The physical AI models market map: Behind the arms race to control robot intelligence

In 2025, the robotics sector experienced a significant surge in funding, raising a record $40.7 billion—an impressive 74% year-over-year (YoY) increase, which equates to 9% of all venture funding. This growth propelled the sector into the spotlight alongside AI software, marking a new era of innovation in physical AI. Physical AI, the backbone of this progress, enables robots to operate in the physical world and learn from real-world data rather than relying on pre-programmed rules. The race to control robot intelligence is now heating up, with companies vying for dominance in the physical AI models market.

The foundation of this arms race lies in building foundation models for robots to learn across a wide range of tasks. Unlike language models trained on text, physical AI models require extensive real-world robot data to function effectively. This need for data creates an early window for companies to establish a foothold in the market and secure a competitive advantage.

To better understand the landscape, CB Insights Research has created a comprehensive market map that identifies over 70 companies across 10 physical AI model categories. These companies have been organized into distinct groups based on their focus areas, providing valuable insights into the current state of the industry.

One of the key groups is "Data & Simulation," which includes companies specializing in synthetic data generation, real-world robot demonstrations, and simulation platforms. These platforms allow robots to learn in virtual environments, enabling them to adapt to real-world scenarios more effectively. By generating synthetic data, these companies help reduce the time and cost required for training robots, accelerating the development of physical AI systems.

Another significant group is "Model Approaches," which focuses on core AI architectures that enable robots to understand their environment, generate actions, and predict outcomes. These architectures are crucial for enabling robots to make informed decisions and adapt to changing conditions. Companies in this group are working on developing vision language models (VLMs) that allow robots to interpret visual information, vision language action models (VLAs) that generate appropriate actions, and world models that predict the consequences of these actions.

The "Foundation Models" group represents pre-trained robot intelligence that combines various model architectures for general manipulation, autonomous driving, and multi-robot coordination. These models serve as a starting point for robots to learn and adapt to new tasks, significantly reducing the time and resources needed for training. The development of foundation models is a critical area of focus, as they hold the potential to revolutionize the way robots interact with the physical world.

The "Observability" group consists of platforms that monitor deployed robots and feed real-world data back into training to improve performance. This continuous learning loop is essential for enhancing the capabilities of physical AI systems and ensuring they can adapt to new challenges. By leveraging real-world data, these platforms help companies refine their robots' decision-making processes and optimize their performance.

CB Insights customers can explore these companies through a dedicated platform, sorted by their Mosaic health score. The Mosaic score is a predictive metric that evaluates the health and success likelihood of private companies. In markets with five or fewer companies, all companies were included to provide a comprehensive view of the space. It is important to note that this market map is not exhaustive, as the physical AI models market is rapidly evolving, and new players are likely to emerge.

One of the key takeaways from this market map is the critical role of proprietary training data in the success of physical AI models. Companies that can generate and leverage high-quality, real-world data are well-positioned to lead the market. Additionally, the ability to adapt and refine models based on continuous learning and feedback is becoming increasingly important.

In conclusion, the physical AI models market is experiencing a dynamic transformation, driven by the need for robots to operate in the physical world and learn from real-world data. The arms race to control robot intelligence is intensifying, with companies vying for dominance in various categories. As the industry continues to evolve, the focus on data, model architectures, and continuous learning will be crucial in shaping the future of physical AI and robotics.