Robots are learning to handle the physical world in a fundamentally new way, and investors should look at the companies supplying their parts, according to a note from UBS, the Swiss bank.
The note, from the bank's wealth management chief investment office, argues we have entered an era of "physical AI", meaning artificial intelligence that can understand and manipulate real surroundings rather than just process text.
The shift rests on a technique called in-context learning. Older robots had to be painstakingly trained for each task through millions of digital simulations and hard-coded instructions, a process that took weeks.
In-context learning lets a robot watch a human do a job once or twice through a camera, then copy it straight away, without going back for a lengthy retraining cycle.
UBS compared it to giving a chatbot a couple of examples of a writing style, which it then imitates without any change to its code.
For investors, the bank invoked the gold rush maxim of selling shovels rather than digging for gold. It identified three parts of the supply chain that stand to benefit.
The first is precision movement, covering the specialised motors and gears that let a humanoid robot move its fingers accurately or balance on uneven ground, a market dominated by a handful of firms.
The second is sensing, including advanced laser-based depth scanners and "electronic skin" that lets a robot grip a fragile egg or a heavy wrench without crushing or dropping it.
The third is edge computing and power, meaning the low-energy chips that run these models on a moving machine and the advanced batteries that keep it going through long shifts.
UBS said physical AI would create economic value across logistics, manufacturing and healthcare.
The bank concluded that the firms making robots' gears, sensors and chips should grow more valuable over time.
