Physical AI enters a scale-up phase for humanoid robot commercialization

Overview

This report focuses on the global commercialization of Physical AI and humanoid robots. Nomura believes that Boston Dynamics' investor demonstration and Figure AI's autonomous multi-day run in mid-May 2026 mark a key industry transition from R&D validation to scaled operational assets. The core conclusion is that humanoid robots are approaching deployment in logistics and industrial settings, and competition will center on mass-production costs, supply-chain simplification, scenario data accumulation, and proprietary model flywheels.

Core views

The report's core views are as follows: first, Boston Dynamics' all-electric Atlas lowers manufacturing and application risk through standardized components and heavy-load capability; second, Figure AI has demonstrated a viable commercialization path through logistics endurance, BotQ factory capacity, and BMW site deployment; third, compared with automotive lines, logistics fulfillment centers have lower precision and liability risk and larger demand scale, making them likely the most important near-term commercialization breakthrough; fourth, for labor-intensive fulfillment platforms such as Coupang, shifting from labor to RaaS may compress variable operating costs and improve margin resilience.

Analysis framework

The report uses an event-driven commercialization framework for the industry chain, assessing Boston Dynamics and Figure AI's demonstrations, mass-production capacity, on-site deployment data, payload capacity, logistics TAM, and RaaS economics, and mapping these changes to automotive manufacturing, industrial automation, and internet retail fulfillment scenarios.

Methodology notes

• Industry commercializationFrom technical feasibility to commercial scalability

  From an R&D cost center to a scalable operational asset

  The report argues that the key question for humanoid robot investment is no longer simply whether an action can be completed, but whether it can be brought into real logistics and industrial scenarios in a replicable, low-cost, and sustainable way.

• Unit economicsBOM standardization and mass-production cost reduction

  Standardized actuators and high-pressure die casting reduce marginal costs

  Boston Dynamics reduced supply-chain and capex complexity by consolidating a complex hydraulic legacy into two standardized commercial actuators; Figure AI shifted from low-volume machining to automotive-grade high-volume die casting to lower the marginal cost per robot.

• Data flywheelProprietary operational data flywheel

  The more deployments, the more real-world data, and the stronger the model and operational capability

  The report describes industry competition as a winner-take-all contest driven by mass-production unit economics and proprietary data flywheels.

• Rating methodologyNomura equity research rating system

  Buy, Neutral, and Reduce relative-to-benchmark ratings

  Nomura's Buy means the analyst expects the stock to outperform the benchmark over the next 12 months; if a target price is discussed, it represents the 12-month stock price forecast.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• COUPANG INC (CPNG.US)

  Potential beneficiary; mentioned in the report as CPNG US, Buy

  Strengths

  The logistics fulfillment scenario is large in scale, and if RaaS humanoid robots are adopted, variable labor-intensive operating costs may be compressed and margin resilience improved.

  Weaknesses

  The report does not provide Coupang-specific deployment progress, investment size, or financial estimates.

  Comparison

  Compared with automotive lines, logistics scenarios have lower precision requirements and less safety and liability friction, but a larger task volume.

  Risks

  The pace of robot commercialization, unit cost declines, on-site reliability, integration difficulty, and resistance to labor substitution could affect the realization of benefits.

• Boston Dynamics / Hyundai

  Unlisted core industry participant; validator of heavy-duty industrial tasks

  Strengths

  Atlas demonstrated a 23kg carrying capability, and the training model can adapt to up to 45kg; standardized actuators help reduce BOM and supply-chain complexity.

  Weaknesses

  The report does not disclose order scale, pricing, gross margin, or specific commercial contracts.

  Comparison

  Compared with light-load competitors, Boston Dynamics is better suited to heavy industrial tasks such as automotive parts sorting and steel-related handling.

  Risks

  High-load scenarios require higher standards for safety, reliability, cost, and on-site integration.

• Figure AI

  Unlisted core industry participant; validator of logistics and manufacturing site deployment

  Strengths

  It demonstrated six consecutive days of autonomous endurance, increased BotQ capacity to one unit per hour, and autonomously moved more than 90,000 critical components at BMW's Spartanburg plant over 11 months with zero teleoperation.

  Weaknesses

  It still needs to prove quality consistency, service costs, and cross-customer replication after large-scale mass production.

  Comparison

  Compared with automotive lines, logistics fulfillment centers may reach scaled TAM more quickly; compared with Boston Dynamics, Figure AI emphasizes operational endurance and capacity ramp-up.

  Risks

  Mass-production ramp-up, customer conversion, RaaS pricing, on-site downtime costs, and intensifying competition are key uncertainties.

Key data

• Boston Dynamics demonstration date2026-05-18The investor presentation explained the commercialization strategy for the all-electric Atlas.
• Figure AI autonomous endurance demonstration6 consecutive days, starting 2026-05-14Used to demonstrate the operational endurance needed to replace human shifts.
• Atlas demonstration payload23kg (50lbs)Atlas used whole-body control to carry a mini fridge.
• Atlas training payload adaptabilityUp to 45kg (100lbs)The reinforcement learning model adapted to heavier weights during training, showing software robustness to weight changes.
• Figure AI capacity increaseFrom 1 unit per day to 1 unit per hourCorresponds to annualized capacity of more than 12,000 units.
• BMW site deployment11 months, 90,000+ critical components, zero teleoperationEvidence of de-risking in Figure AI's real factory deployment of humanoid robots.
• Nomura rating distributionBuy 57%; Neutral 41%; Reduce 2%Nomura Group's global equity research rating distribution disclosure as of 2026-03-31.

Impact & implications

If the report's view proves correct, the value focus of the humanoid robot industry chain will shift from lab demonstrations to large-scale manufacturing, scenario deployment, and closed-loop operational data. Logistics fulfillment may benefit first because its precision requirements are lower than automotive assembly lines but its task volume is much larger, making it suitable for RaaS to replace part of labor costs. For internet retail and fulfillment platforms such as Coupang, robot deployment may improve long-term variable cost structure; for the automotive and industrial automation sectors, payload capacity and deployment data will determine whether suppliers can enter high-value tasks.

Risks

• Humanoid robots may still face reliability, safety, maintenance, and on-site integration challenges when moving from demonstrations to large-scale commercial deployment.
• If unit cost declines are slower than expected, the economics of the RaaS model may be insufficient for customers.
• High-precision scenarios such as automotive lines involve greater safety and liability friction, so commercialization may progress more slowly than in logistics scenarios.
• Winner-take-all competition may quickly marginalize participants that lag in technology, data, and manufacturing capabilities.
• The report does not provide Coupang-specific financial sensitivity, target price, or earnings forecasts, so the investment mapping remains thematic and scenario-based.

What to watch

• Whether Boston Dynamics' all-electric Atlas secures verifiable commercial orders and batch deliveries.
• Whether Figure AI's BotQ factory can stably maintain one unit per hour and annualized capacity above 12,000 units.
• Figure AI's replication progress beyond BMW and into logistics fulfillment scenarios.
• Whether Coupang and other fulfillment platforms publicly disclose RaaS or humanoid robot pilots, cost savings, and efficiency metrics.
• The downtrend in humanoid robot BOM, actuator standardization, high-volume die casting, and maintenance costs.
• Whether real-world data accumulation creates a sustained flywheel in model performance and operational efficiency.