Robot Foundation Models Market (2026-2036)

The growth of this market is mainly driven by breakthroughs in vision-language-action models, record investment in physical AI, and labor shortages and automation demand across manufacturing and logistics. However, the scarcity of robot training data, safety certification requirements for AI-controlled machinery, valuations far ahead of commercial revenue, and high compute and on-robot processing requirements restrain the growth of this market.

Furthermore, general-purpose humanoid and mobile manipulation robots, licensing of foundation models to robot manufacturers, and robot data infrastructure are expected to offer growth opportunities for the stakeholders in this market. However, evaluating real-world performance, generalizing across robot embodiments, closing the gap between simulation, human video, and real robot data, and managing physical safety and liability remain major challenges impacting the growth of this market. Additionally, dual-system model architectures, open-weight robot models, and on-device inference are prominent trends in this market.

The Robot Foundation Models Market comprises large, general-purpose AI models trained on diverse data to give robots perception, reasoning, and motor control that transfer across tasks, environments, and robot bodies, together with the platforms, data, and services used to adapt and deploy them. The market covers vision-language-action (VLA) models that turn camera images and language instructions into robot actions; embodied reasoning models that plan multi-step tasks and understand spatial relationships; world models that predict how the physical world evolves; and locomotion and whole-body control models for legged and humanoid robots. Revenue is measured from model licensing and API access, commercial support for open-weight models, fine-tuning, simulation, and evaluation platforms, robot training data collection and generation, and integration services, including the attributable value of foundation model software embedded in robots sold. Robot hardware, general-purpose AI chips, and traditional task-specific robot programming software are excluded. The ecosystem spans AI labs, robot foundation model startups, robot manufacturers, chip and simulation platform providers, data collection companies, and end users in logistics, manufacturing, healthcare, and homes.

Model capabilities advanced rapidly in 2025. NVIDIA released Isaac GR00T N1 on 17 March 2025 as an open foundation model for humanoid robots, a VLA model with a dual-system architecture in which a vision-language module interprets the scene and instructions and a diffusion transformer generates motor actions, trained on real robot trajectories, human videos, and synthetic data and demonstrated on Fourier GR-1 and 1X humanoids. Google DeepMind introduced Gemini Robotics 1.5 and Gemini Robotics-ER 1.5 on 25 September 2025, describing the former as its most capable VLA model, able to think before acting and transfer skills across embodiments, and the latter as a state-of-the-art embodied reasoning model evaluated on 15 academic benchmarks and made available through the Gemini API. Physical Intelligence open-sourced its π0 model and released π0.5, designed to perform tasks in entirely new environments.

Investment has reached unprecedented levels. On 16 September 2025, Figure announced more than USD 1 billion in committed Series C capital at a USD 39 billion post-money valuation, to scale its Helix VLA model, expand its BotQ manufacturing, build NVIDIA GPU infrastructure, and launch large-scale data collection from human video and multimodal sensors. Physical Intelligence raised USD 600 million at a USD 5.6 billion valuation in November 2025, Skild AI announced close to USD 1.4 billion in funding at a valuation above USD 14 billion in January 2026, and FieldAI raised USD 314 million at a USD 2 billion valuation in 2025. At the same time, demand for automation remains strong: the International Federation of Robotics reported 542,000 industrial robot installations worldwide in 2024, with China accounting for more than half.

The market faces fundamental constraints. Unlike language models trained on hundreds of trillions of tokens of text, robot models have limited real-world training data: the Open X-Embodiment collaboration, published in 2023, pooled about one million trajectories from 22 robot embodiments across 21 institutions. Developers pursue different strategies to close the gap, from Physical Intelligence's emphasis on real-world human-in-the-loop data to simulation-heavy training and learning from human video. Regulation is also tightening: the EU Machinery Regulation, applying from 20 January 2027, introduces requirements for machinery with self-evolving behavior and AI-based safety functions.