Humanoid robots are now operating inside BMW factories. Tesla Optimus is working on Tesla’s own production floors. Amazon, Mercedes-Benz, and several of the world’s largest manufacturers are actively running Physical AI pilots.

Yet the most important number in Physical AI today may not be how many robots have been demonstrated.

It may be how few have actually been deployed at scale.

That gap between demonstration and dependable deployment is quickly becoming the defining challenge of the industry.

The reason is simple. The physical world is far harder than the digital one.

A Physical AI system has to perceive its environment, understand what is changing around it, decide how to respond, and execute that action in real time. Every object is different. Every environment is different. Every edge case matters.

Which is why the biggest bottlenecks in Physical AI today are not capability alone. They are data, simulation, reliability, dexterity, and safety.

And that is also where some of the most interesting startup opportunities are emerging.

In the latest edition of DeepTech Dispatch, we unpack:
• What Physical AI actually is
• Why the industry is advancing so rapidly today
• The deployment challenge holding back large-scale adoption
• Why India could become one of the most important proving grounds for Physical AI globally
• Where we see the next generation of startup opportunities emerging

The challenge is no longer proving that Physical AI can work. It is proving that it can work reliably, repeatedly, and economically in the real world.