Industrial & Physical AI:
Global Dependencies 2026–2030

Simulation · Digital Twins · Robotics · Embodied AI — Who Owns the Stack

Analysis as of February 2026 · Companion to AI Chip Buildout Analysis

Framing — Three Strands of AI Infrastructure Buildout This analysis is Strand 1 of 3. A complete picture of the AI buildout investment landscape requires examining three distinct but interconnected strands. This report covers the industrial/physical AI layer — simulation, digital twins, robotics, and embodied intelligence. The remaining two strands — agentic infrastructure software and agentic application software — will be covered in separate analyses.
Strand 1 — This Report
Industrial / Physical AI
Simulation, digital twins, robotics, embodied AI. Where software meets physics. The $50T manufacturing & logistics economy goes digital.
Strand 2 — Separate Report
Agentic Infrastructure Software
Orchestration layers, foundation model platforms, compute middleware, vector databases, inference serving. The plumbing of agentic systems.
Strand 3 — Separate Report
Agentic Application Software
Vertical AI agents, copilots, autonomous workflows, SaaS disruption. Where agentic AI meets the end user and displaces legacy software.

Core ThesisThe Next Phase: AI Meets the Physical World

The first wave of AI investment (2023–2025) was dominated by compute infrastructure — GPUs, data centers, and the chip supply chain. The second wave (2026–2030) is where AI connects to the physical world. Jensen Huang frames it as the "$50 trillion manufacturing and logistics industries" going digital. Every factory, every warehouse, every robot, every vehicle will have a digital twin, a simulation environment, and eventually autonomous decision-making capability.

This analysis maps who owns the critical layers of this physical AI stack — from simulation software and digital twin platforms through industrial automation and robotics to the emerging embodied AI frontier. Unlike the chip supply chain where bottlenecks are physical (photoresist, EUV lithography), the bottlenecks here are software-and-domain-knowledge based: decades of accumulated physics models, validated engineering workflows, and installed-base lock-in across millions of factory floors.

The landmark event is the completion of Synopsys' $35B acquisition of Ansys (July 2025), which merges silicon-level EDA with system-level multiphysics simulation — signaling that the industry views the convergence of chip design and physical-world simulation as a single investable thesis. Combined TAM: $31B and growing.

Value Chain ArchitectureThe Six-Layer Physical AI Stack

Layer 1
Simulation & Multiphysics
🇺🇸 US + 🇫🇷 France
~75%
Layer 2
EDA & Silicon Design
🇺🇸 US
>85%
Layer 3
Digital Twin Platforms
🇩🇪 Germany + 🇺🇸 US
~60%
Layer 4
Industrial Automation
🇩🇪 DE + 🇯🇵 JP + 🇺🇸 US
~70%
Layer 5
Robotics Hardware
🇯🇵 JP + 🇨🇳 CN
~65%
Layer 6
Embodied AI & Physical Intelligence
🇺🇸 US + 🇨🇳 CN
~80%

Dependency MatrixCountry Positions Across the Stack

CountrySimulation & CAEEDA / Silicon DesignDigital Twin PlatformsIndustrial AutomationRobotics HWEmbodied AIPosition
🇺🇸 United StatesDominantDominantStrongStrongEmergingDominantFull-stack hegemon
🇩🇪 GermanyStrongStrongDominantDominantStrongEmergingIndustrial middleware king
🇫🇷 FranceDominantEmergingStrongEmergingEmergingEmergingSimulation aristocrat
🇯🇵 JapanEmergingEmergingEmergingStrongDominantEmergingRobotics legacy power
🇨🇳 ChinaConstrainedConstrainedEmergingEmergingStrongStrongScale deployer, IP-light
🇰🇷 South KoreaConstrainedConstrainedEmergingStrongEmergingEmergingAdvanced mfg buyer

Market DimensionsThe Numbers: Feb 2026 Snapshot

Digital Twin Market (2025)
$24B
→ $150B by 2030
CAGR ~35–48%
Simulation Software (2024)
$72B
→ $172B by 2033
CAGR ~11.4%
EDA Software (2024)
$14B
→ $34B by 2033
CAGR ~10.1%
Industrial Robotics (2024)
$33B
5.5M units by 2026
China: 57% domestic
Synopsys+Ansys TAM
$31B
Combined EDA+Physics
$35B deal closed Jul 2025
Physical AI Addressable
$50T
Mfg + Logistics base
Per Nvidia (Huang, 2025)
CAE / CFD Market
$15B
→ $39B by 2032
CAGR ~14.4%
Humanoid Robots (2025)
~16K
Units sold globally
China majority producer

Country ProfilesSix-Node Analysis: Who Owns What

🇺🇸
United States
Full-Stack Physical AI Hegemon
The US dominates three of six layers and holds strong positions in the remaining three. In simulation and CAE, the Synopsys+Ansys combination (completed July 2025, $35B) creates a silicon-to-systems design monopoly targeting a $31B TAM with first integrated multiphysics-EDA tools shipping H1 2026. Cadence extends into system-level simulation through its Reality Digital Twin platform. On the physical AI platform layer, NVIDIA's Omniverse has become the de facto operating system for industrial digital twins — Siemens, Dassault Systèmes, Rockwell, Schaeffler, PepsiCo, and BMW all build on its libraries. NVIDIA's Cosmos world foundation models and PhysicsNeMo AI physics engine position the company as the foundational compute platform for the entire physical AI stack, analogous to its CUDA dominance in training. Rockwell Automation (FactoryTalk Twin Studio) and PTC (ThingWorx + Creo) provide the industrial IoT middleware connecting factory-floor data to digital twins. In embodied AI, the US leads through NVIDIA Isaac (robot simulation), Boston Dynamics (Spot, Atlas), and an ecosystem of startups (Bright Machines, Figure, AIGEN). The risk: US manufacturing base is thinner than Germany/Japan/China, meaning much of the physical-world deployment and data generation happens offshore.
EDA + Simulation Share
>85% global EDA
Digital Twin Platform
~38% N. America share
Key Event
Synopsys+Ansys closed
Physical AI Engine
Omniverse / Cosmos
Key Players
Synopsys (SNPS)EDA + Ansys multiphysics. $31B combined TAM. First integrated tools H1 2026.
NVIDIA (NVDA)Omniverse, Cosmos WFMs, PhysicsNeMo, Isaac Sim. Platform layer for all physical AI.
Cadence (CDNS)Reality Digital Twin, Allegro + Omniverse integration. EDA #2.
Rockwell (ROK)FactoryTalk Twin Studio. Industrial automation + digital twin for discrete mfg.
PTC (PTC)Creo CAD, ThingWorx IoT, Vuforia AR. Design-to-factory-floor digital thread.
Altair (ALTR)HyperWorks, CFD, Omniverse real-time CAE integration. Simulation democratizer.
Autodesk (ADSK)Fusion, Revit, AEC digital twins. Construction + infrastructure simulation.
Verdict: Indispensable — Software & Platform Monopolist
🇩🇪
Germany
Industrial Middleware King
Germany's position in physical AI is unique and structurally underappreciated. While the US owns the software platforms, Germany owns the industrial connective tissue — the PLM, MES, automation, and factory-floor systems that those platforms must integrate with. Siemens is the single most important company in this landscape. Its Xcelerator platform, Teamcenter PLM, Simcenter simulation, MindSphere IoT, and the newly launched Digital Twin Composer (CES 2026) make it the default industrial operating system for global manufacturing. The Siemens-NVIDIA partnership — building the "Industrial AI Operating System" — is the most consequential alliance in the physical AI space. Siemens' Erlangen electronics factory is the first blueprint for a fully AI-driven adaptive manufacturing site (2026). In EDA, Siemens (ex-Mentor Graphics) holds #3 position. Infineon provides power semiconductors for every physical AI system. Bosch and Trumpf contribute sensors and precision manufacturing. Germany's Industry 4.0 head start — particularly the installed base of ~70K Siemens Digital Industries employees serving millions of factory installations — creates a domain-knowledge moat that pure software companies cannot replicate.
Digital Twin Market DE
$2.6B 2026E
Siemens DI Workforce
~70K employees
Key Players
Siemens (SIE.DE)Xcelerator, Digital Twin Composer, Simcenter, NVIDIA partnership. The industrial OS.
Infineon (IFX.DE)#1 power semiconductors. Every robot, EV, factory controller needs power management.
KUKAIndustrial robotics (Midea/China-owned). Augsburg HQ, deep automotive integration.
BoschSensors, industrial IoT, factory automation. Private.
SchaefflerOmniverse digital twin platform. Goal: 50% of plants in Omniverse by 2030.
Verdict: Indispensable — Factory OS + Automation
🇫🇷
France
Simulation Aristocrat
France punches far above its weight through a single company: Dassault Systèmes and its 3DExperience platform. Dassault's CATIA (3D design), SIMULIA (simulation), DELMIA (manufacturing), and ENOVIA (PLM) constitute one of the two globally dominant product lifecycle management ecosystems (alongside Siemens). The freshly expanded NVIDIA-Dassault partnership (Feb 2026) integrates Omniverse libraries with 3DExperience to build validated, physics-based industrial digital twins. NVIDIA will use Dassault's MBSE technology to design AI factories (starting with Rubin platform). Dassault's strength lies in aerospace (Boeing, Airbus), automotive (Toyota, BMW), and life sciences — industries where physics simulation accuracy is non-negotiable. Schneider Electric adds complementary strength in power management and industrial IoT (EcoStruxure). The risk: Dassault is effectively the entire thesis for France — there is no second pillar.
Key Platform
3DExperience
NVIDIA Partnership
Expanded Feb 2026
Key Players
Dassault Systèmes (DSY.PA)3DExperience, SIMULIA, CATIA, DELMIA. Aerospace/auto/pharma simulation leader.
Schneider ElectricEcoStruxure platform. Smart manufacturing, power management, industrial IoT.
Verdict: Dominant in Simulation — Single-Company Thesis
🇯🇵
Japan
Robotics Legacy Power
Japan is the world's largest and most mature robotics ecosystem. FANUC (1M+ industrial robots produced), Yaskawa, Kawasaki Heavy, and Denso constitute a global supply infrastructure that is "nearly impossible to replicate." This is not just the robots — it's the intricate layers of suppliers, integrators, sensors, and decades of accumulated "industrial memory." Japan's robotics companies power automotive production lines from Detroit to Shenzhen. However, Japan's position in the software layers (simulation, digital twin, EDA) is structurally weak — it is primarily a consumer of US and European engineering software. The opportunity is in convergence: as physical AI requires robots that see, reason, and act in unstructured environments, Japan's hardware expertise becomes the critical complement to US AI software. The risk is that without ownership of the intelligence layer, Japan's hardware becomes commoditized from below by Chinese competitors (Estun, EFORT) who are rapidly closing the gap in lower-complexity applications.
FANUC Output
1M+ robots produced
Global Robot Density
#1 ecosystem depth
Key Players
FANUC (6954.T)World's #1 industrial robot maker. 1M+ units. CNC, factory automation.
Yaskawa (6506.T)#2 global robotics. Motoman series. Servo drives + motion control.
Kawasaki HeavyIndustrial + collaborative robotics. Aerospace crossover.
Keyence (6861.T)Machine vision, sensors, factory automation. Highest-margin industrial co in Japan.
DensoCompact industrial robots. Toyota ecosystem. Automotive assembly specialist.
Verdict: Hardware Legacy — Needs AI Partnership to Maintain Position
🇨🇳
China
Scale Deployer, IP-Light
China presents the most complex picture in physical AI. It is simultaneously the world's largest consumer of industrial robots (cumulative installed base surpassing 5M units in 2025, 57% now domestically produced), the fastest-moving in humanoid robotics deployment (majority of ~16K units sold in 2025), and the most aggressive in state-directed AI integration (the "AI+" initiative targets 70% AI penetration across key sectors by 2027). Yet China remains structurally dependent on Western simulation and EDA software — Siemens, Dassault, Synopsys, and Cadence tools run in nearly every Chinese factory and design center. Beijing's response is a dual strategy: massive subsidies ($20B+ allocated to robotics in 2024–2025) plus localization pressure on software tools. Embodied AI champions are emerging — UBTech (humanoids deployed at Zeekr factory, powered by DeepSeek R1), Unitree ($5,900 humanoid), DJI (drones), Baidu Apollo (autonomous vehicles). The key dynamic: China may lack the simulation IP stack but compensates with unmatched deployment scale and data generation. Every factory floor running Chinese robots generates training data that feeds back into embodied AI models — a flywheel the West cannot match in volume.
Installed Robot Base
5M+ units (2025)
Domestic Production
57% of demand
Robot Industry Rev
$33B 2024
State Subsidy
$20B+ 2024–2025
Key Players
UBTechWalker S2 humanoid. Zeekr factory deployment. DeepSeek-R1 powered reasoning.
UnitreeLow-cost humanoid ($5,900). Mass production pioneer. H1/G1 series.
Estun (002747.SZ)"Little Dragon" of industrial robots. Servo motors → full robot stack. SCARA #2 domestic.
DJIDrone monopolist. Agricultural + industrial inspection. ~70% global consumer drone share.
Baidu ApolloAutonomous driving platform. L4 robotaxi fleets in multiple Chinese cities.
Huawei FusionPlantIndustrial data infrastructure. AI solutions for 65% of state-owned enterprises.
Verdict: Scale Challenger — Deployment Speed vs. IP Dependency
🇰🇷
South Korea
Advanced Manufacturing Buyer
South Korea's physical AI position derives from its world-class manufacturing density — particularly in semiconductors (Samsung, SK Hynix fabs), shipbuilding (Hyundai Heavy, Samsung Heavy), automotive (Hyundai/Kia), and consumer electronics (Samsung, LG). These are among the most sophisticated deployers of digital twin and factory simulation technology globally, making Korea a critical demand node even though it originates relatively little physical AI platform software. Hyundai Robotics is emerging as a serious industrial robot OEM, and HD Hyundai's ownership of Boston Dynamics gives Korea a direct stake in the embodied AI frontier. Samsung's semiconductor fabs are among the most digitized industrial environments on earth. Korea's role: the world's most advanced buyer-deployer, with potential to move up-stack through Hyundai's robotics ambitions.
Robot Density
#2 global per capita
Key Integrator
HD Hyundai + BostonD
Key Players
Hyundai RoboticsIndustrial robots. HD Hyundai parent owns Boston Dynamics.
SamsungWorld's most digitized fabs. Smart factory technology. Heavy PLM software consumer.
Doosan RoboticsCollaborative robots. IPO 2023. Growing international footprint.
Verdict: Advanced Deployer — Up-stack Potential via Hyundai/Boston Dynamics

Structural ConstraintsCritical Bottlenecks: 2026–2030

🔧
Simulation-to-Reality Gap
The biggest barrier to physical AI scale: models trained in simulation fail in the real world. Bridging this "sim-to-real" gap requires photorealistic rendering (NVIDIA RTX), accurate physics engines (PhysX, Warp), and massive real-world validation data that only actual factory/vehicle/robot deployments generate. China's deployment scale gives it a data advantage the West must match through better simulation fidelity.
📐
Domain Knowledge Lock-in
Physics simulation is not a commodity. Decades of validated models in CFD, structural analysis, electromagnetics, and thermal dynamics are embedded in tools from Ansys/Synopsys, Dassault SIMULIA, and Siemens Simcenter. New entrants cannot replicate this validation corpus. The Synopsys+Ansys merger deepens this moat by fusing chip-level and system-level physics into a single validated stack.
🏭
Installed Base Inertia
Millions of factories worldwide run on Siemens PLCs, Rockwell controllers, and Dassault/Siemens PLM systems. Switching costs are measured in years and hundreds of millions of dollars. Digital twin platforms must integrate with existing infrastructure — they cannot replace it. This massively favors incumbents (Siemens, Dassault, Rockwell) over cloud-native challengers.
🤖
Robotics Intelligence Gap
Current industrial robots are highly capable at structured, repetitive tasks but lack visual reasoning and adaptability for unstructured environments. Vision-Language-Action (VLA) models (500M–7B parameters as of mid-2025) are the bridge, but commercially viable autonomous humanoid robotics remain 3–5 years out. Hardware is ahead of the intelligence software.
🔒
Data Interoperability (OpenUSD)
Industrial digital twins require data from dozens of incompatible systems: CAD, PLM, MES, SCADA, IoT sensors. NVIDIA's push for OpenUSD as the universal standard is gaining traction (Siemens, Dassault, Pixar/Apple backing), but the "Tower of Babel" problem across factory-floor data formats remains the #1 deployment friction.
⚡
GPU Compute for Simulation
Real-time, physics-accurate digital twins require massive GPU compute — NVIDIA RTX PRO 6000 (Blackwell) for rendering, CUDA-X for simulation acceleration. Siemens is GPU-accelerating its entire simulation portfolio. This creates a secondary demand loop for NVIDIA silicon: not just for AI training, but for industrial simulation.

Strategic ArchitectureThe Alliance Map: Who Partners With Whom

NVIDIA is the Platform — Everyone Else Orbits
The single most important structural observation: NVIDIA's Omniverse has become the gravitational center of the physical AI stack. Both Siemens and Dassault Systèmes — the two largest industrial software companies — have entered deep platform partnerships with NVIDIA rather than competing. This is the CUDA lock-in strategy replicated at the industrial layer. Every partner that builds on Omniverse libraries creates dependency on NVIDIA's GPU compute stack.

Alliance 1: Siemens × NVIDIA — "The Industrial AI Operating System." Announced CES 2026. GPU acceleration across Siemens' entire simulation portfolio. Digital Twin Composer built on Omniverse. First AI-driven factory blueprint: Erlangen 2026. Joint design for next-gen AI factories. The most comprehensive industrial partnership in the space.

Alliance 2: Dassault Systèmes × NVIDIA — Expanded Feb 2026. Omniverse libraries integrated into DELMIA (factory digital twins). NVIDIA will use Dassault's MBSE for Rubin AI factory design. BioNeMo + Biovia for molecule/materials discovery. 3DExperience becomes an Omniverse-native platform.

Alliance 3: Synopsys + Ansys (Merged) — $35B closed July 2025. Silicon-to-systems simulation. Multiphysics fused into EDA stack. First combined tools H1 2026. Targets automotive, aerospace, industrial — the same verticals as Siemens and Dassault, but from the chip upward rather than the factory downward.

Alliance 4: China's Stack — Less coordinated but rapidly assembling: DeepSeek/Baidu AI models + UBTech/Unitree robotics hardware + Huawei FusionPlant industrial data infrastructure + massive state subsidies. The differentiator: deployment scale (5M+ robots installed) generating proprietary training data. Weakness: still dependent on Western simulation and EDA tools.

ScenariosRisk Scenarios: 2026–2030

Bull: Physical AI Accelerates
Sim-to-real gap narrows faster than expected via Cosmos world models and PhysicsNeMo. PepsiCo's 20% throughput gain from Siemens Digital Twin Composer becomes the norm — every Fortune 500 manufacturer deploys digital twins by 2028. Humanoid robots reach commercial viability by 2029. Digital twin market hits $150B+ by 2030. NVIDIA's Omniverse becomes as essential to factories as Windows is to offices. Germany and France software companies re-rate as "AI infrastructure."
Bear: Deployment Friction Dominates
Installed base inertia proves overwhelming — factories built on 20-year-old SCADA systems resist digital twin integration. Data interoperability remains the "Tower of Babel" despite OpenUSD push. Sim-to-reality gap persists, limiting ROI of digital twins to visualization rather than autonomous decision-making. Humanoid robots remain lab curiosities. Chinese AI localization pressure fragments the global simulation software market. Digital twin market grows but at the low end (~$80B by 2030).
Wild Card: China Leapfrogs via Embodied AI
China's deployment-first strategy pays off: 5M+ robots generate unmatched real-world training data, enabling Chinese VLA models to surpass simulation-trained Western models in practical effectiveness. UBTech/Unitree humanoids reach $5K price points while Western equivalents remain >$20K. China captures the embodied AI layer while the West retains the simulation/design layer — a mirror of the chip supply chain split. Geopolitical implications: two incompatible physical AI ecosystems emerge.
Wild Card: Simulation Software Gets AI-Disrupted
AI-native physics solvers (neural operators, physics-informed neural networks) begin replacing traditional FEA/CFD solvers for certain classes of problems — delivering 1000× speedups at 95% accuracy. This would undermine the validation-corpus moat protecting Ansys/Synopsys, Dassault SIMULIA, and Siemens Simcenter. Early signs: NVIDIA PhysicsNeMo, Neural Concept, and Google DeepMind's physics research. Timeline: unlikely before 2028 for safety-critical applications, but possible earlier for consumer products and facility optimization.

SummaryCountry Verdicts

CountryRoleStructural AdvantageKey RiskVerdict
🇺🇸 United StatesFull-stack hegemonEDA monopoly + Omniverse platform + embodied AI researchThin manufacturing base; deployment data generated offshoreIndispensable
🇩🇪 GermanyIndustrial middleware kingSiemens installed base + Industry 4.0 domain knowledge + automationSlow AI adoption; KUKA now Chinese-owned; demographic headwindsIndispensable
🇫🇷 FranceSimulation aristocratDassault 3DExperience dominance in aerospace/auto/pharmaSingle-company thesis; no robotics or automation pillarDominant (narrow)
🇯🇵 JapanRobotics legacy powerFANUC/Yaskawa/Keyence — irreplaceable hardware ecosystemWeak in software/AI; Chinese competitors closing gap on hardwareImportant — Needs AI Partner
🇨🇳 ChinaScale deployer5M+ robots deployed; unmatched real-world data; state backingDependent on Western simulation/EDA; export controls escalation riskChallenger — Watch Closely
🇰🇷 South KoreaAdvanced buyerWorld-class deployer; Hyundai/Boston Dynamics bridgeLittle originating IP in simulation/platform layerDeployer — Upside via Robotics
The Meta-Observation: Software Eats the Factory
The physical AI buildout inverts the traditional industrial power map. Countries that dominate physical manufacturing (China, Japan, Korea) are consumers of Western engineering software. Countries that dominate engineering software (US, Germany, France) have thinner manufacturing bases. The structural tension: the software layer captures most of the value (Synopsys trades at 40×+ P/E, Siemens Digital Industries at 25×+), but the deployment layer generates the data that feeds the next generation of AI models. China's bet is that data > software over the long run. The West's bet is that validated physics simulation > brute-force data. The Synopsys+Ansys merger, the Siemens-NVIDIA and Dassault-NVIDIA alliances all point to the same conclusion: the incumbents believe the answer is both — and are building platforms that fuse simulation with AI-generated physics.
Analysis as of February 2026. This report covers Strand 1 (Industrial / Physical AI) of a three-part AI buildout series. Strand 2 (Agentic Infrastructure Software) and Strand 3 (Agentic Application Software) to follow in separate analyses. All market figures from cited research reports and company disclosures. This is not investment advice.