UBS says 'physical AI' is here, and shares 3 picks-and-shovels sectors of the market to play the trade.

When you think of AI, you likely think of the chatbot on the phone in your pocket, beaming information from far off data centers that power large language models like ChatGPT or Claude.

But AI is striving to become more tangible — humanoid, in fact.

Ulrike Hoffmann-Burchardi, the global head of equities at UBS, says that humanoid robots are increasingly able to use "in-context" learning, where they learn how to do something by observing an action or task.

In a recent note to clients, Hoffmann-Burchardi said that "we are now at the precipice of moving away from preprogrammed automation and stepping into genuine physical autonomy."

The theme presents a new frontier of AI investment, she said, highlighting three picks-and-shovels sectors of the AI robotics trade to play.

1. Actuators and precision gears

Actuators drive movement in a robot, and gears facilitate that movement. These parts will be crucial to humanoid robots' agility.

"These are highly specialized components that allow a robot to move its fingers with millimeter precision or balance its weight on an uneven floor," Hoffmann-Burchardi wrote in the note. "Manufacturing capacity is heavily concentrated in just a few specialized firms worldwide."

While Hoffmann-Burchardi didn't name any particular companies, examples of stocks in this space include Rockwell Automation (ROK) and ABB (ABBNY).

2. The sensor and perception layer

Sensors are what will allow robots to take in data in real time.

This can come in the form of cameras, so a robot can see and copy an action. But it also means a robot's "skin," which allows it to be able to feel and measure.

"For a robot to pick up a fragile egg or a heavy iron wrench without breaking or dropping it, it needs a type of skin that can measure pressure and friction instantly," Hoffmann-Burchardi said.

Some examples of firms in the space include Synaptics Incorporated (SYNA), Ouster (OUST), and Luminar Technologies (LAZR).

3. Edge computing and power infrastructure

These robots will need computing power to process information, as well as power sources that can fuel the robots for long periods of time.

Robotic-specific hardware will be needed, Hoffmann-Burchardi said.

Examples include Nvidia (NVDA), Arm Holdings (ARM), and Advanced Micro Devices (AMD).

Your guide to what's moving markets