Capgemini report finds physical AI adoption accelerating, but scaling challenges persist

Capgemini Research Institute has released a new report examining the rise of physical AI in robotics, pointing to growing enterprise adoption alongside persistent barriers to large-scale deployment.

The report defines physical AI as the next stage of artificial intelligence, where systems move beyond digital environments to perceive, reason, and act in the physical world. In robotics, this marks a shift from pre-programmed automation to machines capable of operating autonomously in dynamic, real-world settings.

Adoption moves from experimentation to early deployment

Enterprise engagement with physical AI is already widespread. According to the report, 79% of organisations are exploring or implementing physical AI, with 27% actively deploying or scaling solutions.

Momentum is driven by both technological and structural factors. Advances in foundation models, simulation, and edge computing are improving robot intelligence and shortening development cycles. At the same time, labour shortages and rising costs are pushing organisations to automate more complex physical tasks.

Use cases are expanding beyond controlled environments into areas previously difficult to automate. These include logistics operations such as pick-and-place and order fulfilment, as well as sector-specific applications in manufacturing, healthcare, and insurance.

The report notes that 60% of executives believe physical AI will enable robotics adoption in scenarios that were previously impractical.

Productivity gains and operational resilience drive interest

Executives expect physical AI to deliver improvements across productivity, efficiency, and quality, alongside greater operational flexibility. The ability to reconfigure workflows quickly is seen as a key advantage over traditional automation systems.

Workforce constraints remain a central driver. Labour shortages, particularly in sectors such as agriculture, logistics, and retail, are cited as the primary catalyst for investment, outweighing labour cost considerations.

Beyond operational gains, physical AI is also linked to broader industrial policy objectives. Around 43% of executives see it as a way to support domestic production at scale, aligning with reindustrialisation efforts in regions such as the US and Europe.

Scaling remains constrained by technology and readiness gaps

Despite strong interest, large-scale deployment remains limited. Fewer than 5% of organisations report operating physical AI systems at scale, and nearly eight in ten executives cite challenges in moving beyond pilot stages.

Key barriers include system reliability, limited dexterity, and the availability of real-world training data. Safety and cybersecurity concerns also complicate deployment, particularly in environments where robots interact closely with people.

Operational readiness is another constraint. Scaling requires new approaches to system integration, governance, and workforce training, areas where many organisations are still early in development.

Humanoid robots remain a longer-term prospect

While interest in humanoid robots is high, the report positions them as a longer-term investment rather than a near-term deployment priority.

More than 70% of executives cite technical immaturity as a barrier, alongside high costs and unclear return on investment. Public acceptance is also flagged as a potential obstacle, with more than 60% of respondents identifying it as a concern.

In the near term, growth is expected to come from established robot form factors such as autonomous mobile robots and industrial robotic arms, which are already proving viable in specific applications.

Physical AI enters a transitional phase

The report concludes that physical AI has reached an inflection point, with technological progress and market demand converging to enable broader adoption.

However, the transition from pilot to scale remains uneven. Organisations are moving beyond experimentation, but sustained deployment will depend on improvements in reliability, safety, and operational integration, rather than further advances in model capability alone.