McKinsey & Company · Published 2026-09-10

Physical AI: The hidden value pools leaders are missing

privateReview: unreviewedConfidence: high
physical AIroboticsautomationsupply chainvalue poolsphysical-aisim-to-realevaluationdata-flywheeldeploymentsupply-chainvalue-chain
Open on YouTube ↗

Summary

Overview

  • Speaker: Lucia Rahilly, Ani Kelkar, Mark Patel
  • Channel: McKinsey & Company
  • Main topic: Physical AI and its economic potential, value pools, and strategic implications for business leaders
  • Purpose: To educate business leaders on the shift towards Physical AI, identify hidden economic value pools, analyze hardware and software value chain distributions, and provide actionable strategies for adopting physical AI technologies. McKinsey Global Editorial Director Lucia Rahilly hosts partners Ani Kelkar and Mark Patel to discuss Physical AI—the transition of AI systems from digital/virtual spaces into physical operations, autonomous vehicles, manufacturing, robotics, and industrial environments. They explore macroeconomic figures showing physical work accounts for 44% of global working hours, detail the addressable market scale up to $370B by 2040, outline China's supply chain lead in EV and robotics components, and provide business leaders with a three-step framework to build, invest, and embrace physical AI.

Topic Map

Introduction to Physical AI

  • Explanation: Lucia Rahilly introduces the concept that machines are moving from following static instructions to sensing, deciding, and acting in the physical world.
  • Key claims:
    • AI is moving beyond the virtual space into autonomous vehicles, factories, warehouses, and medical systems.
    • Humanoid robots capture immense attention but are expected to account for a tiny fraction (2% by 2045) of physical AI value.
  • Examples:
    • Robots moving from pre-programmed repetitive trajectories to goal-fulfilling adaptive intelligence.
  • Terminology:
    • Physical AI
    • Humanoid robot
    • Sense-decide-act architecture
  • Why it matters: Business leaders often focus solely on humanoid robots while missing the broader ecosystem of physical AI applications.

The Scale of the Physical Economy

  • Explanation: Mark Patel outlines the massive scale of physical work in the global economy relative to digital work.
  • Key claims:
    • 44% of all paid work hours globally across sectors are tied to physical work.
    • An average person in the US spends 12 hours a week on unpaid household physical work.
    • Out of 60 million full-time worker equivalents of annual tasks, only 13% are automatable today.
  • Examples:
    • Cleaning, cooking, manufacturing, construction, transportation, and warehousing.
  • Terminology:
    • Physical economy
    • Paid work hours
    • Automataable tasks
  • Why it matters: Demonstrates the immense addressable market and scale of physical work that physical AI can impact.

Market Sizing and Addressable Value

  • Explanation: Ani Kelkar breaks down McKinsey's robotics market model scaling up to $370B by 2040.
  • Key claims:
    • General-purpose robotics could scale to $370B by 2040, compared to less than $1B today.
    • 60% of the market is in logistics and discrete manufacturing.
    • 40% is in pick & place and simple tool manipulation tasks.
  • Examples:
    • Logistics ($115B), discrete manufacturing ($100B), labor services ($50B), consumer packaged goods ($30B).
  • Terminology:
    • Total Addressable Market (TAM)
    • General-purpose robotics
    • Task platform
  • Why it matters: Provides clear valuation metrics and timeline expectations for market expansion.

Closing the Last-MM Gap and Data Reliability

  • Explanation: Mark Patel explains the four-step framework required to get models from demo to production reliability.
  • Key claims:
    • Demos do not equal production reliability; real-world deployment data is required to bridge the last-mm gap.
    • Simulation and synthetic data accelerate initial training, but real-world edge cases are essential.
    • Adoption, expansion, and invention are the three historical forms of value creation.
  • Examples:
    • Fine-tuning models with real-world data to close the dexterous manipulation gap.
  • Terminology:
    • Last-mm gap
    • Synthetic data
    • Data pyramid
    • Model fine-tuning
  • Why it matters: Explains why companies cannot rely solely on demos and must invest in real-world data integration.

Hardware Supply Chain and EV Overlaps

  • Explanation: Mark Patel discusses China's early lead in robotics hardware supply chains driven by the EV industry.
  • Key claims:
    • China's electric vehicle (EV) supply chain gives it a decisive early lead in robotics hardware capacity.
    • High robotics overlap exists in power electronics, harmonic drives, motors, permanent magnets, precision bearings, and actuators.
    • Unit cost per unit of performance will define the winning supply chain.
  • Examples:
    • EV power systems and motor controls overlapping with robotics actuators.
  • Terminology:
    • Supply chain overlap
    • Power electronics
    • Harmonic drives
    • Precision bearings
  • Why it matters: Highlights geopolitical and supply chain advantages in manufacturing physical AI hardware.

Value Pool Distribution across Hardware, Software, and Services

  • Explanation: Ani Kelkar presents the value potential heatmap across value chain segments by 2035.
  • Key claims:
    • Hardware (training hardware, platform hardware, form factor hardware) captures 25-35% of value.
    • Software (training intelligence, cloud applications, onboarding software) captures 20-30% of value.
    • Deployment and field operations capture the remaining value, with software/operations in field ops capturing 15-25%.
  • Examples:
    • Software layer improvements in redesigning workflows and enterprise adoption.
  • Terminology:
    • Value chain segment
    • Embodied Physical AI
    • Value heatmap
  • Why it matters: Helps business leaders decide whether to invest in hardware, software, or deployment services.

Key Points

Shift from instruction-following to sense-decide-act

  • Explanation: Robots are evolving from performing fixed repetitive tasks to perceiving environments and making autonomous decisions.
  • Evidence: Autonomous laundry folding and navigation robots demonstrating real-time adaptation.
  • Practical implication: Companies can deploy systems into unstructured and complex environments.

The Last-MM Gap in Physical AI

  • Explanation: Moving from controlled laboratory demos to robust production reliability requires real-world data collection.
  • Evidence: Data pyramid model showing real-world deployment data feeds edge-case improvements.
  • Practical implication: Invest in data collection and continuous fine-tuning loops rather than buying single demo solutions.

EV Supply Chain Synergy

  • Explanation: Electric vehicle manufacturing capabilities directly translate into robotics component manufacturing advantages.
  • Evidence: China's high robotics component overlap with EV supply chains across actuators and bearings.
  • Practical implication: Assess hardware sourcing strategies through the lens of existing EV manufacturing ecosystems.

Frameworks, Models & Processes

The Four-Step Production Reliability Framework

  • How it works: A continuous loop to take models from initial skill to production reliability.
  • Components:
    • Build reliable robot hardware capable of dexterous manipulation
    • Generate high-fidelity, diverse, and large-volume data to feed model performance
    • Create next-gen model architecture with greater accuracy
    • Deploy robot to real-world with reliable mission execution
  • When to use: When scaling physical AI models from simulation/demo to industrial production.

How Business Leaders Can Capture Long-Term Value

  • How it works: Three-pronged strategic approach for executives entering physical AI.
  • Components:
    • Build: Develop differentiated Physical AI solutions
    • Invest: Develop a Physical AI investment strategy
    • Embrace: Adopt Physical AI to transform operations
  • When to use: When formulating enterprise AI and robotics transformation roadmaps.

Examples & Case Studies

Boston Dynamics and autonomous laundry-folding robots operating in dynamic spaces.

  • Illustrates: The shift from pre-programmed trajectory automation to sense-decide-act physical AI systems.
  • Lesson: Machines can adapt to irregular objects and moving environments when endowed with physical AI.

Actionable Takeaways

  • Immediate:
    • Focus on tasks rather than robots when evaluating business value.
    • Audit internal workflows to identify automatable physical tasks.
  • Strategic:
    • Incorporate physical AI into enterprise transformation and investment strategies.
    • Evaluate hardware supply chain resilience and EV-robotics component overlaps.
  • Questions to investigate:
    • Where are our biggest labor bottlenecks in physical operations?
    • How are we capturing and utilizing real-world deployment data?

Claims Worth Verifying

  • General-purpose robotics could scale to $370B by 2040. (Market forecast)
  • 44% of all paid work hours globally are tied to physical work. (Economic statistic)
  • Humanoid robots are expected to account for just 2% of physical AI value by 2045. (Industry projection)

Notable Quotes

"The quest to automate physical work isn't new; we humans have been at it for hundreds of years at this point." (at 6:54) "Every AI transformation at its heart is a people transformation." (at 1:56) "Physical work remains essential today." (at 16:47)

Compressed Summary

  • Physical AI transitions machines from static programming to sense-decide-act capabilities.
  • 44% of global paid work hours are tied to physical work, representing a massive addressable market.
  • Robotics TAM could scale to $370B by 2040.
  • EV supply chains provide countries like China with a decisive lead in robotics hardware components.
  • Capturing long-term value requires building solutions, investing in strategy, and embracing operational transformation.
  • Keywords: physical ai, robotics, automation, supply chain, value pools
  • Core insight: Physical AI extends artificial intelligence into the physical economy, unlocking hundreds of billions in value by transforming how manual and operational tasks are executed.

Core insights

6
Mechanismmedium noveltymoderate evidence

Demos are not production reliability: synthetic/simulation data accelerates initial training, but robust deployment requires real-world edge-case data collected in the field, framed as the 'last-mm gap' and a 'data pyramid' where deployment data feeds back into fine-tuning.

Why it matters

Treats deployment as a data-generation and continuous-fine-tuning stage rather than the end of the pipeline, so reliability work belongs in the runtime/field-operations layer, not only in offline training.

Generalization

Any agent or model that must operate in unstructured conditions needs a closed loop from production edge cases back into training; the gap between benchmark/demo performance and field reliability is a first-class engineering target.

Demos do not equal production reliability; real-world deployment data is required to bridge the last-mm gap.
Open source video
Simulation and synthetic data accelerate initial training, but real-world edge cases are essential.
Open source video
Invest in data collection and continuous fine-tuning loops rather than buying single demo solutions.
Open source video
Mental Modelhigh noveltymoderate evidence

Value in physical AI is distributed, not concentrated in the headline form factor: hardware captures 25-35%, software 20-30%, and deployment/field operations carry the remainder with software/operations in field ops at 15-25% by 2035, while humanoid robots are projected at only ~2% of value by 2045.

Why it matters

Attention (humanoids) is inversely related to value (deployment, field ops, narrow task platforms), so investment and architecture decisions anchored on the visible artifact misallocate effort.

Generalization

Emerging technology value often sits in integration, deployment, and operations layers rather than the most visible component; map value pools explicitly before choosing where to build.

Humanoid robots capture immense attention but are expected to account for a tiny fraction (2% by 2045) of physical AI value.
Open source video
Hardware (training hardware, platform hardware, form factor hardware) captures 25-35% of value.
Open source video
Deployment and field operations capture the remaining value, with software/operations in field ops capturing 15-25%.
Open source video
Empirical Resultmedium noveltymoderate evidence

Demand is concentrated in narrow, mundane task classes: 60% of the projected robotics market is logistics and discrete manufacturing, and 40% is pick & place and simple tool manipulation, modeled to reach $370B by 2040 from under $1B today.

Why it matters

Value accrues to constrained task platforms, not general-purpose capability, which favors scoped, decomposable task abstractions over open-ended autonomy claims.

Generalization

Deploying agents against well-bounded, high-volume task classes yields more value than chasing general capability; define the task platform boundary explicitly.

General-purpose robotics could scale to $370B by 2040, compared to less than $1B today.
Open source video
60% of the market is in logistics and discrete manufacturing.
Open source video
40% is in pick & place and simple tool manipulation tasks.
Open source video
Empirical Resultlow noveltymoderate evidence

The addressable scale is far larger than current automation: physical work accounts for 44% of global paid work hours and 60 million full-time worker equivalents of annual tasks, yet only 13% are automatable today.

Why it matters

The 13%-vs-44% gap quantifies the headroom and implies most value will be unlocked by gradual expansion rather than current deployable capability.

Generalization

When current automatable share is far below the total task share, expect incremental capability expansion to be the dominant value driver, not a single breakthrough.

44% of all paid work hours globally across sectors are tied to physical work.
Open source video
Out of 60 million full-time worker equivalents of annual tasks, only 13% are automatable today.
Open source video
Mechanismhigh noveltymoderate evidence

Supply-chain advantage flows from shared component ecosystems: China's EV supply chain overlaps with robotics across power electronics, harmonic drives, motors, permanent magnets, precision bearings, and actuators, and the winning supply chain is defined by unit cost per unit of performance.

Why it matters

Frames hardware competitiveness as an existing-ecosystem leverage problem and introduces 'cost per unit of performance' as the operative supply-chain metric.

Generalization

Reuse of an adjacent manufacturing base can determine early dominance in a new hardware category; metric selection (cost per unit performance) shapes which supplier wins.

China's electric vehicle (EV) supply chain gives it a decisive early lead in robotics hardware capacity.
Open source video
High robotics overlap exists in power electronics, harmonic drives, motors, permanent magnets, precision bearings, and actuators.
Open source video
Unit cost per unit of performance will define the winning supply chain.
Open source video
Architecturelow noveltyweak evidence

The architectural shift is from instruction-following to goal-fulfilling sense-decide-act behavior, with historical value creation categorized as adoption, expansion, and invention.

Why it matters

Reframes the system boundary: perception, decision, and action become the loop that replaces static pre-programmed trajectories, which is a control-loop rather than a pipeline.

Generalization

Moving from static instruction execution to closed perception-decision-action loops is the recurring architecture pattern when systems enter unstructured environments.

Robots moving from pre-programmed repetitive trajectories to goal-fulfilling adaptive intelligence.
Open source video
Adoption, expansion, and invention are the three historical forms of value creation.
Open source video

Deep dives

4

Closing the last-mm gap: data requirements and reliability curves

Research question

What volume and diversity of real-world deployment data is required to close the last-mm gap for a given manipulation skill, and how does reliability scale with field-data volume and fine-tuning iterations?

Why

The last-mm gap is the difference between demo and production reliability, and quantifying the data-reliability relationship would turn field deployment from a cost center into a measurable engineering investment.

Demos do not equal production reliability; real-world deployment data is required to bridge the last-mm gap.
Open source video
Simulation and synthetic data accelerate initial training, but real-world edge cases are essential.
Open source video
Source video

Decomposing the 87% non-automatable physical tasks

Research question

Why is only 13% of physical tasks automatable today, and which technical constraints (dexterity, variability, data requirements) bound the remaining 87%?

Why

Identifying the binding constraints would focus R&D on the highest-leverage technical bottlenecks rather than broad capability claims.

Out of 60 million full-time worker equivalents of annual tasks, only 13% are automatable today.
Open source video
44% of all paid work hours globally across sectors are tied to physical work.
Open source video
Source video

Operationalizing cost per unit of performance for robotics supply chains

Research question

How should 'unit cost per unit of performance' be operationalized as a measurable metric for robotics components and supply chains, and does it predict which suppliers win?

Why

The metric is proposed as the defining factor for supply chain competitiveness, but without a concrete definition and benchmark, it cannot guide sourcing or investment decisions.

Unit cost per unit of performance will define the winning supply chain.
Open source video
High robotics overlap exists in power electronics, harmonic drives, motors, permanent magnets, precision bearings, and actuators.
Open source video
Source video

Value pool migration across hardware, software, and field operations

Research question

How will value pool shares across hardware, software, and field operations shift over time as software improves and hardware costs fall?

Why

The static 2035 heatmap does not indicate whether value will migrate between layers, which is critical for long-term investment and build-vs-invest decisions.

Hardware (training hardware, platform hardware, form factor hardware) captures 25-35% of value.
Open source video
Deployment and field operations capture the remaining value, with software/operations in field ops capturing 15-25%.
Open source video
Source video

Article ideas

4

The Humanoid Mirage: Where Physical AI Value Actually Lives

The disproportionate attention on humanoid robots misallocates capital and talent; the real value pools are in deployment, field operations, and narrow task platforms like pick & place and simple tool manipulation.

Angle

Contrasts public narrative with McKinsey's value distribution, urging leaders to map value pools before choosing where to build.

Source video

Deployment Is the New Training: Engineering the Last-mm Gap in Physical AI

Production reliability in physical AI cannot be achieved by simulation alone; it requires treating field deployment as a continuous data-generation and fine-tuning loop, making runtime engineering a first-class discipline.

Angle

Argues that the last-mm gap is an engineering target, and outlines the feedback plumbing needed between production and training pipelines.

Source video

The 13% Ceiling: Why Physical AI's Growth Will Be Incremental

With only 13% of physical tasks automatable today, value will accrue through gradual capability expansion across bounded task classes rather than through a single general-purpose breakthrough, so strategy should focus on scoped task platforms.

Angle

Uses the 13% figure and the 44% physical work share to argue for incremental deployment over revolutionary hype.

Source video

The EV Inheritance: How Adjacent Supply Chains Decide Robotics Winners

Robotics hardware competitiveness is being inherited from the EV supply chain through shared components, and the winning supply chain will be defined by unit cost per unit of performance, not by novel form factors.

Angle

Examines component overlap and cost-performance metric as the new basis of competition.

Source video

Project ideas

4

Field Data Flywheel for Pick & Place

beyond-evals

For a pick & place task, reliability improves at a measurable rate as a function of real-world edge-case data volume collected in deployment, and fine-tuning with field data closes the last-mm gap faster than additional simulation data.

Proof of concept

Deploy a pick & place robot in a controlled warehouse-like environment; instrument it to log all edge cases; run periodic fine-tuning with collected data; measure task success rate vs. data volume. Compare against a simulation-only baseline that receives an equivalent amount of synthetic data.

Measurement

Task success rate (e.g., picks per hour without human intervention) as a function of number of edge-case samples and fine-tuning iterations; compare against simulation-only baseline.

Source video

Cost-per-Performance Benchmark for Robotics Components

new

A standardized benchmark suite mapping component cost (e.g., actuators, harmonic drives) to performance on a set of manipulation tasks will reveal that suppliers with lower cost per unit of performance are not the cheapest per unit, and that EV-derived components dominate on this metric.

Proof of concept

Build a benchmark suite of 5 manipulation tasks (pick & place, peg insertion, etc.); measure performance (success rate, cycle time) for actuators from different suppliers; compute cost per unit performance; compare EV-derived vs. purpose-built components.

Measurement

Cost per successful task completion normalized by cycle time; ranking of suppliers.

Source video

Value Pool Migration Simulator

new

Under scenarios of improving software capability and falling hardware costs, the share of value captured by software and field operations will increase relative to hardware, and the humanoid form factor will remain a small fraction of total value.

Proof of concept

Build a simple economic model with parameters (software improvement rate, hardware cost decline, adoption curves) to simulate value pool shares from 2025 to 2045; calibrate with available market data; test sensitivity.

Measurement

Projected value shares by layer over time; sensitivity to key parameters.

Source video

Task Decomposition for Automatability

beyond-evals

By decomposing physical tasks into primitives (dexterity, variability, data requirements), the 13% automatable share can be explained primarily by variability and data requirements rather than dexterity.

Proof of concept

Take a sample of physical tasks from logistics and manufacturing; classify by dexterity, variability, and data availability; correlate with current automation feasibility; identify binding constraints.

Measurement

Correlation between task properties and automation feasibility; classification accuracy of a predictive model.

Source video

Architectural implications

3

Deployment and field operations are a distinct value pool (15-25% via software/operations in field ops), not just a support function.

Before

Field deployment treated as one-time integration; training data generated offline from simulation only.

After

Runtime ingests field edge cases as first-class data, feeding continuous fine-tuning loops back into the model.

Consequence

Responsibility for data collection, telemetry, and retraining moves into the operational layer, requiring feedback plumbing between production and training pipelines.

Source video

Value concentrates in narrow task classes (pick & place, simple tool manipulation) rather than general-purpose form factors.

Before

Building toward broad general-purpose capability as the primary abstraction.

After

Scoping capability around a task platform with explicit boundaries and decomposable skills.

Consequence

System design optimizes for bounded task performance and throughput rather than open-ended generalization, changing evaluation and tooling choices.

Source video

Hardware, software, and deployment each capture a comparable but non-dominant share of value (25-35%, 20-30%, 15-25%).

Before

Assuming a single layer (e.g., model or robot) captures most of the economics.

After

Allocating investment across a layered stack with explicit value-per-layer mapping.

Consequence

Teams must decide layer ownership deliberately, since no single layer monopolizes returns.

Source video

Tradeoffs and failure modes

4

Simulation and synthetic data vs real-world data

Benefit

Simulation and synthetic data accelerate initial model training at scale and low cost.

Cost or risk

They fail to cover real-world edge cases, so models remain unreliable in production until field data closes the last-mm gap.

Simulation and synthetic data accelerate initial training, but real-world edge cases are essential.
Open source video
Source video

Humanoid form factor vs broader physical AI ecosystem

Benefit

Humanoid robots attract disproportionate attention and investment mindshare.

Cost or risk

They are projected to be a tiny fraction of physical AI value, so focus on them risks missing larger hidden value pools.

Humanoid robots capture immense attention but are expected to account for a tiny fraction (2% by 2045) of physical AI value.
Open source video
Source video

Concentrated EV-derived supply chain advantage

Benefit

Overlapping EV and robotics components give incumbents a decisive early lead in hardware capacity.

Cost or risk

Concentration in a single ecosystem creates dependency and geopolitical exposure for robotics hardware sourcing.

China's electric vehicle (EV) supply chain gives it a decisive early lead in robotics hardware capacity.
Open source video
Source video

Demo-driven evaluation vs production reliability

Benefit

Demos prove a capability is feasible and are fast to produce.

Cost or risk

Demos overstate deployability, leading buyers to adopt solutions that fail outside controlled conditions.

Demos do not equal production reliability; real-world deployment data is required to bridge the last-mm gap.
Open source video
Source video

Open questions

5

What volume and diversity of real-world deployment data is required to close the last-mm gap for a given manipulation skill?

Why unresolved

The summary names the data pyramid and the last-mm gap but gives no quantitative relationship between field-data volume and reliability gains.

Research direction

Instrument a field deployment to measure reliability as a function of real-world edge-case data collected and fine-tuning iterations.

Source video

How should 'unit cost per unit of performance' be operationalized as a measurable metric for robotics components and supply chains?

Why unresolved

The summary states that this metric defines the winning supply chain but does not define performance units or how to compare them.

Research direction

Define a benchmark suite mapping component cost to standardized task performance to enable cross-supplier comparison.

Source video

How will value pool shares across hardware, software, and field operations shift over time as software improves?

Why unresolved

The summary gives a static 2035 heatmap without dynamics or the conditions that would shift value between layers.

Research direction

Model layer-value migration under scenarios of improving software capability and falling hardware costs.

Source video

Why is only 13% of physical tasks automatable today, and which technical constraints bound the remaining 87%?

Why unresolved

The summary reports the figure but does not decompose which task properties (dexterity, unstructured environments, edge cases) block automation.

Research direction

Decompose non-automatable tasks by dexterity, variability, and data requirements to identify the binding constraints.

Source video

Do historical value-creation modes (adoption, expansion, invention) predict which physical AI entrants capture durable value?

Why unresolved

Presented as a framing claim with no supporting evidence or mapping to physical AI companies.

Research direction

Map current physical AI players to adoption/expansion/invention categories and track value capture over time.

Source video

Key claims

8
predictionVerification needed

General-purpose robotics could scale to $370B by 2040, from less than $1B today.

Evidence

General-purpose robotics could scale to $370B by 2040, compared to less than $1B today.

Question

What modeling assumptions and adoption curves underpin the $370B by 2040 figure?

Source video
predictionVerification needed

Humanoid robots are expected to account for only about 2% of physical AI value by 2045.

Evidence

Humanoid robots capture immense attention but are expected to account for a tiny fraction (2% by 2045) of physical AI value.

Question

What market model produces the 2% humanoid share by 2045, and how sensitive is it to hardware cost declines?

Source video
factualVerification needed

Physical work accounts for 44% of all paid work hours globally.

Evidence

44% of all paid work hours globally across sectors are tied to physical work.

Question

Which sectors and geographies are included, and how is 'physical work' defined and measured?

Source video
factualVerification needed

Only 13% of 60 million full-time worker equivalents of annual physical tasks are automatable today.

Evidence

Out of 60 million full-time worker equivalents of annual tasks, only 13% are automatable today.

Question

What automation-capability threshold defines 'automatable' and how was the 13% computed?

Source video
comparativeVerification needed

60% of the robotics market is in logistics and discrete manufacturing.

Evidence

60% of the market is in logistics and discrete manufacturing.

Question

Is this market share measured by revenue, unit volume, or task count?

Source video
predictionVerification needed

Hyperscaler-scale value split: hardware captures 25-35%, software 20-30%, and software/operations in field ops 15-25% of physical AI value by 2035.

Evidence

Hardware (training hardware, platform hardware, form factor hardware) captures 25-35% of value.

Question

How are value-chain segment boundaries defined and are these gross margins or revenue shares?

Source video
causalVerification needed

China's EV supply chain gives it a decisive early lead in robotics hardware capacity due to component overlap.

Evidence

China's electric vehicle (EV) supply chain gives it a decisive early lead in robotics hardware capacity.

Question

How much of robotics hardware capacity is genuinely shared with EV supply chains versus purpose-built?

Source video
causalVerification not requested

Demos do not equal production reliability; real-world deployment data is required to bridge the last-mm gap.

Evidence

Demos do not equal production reliability; real-world deployment data is required to bridge the last-mm gap.

Source video

Connections

5