Is a robot brain worth more than its body? Physical AI funding has reached $26.9 billion this year

Grace Gong gave a keynote on physical AI at AWS's ETHIS AI Summit. Physical AI refers to AI systems that can enter the physical world, perceive, and act; robots are a typical embodiment. She cited Jensen Huang: Physical AI already brings Nvidia nearly $10 billion in revenue each year, making it Nvidia's second most important growth category. Someone replied under the tweet that this number came earlier than expected.

Why now? She gave three unlocks.

1. Robot foundation models. One brain, many bodies. SkildAI's model controls quadruped robots, humanoid robots, and robotic arms without needing to know the body form in advance.
2. Simulation. Robots first learn from internet videos and millions of hours of simulation before touching the real world.
3. Cheaper bodies. The EV supply chain has driven down the prices of actuators, batteries, and sensors.

Then look at the money. Data from Harmonic.ai shows about 1,300 venture-backed physical AI companies, excluding general AI labs and defense prime contractors. Funding was $3.8 billion in 2022, $10 billion in 2025, and $26.9 billion in 2026—and the year is not over yet.

Funding is 7x, while deal count is only 1.8x. Money is concentrating in a few winners. In 2026 alone, there were 72 funding rounds of $100 million or more, exceeding the total from 2022 to 2025. 30% of the companies were founded in 2023 or later. This is new company formation, not old robotics companies rebranding.

She listed several companies she pays attention to. Bedrock Robotics was covered last week on Venture with Grace for its $600 million Series C; the company validated demand before building. DynaRobotics' founder recently spoke at Smart AI Summit, and RLWRLD_ai's founder and president attended Venture with Grace CEO club dinners. She believes both are doing interesting work in physical AI.

What are the four layers where value will land?

1. Brain. Skild, physical_int. She calls it the Android proposition for robots. The brain's funding level is now the same as the body's.
2. Body. Figure, Apptronik, Unitree, NEURA. Whoever deploys the most robots collects the most real-world data. Figure is valued at about $39 billion.
3. Picks and shovels. Simulation, teleoperation, data, security, and compute.
4. Vertical deployers. Warehousing, construction, agriculture, data centers, and defense.

She ended with a question: Will the robot brain be worth more than the body? In smartphones, the operating system won; in cars, the body won. Which way will robots go?

The event was organized by Ann from ETHIS and the AWS team.