The global robot foundation models market was valued at USD 0.62 billion in 2025. It is estimated at USD 0.99 billion in 2026 and is projected to reach USD 24.85 billion by 2036, growing at a CAGR of 38.0% over the 2026-2036 forecast period. The report examines model architectures, robot data, hardware, safety rules, business models, investment and M&A activity, and competition.
The study covers broad AI models that give robots transferable perception, reasoning, and motor skills, plus the tools, datasets, and services that tailor these models to specific machines and put them into use. Model types include vision-language-action (VLA) models that convert camera input and instructions into motion, embodied reasoning models that plan multi-step tasks, world models that predict how scenes change, and models that control walking and full-body movement. Revenue is counted from licenses and APIs, paid support for open models, development platforms, training data, integration work, and the share of robot value attributable to embedded model software. Robot hardware, AI chips, and traditional task-specific programming tools are outside the scope.
According to company announcements, capital is flowing into the field at a record pace: Figure's September 2025 Series C brought in more than USD 1 billion and valued the company at USD 39 billion after the round, Physical Intelligence closed a USD 600 million round in November 2025 that valued it at USD 5.6 billion, and Skild AI secured nearly USD 1.4 billion in January 2026 at a valuation of more than USD 14 billion. Underlying automation demand is large: according to the International Federation of Robotics, 542,000 industrial robots were installed globally in 2024, and China took more than half.
Training data is the biggest constraint, as the 2023 Open X-Embodiment project, a joint effort of 21 institutions, gathered roughly one million real trajectories across 22 kinds of robots, compared with about 300 trillion tokens of public text for language models according to Epoch AI. Safety rules also slow deployment: the EU Machinery Regulation (EU) 2023/1230, which applies from 20 January 2027, sets requirements for machinery with self-evolving behavior based on machine learning, and standards for learned robot policies in unstructured settings are still immature. Valuations well ahead of revenue create funding risk, and high compute needs for training and on-robot inference add cost. Measuring real-world performance, transferring models across robot bodies, finding the right mix of simulated, video, and real robot data, and handling injury risk and legal liability remain key challenges.
General-purpose humanoids and mobile manipulators are the largest opportunity, helped by cheaper hardware such as Unitree's G1, which the company introduced in 2024 with a base price near USD 16,000, according to Unitree. Licensing models to robot makers is another, since most manufacturers cannot afford to train their own models, and robot data infrastructure, covering teleoperation, human video capture, simulation, and synthetic data, is drawing strong investment. Key trends include dual-system designs that pair slow reasoning with fast motor control, open-weight robot models such as GR00T N1 and π0, and models built to run directly on the robot, such as Gemini Robotics On-Device.
By model type, the report covers vision-language-action models, embodied reasoning models, world models, and locomotion & whole-body control models. VLA models are expected to hold the largest share in 2026, as they serve as the core control layer that turns perception and instructions into manipulation, and as leading products such as GR00T N1, Gemini Robotics 1.5, π0, and Figure's Helix are built on this design. World models are expected to grow at the highest CAGR, because they can generate training data, support simulation, and help robots plan by predicting how scenes will change, which addresses the shortage of real robot data. Embodied reasoning models are becoming widely available, with Gemini Robotics-ER 1.5 offered to developers through the Gemini API.
By offering, the study examines models (model licensing & APIs and embedded model software), platforms & tools (fine-tuning & post-training platforms, simulation platforms, and evaluation & fleet learning tools), robot data (teleoperation data, human video data, and synthetic data), and services. Models are expected to account for the largest share in 2026, reflecting licensing and API fees and the value of model software shipped inside robots. Robot data is expected to grow at the highest CAGR, since data scarcity is the main limit on model quality and developers such as Figure and NVIDIA are spending heavily on teleoperation, human video, and synthetic data.
The robot type analysis spans humanoids, mobile manipulators, industrial arms, autonomous mobile robots, quadrupeds, and others. Industrial arms and mobile manipulators together are expected to account for the largest share in 2026, as most current commercial deployments of learned manipulation are in factories and warehouses, where these robots are already installed in large numbers. Humanoids are expected to grow at the highest CAGR, supported by heavy investment in companies such as Figure, Tesla, Agility Robotics, and Apptronik, falling hardware prices, and foundation models designed specifically for humanoid bodies.
By deployment, the report covers cloud, on-robot/edge, and hybrid models. Hybrid deployment is expected to hold the largest share in 2026, as developers run heavy reasoning in the cloud while keeping fast motor control on the robot, matching the dual-system designs now common in the field. On-robot/edge deployment is expected to grow at the highest CAGR, as on-device models reduce latency, keep robots working without network access, and address privacy concerns in homes and workplaces.
The application analysis includes logistics & warehousing, manufacturing & assembly, household & domestic, healthcare & eldercare, retail, agriculture, construction & inspection, and research. Logistics & warehousing is expected to account for the largest share in 2026, since picking and handling a wide range of items is well suited to learned manipulation and e-commerce companies are among the earliest adopters. Household & domestic use is expected to grow at the highest CAGR, as humanoid makers target homes and models that work in new environments become more capable, although safety certification is likely to slow early home deployment. Manufacturing and healthcare and eldercare are further important uses.
By end user, the report covers robot OEMs, logistics & e-commerce companies, automotive & manufacturing companies, healthcare providers, consumers, and research institutions. Robot OEMs are expected to be the largest end user group in 2026, as they license or post-train foundation models to add general-purpose capabilities to their robots and pay for the model software embedded in units they sell. Logistics and e-commerce companies and automotive and manufacturing firms are major direct buyers, judging models on measured pick rates and uptime in their own operations, while consumer demand depends on the arrival of home robots.
North America is expected to remain the largest regional market in 2026 because the United States hosts most leading developers, including NVIDIA, Figure, San Francisco-based Physical Intelligence, Pittsburgh-based Skild AI, FieldAI, Tesla, Agility Robotics, Apptronik, and Toyota Research Institute, and its logistics and e-commerce companies are among the first large users of learned robot manipulation. Canada adds Sanctuary AI and strong robotics research institutions.
Asia-Pacific is expected to grow at the highest CAGR during the forecast period, led by China, where about 295,000 industrial robots went into service in 2024 according to the International Federation of Robotics, and where humanoid companies including Unitree, AgiBot, UBTECH, and Fourier are scaling quickly. South Korea has the highest industrial robot density in the world, Japan is investing in robots for factories and eldercare, and Singapore, Australia, and India are expanding automation. Europe holds a significant share, shaped by the EU Machinery Regulation and by Google DeepMind in London, Norway's 1X Technologies, and Germany's Neura Robotics. In Latin America, Mexico's automotive and electronics plants and nearshoring support demand, while in the Middle East & Africa, Israel's Mentee Robotics, acquired by Mobileye in early 2026, and investment by the UAE and Saudi Arabia point to emerging opportunities.
The market is young and crowded, with big technology firms, heavily funded start-ups, humanoid builders, industrial robot makers, and Chinese robotics firms all competing. Buyers weigh how general and reliable a model is, its data and compute backing, hardware ties, openness, and results in live deployments. Figure and Tesla pair in-house models with their own robots, whereas Physical Intelligence, Skild AI, NVIDIA, and Google DeepMind aim to supply models for many robot brands. Recent moves include the open release of GR00T N1 in March 2025, Google DeepMind's launch of Gemini Robotics 1.5 in September 2025, Mobileye's acquisition of Mentee Robotics, and FieldAI's 2025 funding round at a USD 2 billion valuation, according to the company.
Key companies profiled in the report include NVIDIA Corporation (U.S.), Google DeepMind (U.K./U.S.), Physical Intelligence, Inc. (U.S.), Skild AI, Inc. (U.S.), FieldAI (U.S.), Toyota Research Institute (U.S.), Covariant (Amazon) (U.S.), Figure AI, Inc. (U.S.), Tesla, Inc. (U.S.), Agility Robotics, Inc. (U.S.), Apptronik, Inc. (U.S.), 1X Technologies AS (Norway), Neura Robotics GmbH (Germany), Sanctuary AI (Canada), Mentee Robotics (Mobileye) (Israel), Unitree Robotics (China), AgiBot (China), UBTECH Robotics Corp. (China), and Fourier Intelligence (China).
The study covers broad AI models that give robots transferable perception, reasoning, and motor skills, plus the tools, datasets, and services that tailor these models to specific machines and put them into use. Model types include vision-language-action (VLA) models that convert camera input and instructions into motion, embodied reasoning models that plan multi-step tasks, world models that predict how scenes change, and models that control walking and full-body movement. Revenue is counted from licenses and APIs, paid support for open models, development platforms, training data, integration work, and the share of robot value attributable to embedded model software. Robot hardware, AI chips, and traditional task-specific programming tools are outside the scope.
According to company announcements, capital is flowing into the field at a record pace: Figure's September 2025 Series C brought in more than USD 1 billion and valued the company at USD 39 billion after the round, Physical Intelligence closed a USD 600 million round in November 2025 that valued it at USD 5.6 billion, and Skild AI secured nearly USD 1.4 billion in January 2026 at a valuation of more than USD 14 billion. Underlying automation demand is large: according to the International Federation of Robotics, 542,000 industrial robots were installed globally in 2024, and China took more than half.
Market Dynamics
Rapid progress in VLA models is the main driver, since a single model can now handle many tasks and robots instead of each task being programmed by hand. NVIDIA released Isaac GR00T N1 as an open foundation model for humanoids in March 2025, Google DeepMind launched Gemini Robotics 1.5, which can reason before acting and carry skills from one robot to another, in September 2025, and Physical Intelligence released π0.5 to work in places it has never seen. Large funding rounds finance the compute, data gathering, and engineering talent behind these models. Hiring difficulties in warehouses, factories, and care homes, along with aging populations in countries such as Japan, are steering buyers toward robots that cope with messy, changing work.Training data is the biggest constraint, as the 2023 Open X-Embodiment project, a joint effort of 21 institutions, gathered roughly one million real trajectories across 22 kinds of robots, compared with about 300 trillion tokens of public text for language models according to Epoch AI. Safety rules also slow deployment: the EU Machinery Regulation (EU) 2023/1230, which applies from 20 January 2027, sets requirements for machinery with self-evolving behavior based on machine learning, and standards for learned robot policies in unstructured settings are still immature. Valuations well ahead of revenue create funding risk, and high compute needs for training and on-robot inference add cost. Measuring real-world performance, transferring models across robot bodies, finding the right mix of simulated, video, and real robot data, and handling injury risk and legal liability remain key challenges.
General-purpose humanoids and mobile manipulators are the largest opportunity, helped by cheaper hardware such as Unitree's G1, which the company introduced in 2024 with a base price near USD 16,000, according to Unitree. Licensing models to robot makers is another, since most manufacturers cannot afford to train their own models, and robot data infrastructure, covering teleoperation, human video capture, simulation, and synthetic data, is drawing strong investment. Key trends include dual-system designs that pair slow reasoning with fast motor control, open-weight robot models such as GR00T N1 and π0, and models built to run directly on the robot, such as Gemini Robotics On-Device.
Segment Analysis
The report analyzes the market by model type, offering, robot type, deployment, application, end user, and geography, helping stakeholders identify where early revenue is forming and where growth will come from.By model type, the report covers vision-language-action models, embodied reasoning models, world models, and locomotion & whole-body control models. VLA models are expected to hold the largest share in 2026, as they serve as the core control layer that turns perception and instructions into manipulation, and as leading products such as GR00T N1, Gemini Robotics 1.5, π0, and Figure's Helix are built on this design. World models are expected to grow at the highest CAGR, because they can generate training data, support simulation, and help robots plan by predicting how scenes will change, which addresses the shortage of real robot data. Embodied reasoning models are becoming widely available, with Gemini Robotics-ER 1.5 offered to developers through the Gemini API.
By offering, the study examines models (model licensing & APIs and embedded model software), platforms & tools (fine-tuning & post-training platforms, simulation platforms, and evaluation & fleet learning tools), robot data (teleoperation data, human video data, and synthetic data), and services. Models are expected to account for the largest share in 2026, reflecting licensing and API fees and the value of model software shipped inside robots. Robot data is expected to grow at the highest CAGR, since data scarcity is the main limit on model quality and developers such as Figure and NVIDIA are spending heavily on teleoperation, human video, and synthetic data.
The robot type analysis spans humanoids, mobile manipulators, industrial arms, autonomous mobile robots, quadrupeds, and others. Industrial arms and mobile manipulators together are expected to account for the largest share in 2026, as most current commercial deployments of learned manipulation are in factories and warehouses, where these robots are already installed in large numbers. Humanoids are expected to grow at the highest CAGR, supported by heavy investment in companies such as Figure, Tesla, Agility Robotics, and Apptronik, falling hardware prices, and foundation models designed specifically for humanoid bodies.
By deployment, the report covers cloud, on-robot/edge, and hybrid models. Hybrid deployment is expected to hold the largest share in 2026, as developers run heavy reasoning in the cloud while keeping fast motor control on the robot, matching the dual-system designs now common in the field. On-robot/edge deployment is expected to grow at the highest CAGR, as on-device models reduce latency, keep robots working without network access, and address privacy concerns in homes and workplaces.
The application analysis includes logistics & warehousing, manufacturing & assembly, household & domestic, healthcare & eldercare, retail, agriculture, construction & inspection, and research. Logistics & warehousing is expected to account for the largest share in 2026, since picking and handling a wide range of items is well suited to learned manipulation and e-commerce companies are among the earliest adopters. Household & domestic use is expected to grow at the highest CAGR, as humanoid makers target homes and models that work in new environments become more capable, although safety certification is likely to slow early home deployment. Manufacturing and healthcare and eldercare are further important uses.
By end user, the report covers robot OEMs, logistics & e-commerce companies, automotive & manufacturing companies, healthcare providers, consumers, and research institutions. Robot OEMs are expected to be the largest end user group in 2026, as they license or post-train foundation models to add general-purpose capabilities to their robots and pay for the model software embedded in units they sell. Logistics and e-commerce companies and automotive and manufacturing firms are major direct buyers, judging models on measured pick rates and uptime in their own operations, while consumer demand depends on the arrival of home robots.
Regional Analysis
The report covers North America, Europe, Asia-Pacific, Latin America, and the Middle East & Africa, with analysis of 19 countries, Norway and the Nordic countries, and five rest-of-region groups. The regional analysis considers AI and robotics investment, the presence of model developers and robot makers, industrial robot adoption, labor market conditions, and safety regulation.North America is expected to remain the largest regional market in 2026 because the United States hosts most leading developers, including NVIDIA, Figure, San Francisco-based Physical Intelligence, Pittsburgh-based Skild AI, FieldAI, Tesla, Agility Robotics, Apptronik, and Toyota Research Institute, and its logistics and e-commerce companies are among the first large users of learned robot manipulation. Canada adds Sanctuary AI and strong robotics research institutions.
Asia-Pacific is expected to grow at the highest CAGR during the forecast period, led by China, where about 295,000 industrial robots went into service in 2024 according to the International Federation of Robotics, and where humanoid companies including Unitree, AgiBot, UBTECH, and Fourier are scaling quickly. South Korea has the highest industrial robot density in the world, Japan is investing in robots for factories and eldercare, and Singapore, Australia, and India are expanding automation. Europe holds a significant share, shaped by the EU Machinery Regulation and by Google DeepMind in London, Norway's 1X Technologies, and Germany's Neura Robotics. In Latin America, Mexico's automotive and electronics plants and nearshoring support demand, while in the Middle East & Africa, Israel's Mentee Robotics, acquired by Mobileye in early 2026, and investment by the UAE and Saudi Arabia point to emerging opportunities.
Competitive Landscape
The report reviews key growth strategies, competitive benchmarking, the competitive dashboard (industry leaders, market differentiators, vanguards, and emerging companies), and 2025 market share/rank analysis, with profiles of AI platform and foundation model developers, humanoid and general-purpose robot developers, and Asia-Pacific robotics companies covering business overview, financials, product portfolio, strategic developments, and SWOT analysis.The market is young and crowded, with big technology firms, heavily funded start-ups, humanoid builders, industrial robot makers, and Chinese robotics firms all competing. Buyers weigh how general and reliable a model is, its data and compute backing, hardware ties, openness, and results in live deployments. Figure and Tesla pair in-house models with their own robots, whereas Physical Intelligence, Skild AI, NVIDIA, and Google DeepMind aim to supply models for many robot brands. Recent moves include the open release of GR00T N1 in March 2025, Google DeepMind's launch of Gemini Robotics 1.5 in September 2025, Mobileye's acquisition of Mentee Robotics, and FieldAI's 2025 funding round at a USD 2 billion valuation, according to the company.
Key companies profiled in the report include NVIDIA Corporation (U.S.), Google DeepMind (U.K./U.S.), Physical Intelligence, Inc. (U.S.), Skild AI, Inc. (U.S.), FieldAI (U.S.), Toyota Research Institute (U.S.), Covariant (Amazon) (U.S.), Figure AI, Inc. (U.S.), Tesla, Inc. (U.S.), Agility Robotics, Inc. (U.S.), Apptronik, Inc. (U.S.), 1X Technologies AS (Norway), Neura Robotics GmbH (Germany), Sanctuary AI (Canada), Mentee Robotics (Mobileye) (Israel), Unitree Robotics (China), AgiBot (China), UBTECH Robotics Corp. (China), and Fourier Intelligence (China).
How This Report Helps
- Provides reliable market size estimates and long-term forecasts for the global robot foundation models market through 2036.
- Explains how vision-language-action models, world models, and record physical AI investment are changing robot capabilities.
- Assesses the impact of robot data scarcity, safety certification, high valuations, and compute needs on adoption.
- Identifies high-growth opportunities across model type, offering, robot type, deployment, application, end user, and region.
- Tracks developments such as dual-system architectures, open-weight models, on-device inference, and synthetic and human video data.
- Benchmarks model developers and robot makers on technology, data strategy, partnerships, funding, and market position.
- Supports product strategy, licensing, partnership, and investment decisions for AI companies, robot manufacturers, end users, and investors.
Key Questions Answered
- How large is the global robot foundation models market today, and how will it grow through 2036?
- How are vision-language-action and world models changing what robots can do?
- What are the main drivers, restraints, opportunities, challenges, and trends shaping the market?
- How are developers addressing the shortage of real robot training data?
- Which model type, offering, robot type, deployment, application, and end user segments will grow the fastest?
- How will the EU Machinery Regulation and other safety rules affect deployment of AI-controlled robots?
- Which regions and countries offer the most attractive opportunities?
- Who are the leading companies in the market, and how do vertically integrated and platform strategies compare?
- What funding rounds, model releases, and acquisitions are shaping competition?
Table of Contents
1. Introduction
2. Research Methodology
4. Market Overview
5. Robot Foundation Models Market, by Model Type
6. Robot Foundation Models Market, by Offering
7. Robot Foundation Models Market, by Robot Type
8. Robot Foundation Models Market, by Deployment
9. Robot Foundation Models Market, by Application
10. Robot Foundation Models Market, by End User
11. Robot Foundation Models Market, by Geography
12. Competitive Landscape
13. Company Profiles
14. Appendix
LIST OF TABLES
LIST OF FIGURES
Companies Mentioned
- NVIDIA Corporation
- Google DeepMind
- Physical Intelligence, Inc.
- Skild AI, Inc.
- FieldAI
- Toyota Research Institute
- Covariant
- Figure AI, Inc.
- Tesla, Inc.
- Agility Robotics, Inc.
- Apptronik, Inc.
- 1X Technologies AS
- Neura Robotics GmbH
- Sanctuary AI
- Mentee Robotics
- Unitree Robotics
- AgiBot
- UBTECH Robotics Corp.
- Fourier Intelligence
