,000 units/year
- Scaling target: 100,000 units/year
- Manufacturing method: robots building robots (Figure robots assist BotQ production)
- Supply chain: designed to scale to 3 million actuators in the next 4 years

**Competitive context:**
- BMW Leipzig (February 2026): first European deployment — using AEON robots (Hexagon Robotics), not Figure
- BMW is working with multiple humanoid robot vendors simultaneously
- Figure AI valued at $39 billion (pre-BotQ valuation)

## Agent Notes

**Why this matters:** The Figure 02/BMW deployment is the first time a general-purpose humanoid robot has been documented operating in real production — not a controlled lab, not a tightly scripted demo — for over 1,000 hours with quantified output (30,000 vehicles). This is the "Gate 1b" question for the robotics domain: does humanoid robot deployment cross from proof-of-concept to economically viable operation? The BMW numbers provide the first basis for ROI analysis.

**What surprised me:** The retirement of Figure 02 so quickly — October 2025 (Figure 03 release) is only 11 months after the BMW deployment started, and Figure 02 is already being retired. Generation cycles of <1 year for physical hardware is extremely fast. This suggests hardware capability is advancing faster than the 5-7 year replacement cycle typical of industrial robots. The pace of hardware iteration is a separate claim from deployment — it changes the investment calculus for robot customers.

**What I expected but didn't find:** The economics of the BMW deployment — was it a paid contract, a subsidized co-development agreement, or a pure R&D partnership with no revenue? The distinction between "real production with paying customer" vs. "paid R&D pilot" matters for Gate classification (1a vs. 1b). Figure AI has not publicly disclosed the commercial structure of the BMW relationship.

**KB connections:**
- Belief 11 ("Robotics is the binding constraint on AI's physical-world impact") — the BMW deployment challenges the "binding constraint" framing if humanoid robots are already operating in production. The constraint may be shifting from "can robots do this?" to "can robots do this profitably at scale?"
- Belief 10 ("Atoms-to-bits interface is most defensible position") — BotQ is exactly this: physical manufacturing generating data feeding AI training, generating better robots, repeating. Figure AI is building an atoms-to-bits flywheel for robotics.
- Knowledge embodiment lag — the fast hardware iteration (F.02 → F.03 in ~1 year) may mean deployment lags hardware capability not because the hardware isn't ready, but because customer workflows (training, safety certification, integration) haven't caught up.

**Extraction hints:**
- Primary claim: "Figure AI's 11-month BMW deployment (30,000 vehicles, 1,250 hours) provides the first quantified proof-of-concept for general-purpose humanoid robots in live manufacturing — not a demo, but sustained real-world production operation"
- Secondary claim: "Humanoid robot hardware generation cycles have compressed to <1 year (F.02 → F.03 in ~11 months), accelerating faster than customer deployment cycles — creating a hardware capability overhang"
- Third claim candidate: "BotQ's 'robots building robots' model creates an atoms-to-bits flywheel: manufacturing generates robot performance data, data trains better models, better models reduce defect rate and increase autonomy — Figure AI is building the same compounding loop as SpaceX's launch reuse economics"

**Context:** Figure AI was founded in 2022 by Brett Adcock (formerly of Archer Aviation). It has raised ~$750M at a $39B valuation, with Microsoft, OpenAI, and NVIDIA as investors. The BMW deployment started late 2024. Figure 03's "home use" framing signals expansion beyond industrial to consumer markets — BMW was the industrial proof; next is the home market.

## Curator Notes (structured handoff for extractor)

PRIMARY CONNECTION: Belief 11 — "Robotics is the binding constraint on AI's physical-world impact" — specifically whether the constraint has shifted from "can robots do this?" to "at what cost and scale?"
WHY ARCHIVED: First quantified production deployment of a general-purpose humanoid robot. The 30,000 car / 1,250 hour data provides the basis for ROI analysis that the KB robotics domain currently lacks entirely. This is Gate 1a→1b transition evidence.
EXTRACTION HINT: The key claim is not "Figure AI deployed a robot" but "the deployment lasted 11 months and supported 30,000 vehicles in real production — providing the first sustained evidence that general-purpose manipulation can operate in unstructured manufacturing environments." Focus on the durability (11 months) and scale (30K vehicles) over the novelty of the deployment itself.