Executive Summary

While the world has focused its attention for the last three years on generative artificial intelligence, chatbots, and new model releases coming from frontier AI labs, a quieter revolution is taking place that many believe represents the next stage in AI development: the arrival of Physical AI. Like the iPhone’s introduction in 2007, AlexNet’s victory in the 2012 ImageNet competition, and ChatGPT’s release in 2022, analysts and industry representatives believe a similar breakthrough is imminent.

Physical AI “lets autonomous systems like robots, self-driving cars, and smart spaces perceive, understand, and perform complex actions in the real (physical) world.”¹ NVIDIA has declared “in the near future, everything that moves, or that monitors things that move, will be autonomous robotic systems.”² OpenAI reportedly re-opened its robotics division in early 2025 to capitalize on the convergence of AI and robotics, while startups from Shanghai to Silicon Valley building the “brains” of robots are raising hundreds of millions of dollars.³ Electric vehicle makers Tesla and XPeng are racing to develop humanoid robots of their own.⁴ Meanwhile Amazon, which reports having one million robots in operation today, believes “Physical AI is about to change everything for robotics [including] autonomy, manipulation, sortation, and computer vision.”⁵ Adding to this enthusiasm, analysts at Morgan Stanley assert the market for humanoid robots will grow from tens of millions of dollars today to reach $5 trillion by 2050.⁶

Yet the convergence of AI and robotics is so new that the field lacks a shared name, to say nothing of a mature technology stack. Some companies call this convergence “embodied AI” while others prefer “physical AI,” “embodied machine intelligence,” or “generative physical AI.”⁷ It is not at all clear if the hype around AI progress can translate into robots finding their way through the physical world: autonomous three-dimensional navigation of dynamic environments requires a mature software, hardware, and data ecosystem that simply does not exist at scale today. NVIDIA states part of the problem plainly: “Large language models are one-dimensional, able to predict the next token, in modes like letters or words. Image- and video-generation models are two-dimensional, able to predict the next pixel. None of these models can understand or interpret the 3D world.”⁸

The primary challenges facing Physical AI are the same ones that have troubled the robotics industry for generations: technology barriers and economic barriers. Parts of the robotics supply chain remain in their industrial infancy, key hardware technology breakthroughs remain elusive, and even recent advances are not ready for scalable manufacturing. Batteries, motors, sensors, and actuators evolve far more slowly than algorithms and software, and scalable manufacturing requires large amounts of patient capital. In addition, much of the supply chain for robotics components is commoditized, and the relatively slim margins dissuade innovative startups from competing with established incumbents. Adding to these challenges, each robotics company is pursuing its own unique approach, meaning the supply chain of components and parts remains largely non-standardized, hampering scalability and adding cost. The gap between impressive demonstrations in controlled environments and the promise of millions of affordable robots acting independently as they navigate the world is enormous.

The focus of this paper is on characterizing the convergence of Physical AI and robotics, its underlying supply chain, and identifying competitive advantages as well as constraints. This paper provides background on the technology and describes the ecosystem and supply chain of hardware and software suppliers supporting the technology. It then characterizes competitiveness worldwide using bibliometrics, patents, investment data, and industry reports to determine firm leadership, constraints, and breakthroughs across the technology ecosystem from AI foundation models and software to hardware component and robot manufacturers as well as end users. It concludes with a summary of drivers and positive trends, as well as constraints and limiting trends with an eye towards opportunities policymakers interested in promoting the tech industry’s next breakthrough moment can consider.

This paper builds on previous CSET research looking at the robotics patent landscape to characterize competitiveness using CSET’s Map of Science and separate research that proposed a methodology for identifying and characterizing an emerging technology.⁹ It concludes by introducing a template that could be used by policymakers interested in global competitiveness assessment of other emerging technologies.

1. “What is Physical AI?” NVIDIA, https://www.nvidia.com/en-us/glossary/generative-physical-ai/.
2. Madison Huang, “What is NVIDIA’s Three-Computer Solution for Robotics?” NVIDIA (blog), August 8, 2025, https://blogs.nvidia.com/blog/three-computers-robotics/.
3. Michael J. de la Merced, “Physical Intelligence, a Robot A.I. Specialist, Raises Millions From Bezos,”The New York Times, November 4, 2024, https://www.nytimes.com/2024/11/04/business/dealbook/physical-intelligence-robot-ai.html; Amanda Greenwood, “OpenAI’s secret robot plans revealed,” The AI Report, January 13, 2025, https://www.theaireport.ai/articles/openais-secret-robot-plans-revealed; “Chinese embodied AI startup TARS raises $120m in angel funding,” Tech In Asia, March 25 2025, https://www.techinasia.com/news/chinese-embodied-ai-startup-tars-raises-120m-in-angel-funding.
4. Tom Carter, “Tesla’s Chinese EV competitors are racing to build their own Optimus rivals,” Business Insider, November 2024, https://www.businessinsider.com/tesla-chinese-ev-competitors-build-optimus-rivals-2024-11; Aditi Bharade & Cheryl Teh, “China’s spring festival celebration featured a fleet of dancing robots that flexed the country’s advancements in robotics,” Yahoo!Tech, January 29, 2025, https://tech.yahoo.com/science/articles/chinas-spring-festival-celebration-featured-101336663.html; Jack Ewing, “Elon Musk Shows Off Tesla ‘Robotaxi’ That Drives Itself,” The New York Times, October 10, 2024, https://www.nytimes.com/2024/10/10/business/tesla-robotaxi-elon-musk.html.
5. “Amazon VP shares his approach to the future of robotics innovation,” Amazon, April 10, 2025, https://www.aboutamazon.com/news/operations/amazon-robotics-culture; Scott Dresser, “Amazon launches a new AI foundation model to power its robotic fleet and deploys its 1 millionth robot,” Amazon Robotics, June 30, 2025, https://www.aboutamazon.com/news/operations/amazon-million-robots-ai-foundation-model.
6. Pia Singh, “Morgan Stanley says humanoid robots will be a $5 trillion market by 2050. How to play it,” CNBC, April 29, 2025, https://www.cnbc.com/2025/04/29/how-to-play-a-5-trillion-market-for-humanoid-robots-by-2050.html.
7. The trend that appears to have gotten the most traction in the past year, as measured by Google Search, is “Physical AI.” See: “Search: Interest in Physical AI, Embodied AI, and Related Terms,” Google Trends, https://trends.google.com/trends/explore?geo=US&q=physical ai,embodied ai,embodied machine intelligence,generative physical ai&hl=en-US.
8. Huang, “What is NVIDIA’s Three-Computer Solution?”
9. Sara Abdulla, “China’s Robotics Patent Landscape,” Center for Security and Emerging Technology, August 2021, https://cset.georgetown.edu/publication/chinas-robotics-patent-landscape/.