Physical AI: The hidden value pools leaders are missing

AI is fundamentally shifting as machines move from following instructions to sensing, deciding, and acting in the physical world. Humanoid robots are the most visible manifestation of this trend, but they are expected to account for only about 2 percent of the physical AI sector’s value by 2045. Where will the real value come from?

In a recent McKinsey Live event, McKinsey Partner Ani Kelkar and Senior Partner Mark Patel discussed the economic potential of physical AI and outlined how leaders can capture the opportunity.

The importance of the physical economy

The digital economy gets a lot of attention, but much of today’s economy still depends on moving, making, inspecting, maintaining, caring, cleaning, transporting, and operating physical things. In fact, about 44 percent of paid work hours are still tied to physical work across geographies and sectors. Those jobs will likely be reinvented over time as physical AI extends intelligence into places where work is variable, embodied, and harder to codify.

A fundamental shift in function

Traditional robots perform exceptionally well at repetitive tasks in controlled environments but less well when conditions get unpredictable. Now, machines are starting to acquire the ability to sense what is happening around them, interpret that information, determine what needs to be done, and then act in the physical world—for example, by navigating complex environments with crowds of people or inspecting construction sites.

Altogether, the broader physical AI ecosystem could unlock $1 trillion or more in economic value by 2035, with general-purpose robotics scaling to $370 billion by 2040.

Getting from here to there

What will it take to move from impressive demonstrations of physical AI to dependable autonomy at meaningful scale?

The journey to reliable performance takes place in four stages: capability in a demo, reliability in a controlled environment, dependable autonomy in production, and scalable economics across many sites. It will require real-world deployment to provide data on where things break, where humans need to intervene, and what has to be redesigned. Each deployment will provide additional data, creating a learning flywheel and a more diverse set of participants in the value chain.

Beyond productivity gains

When it comes to physical AI, attention has generally been focused on the manual work it can replace. But that’s just part of the story. Real value will happen where physical AI creates not just productivity gains but also higher quality and higher throughput. Achieving this will mean fundamentally redesigning workflows and operating models.

McKinsey analysis suggests that the software and model layers of physical AI alone could represent $330 billion by 2036. The task now is to determine where it can be deployed safely and reliably at scale to create EBIT improvements.

How to capture long-term value

To tap into the immense potential of physical AI, organizations will need to redesign and continually improve their workflows based on deployment experience. Three actions are particularly critical:

1. Focus on the task and the economic problem, not the robot.
2. Build the organizational and data infrastructure to deploy, integrate, and learn from physical AI.
3. Decide where in the value chain you want to play.

Although most of today’s momentum focuses on improving productivity, the opportunities to create disproportionate value extend beyond those gains. Organizations with the imagination to create new operating models will take the lead in physical AI.