[Industry Trends] History Repeats Itself – The Rise of Physical AI

Today, I’d like to talk about an important topic in industry trends: Physical AI.

Narrow Focus for Clarity

Whenever I create content around industrial trends, I try not to cover overly broad areas.
Instead, I focus on specific sectors where a deep dive is more helpful. And today, that specific focus is Physical AI.

The Evolution of AI

To talk about Physical AI, we need to understand how AI has evolved over time.

1. LLM (Large Language Model)

We have LLMs like ChatGPT, which can provide impressively accurate answers when you ask a question.

2. Agents

Taking it a step further, agents don’t just respond to queries—they initiate actions.
For example, you could ask which U.S. state has the best weather in August, and the AI could go beyond just naming a state: it could recommend a hotel and even book it for you.
This ability to act and respond has transformed user experiences dramatically.

3. Physical AI

Now we’re entering the realm of robots powered by AI, which we refer to as Physical AI.
These are robots that don’t need constant instructions—they can clean your home, care for children or the elderly, and perform many tasks autonomously.
Major tech companies, including hyperscalers, are investing heavily to establish the standard for this future.

Two Types of Physical AI

Physical AI generally falls into two categories:

Special-Purpose Robots: designed for specific tasks
General-Purpose (Humanoid) Robots: built to perform a wide range of functions

Special-Purpose Robots

These are robots built for specialized roles like:

Autonomous driving
Surgical procedures
Repetitive manufacturing tasks
Crop harvesting
Delivery services

This category is already seeing robust development across industries.

General-Purpose (Humanoid) Robots

These are humanoid robots designed to handle varied tasks.
Ultimately, they’re expected to dominate industrial usage and become the standard robot form.

Market Growth of Robots

The robot market is expanding exponentially, with robots already in use around the world.

One chart from a 2023 study illustrates robot density by country, showing substantial increases over the past 7 years—South Korea is notably high on that list.

For humanoid robots, even in just five years, the market size is expected to grow by at least 17.5% by 2025.
Some institutions even project growth in the upper 20% range.

In short: The Physical AI market is exploding.

Anatomy of a Humanoid Robot

Humanoid robots have two major components:

Software (the brain): the operating system that controls the entire robot
Hardware (the body): includes motors, batteries, cameras, and actuators—which convert electric energy into mechanical motion at each joint.

The operating system (OS) is what makes the robot’s brain function. Actuators, which are increasingly crucial, allow for precise movement.

Key Players in Robot OS Development

Currently, only major tech companies have the technical sophistication to develop robot OS systems.
Some notable examples:

Open Robotics – ROS2
NVIDIA – Isaac
Tesla – FSD-core

Each company takes a different approach:

Open Robotics & NVIDIA: open-source models to promote adoption and become the industry standard
Tesla: uses a closed system, developing both software and hardware in-house for vertical integration

NVIDIA's Vision: A Closer Look

NVIDIA is arguably the most watched player in this space. Here’s how they’re building out their system:

NVIDIA Cosmos

Cosmos is a virtual 3D environment constructed by NVIDIA.
It simulates the real world as if a camera were capturing it.
For example, you could instruct a 1-meter-high robot to “tour a chemical plant” and Cosmos would create a simulated video.

Isaac Lab

Isaac is a virtual training platform that brings Cosmos environments into Omniverse, a broader simulation world.
Instead of physically repeating an action like pouring water thousands of times, robots practice in this physics-based virtual world.

This drastically reduces cost and time.
Once software is trained and refined in simulation, it can be deployed across different physical robots.
Eventually, only selected companies will gain access to advanced datasets, and those that control the OS may define industry standards.

Top 5 Companies Building Robot OS

Here are the five key companies leading the charge:

Boston Dynamics
Agility Robotics
Figure AI
Unitree Robotics
Tesla

NVIDIA’s Isaac OS is already being used experimentally by the first four companies.
Despite being trained in virtual environments, their robots function effectively in the real world.

Tesla, however, does not use Isaac OS.
They have developed both their own software and custom hardware known as Optimus.
In my view, Tesla currently sits at the cutting edge of the field.

History Repeats Itself

Why is all this important? Because history shows us what happens next.

Back in the 1980s and 1990s, as PCs became common, the number of PC hardware manufacturers decreased.
By the 1990s, IBM had emerged as a dominant force, and eventually only Apple and IBM remained major players.

Then came smartphones in the 2000s, and the same trend occurred:

Multiple OS platforms emerged
Only a few survived: Windows, Android, and iOS

In today's Physical AI race, the same pattern is unfolding.
Only a few hardware companies and OS developers will set the standard and dominate the market.

Smartphones and the AI Parallel

In smartphones:

Samsung dominates Android-based devices
Apple controls both hardware and software, enabling total market dominance

Tesla is playing Apple’s game—developing both its OS and hardware in-house.
While NVIDIA is currently leading the OS field, it hasn’t yet become the definitive industry standard.

In the end, a few hardware makers and one or two software companies will define the landscape.
Just like Google and Apple faced antitrust issues, we may see monopolistic concerns rise in Physical AI.

We might even reach a point where robots can download apps for specialized skills.
For instance, installing an app that helps a robot sew more efficiently.

Big companies will aim to become the industry standard, while smaller firms will turn their services into products they can sell.

Given how rapidly robots are becoming mainstream, it's vital to understand these historical parallels from the PC and smartphone eras.

Recommended Fields of Study

If you're considering a career in the Physical AI industry, here are fields of study that align with its components:

Core Robot Hardware