CES 2026’s surprise theme: machines that listen, watch, decide
“At Nvidia we built three computers and open software stacks for developers to build physical AI, and now Caterpillar is bringing all of this to life,” Deepu Talla said onstage at CES. The line landed because the demo wasn’t a shiny concept car or a floating hologram it was a Cat 306 CR Mini Excavator answering questions in plain language, then moving steel in response.
That moment captures a shift that kept repeating across the show floor: AI stopped being a “cloud feature” and started acting like an interface layer for physical systems. In heavy equipment, that interface has immediate consequences how quickly an operator can set up boundaries, locate functions, and trust what the machine is about to do.
Caterpillar’s pilot, branded Cat AI, runs on Nvidia’s Jetson Thor physical AI platform and leans on speech and on-device inference rather than treating connectivity as a requirement. In the cab, the goal is not autonomous theater; it is an assistant that can translate “how do I get started?” into practical guidance while the machine stays within configured limits. Off the jobsite, Caterpillar’s CEO Joe Creed framed the same idea as a manufacturing strategy, saying the company will “use digital twins before we ever cut steel.” The point is process: simulate the change, then build it once, with fewer surprises.
The show’s other big “listening machine” story arrived in a smaller form factor. Lenovo teased an experimental Motorola wearable under the Project Maxwell codename that continuously captures surroundings and turns them into recommendations and insights. No privacy details came with the pitch, and that absence matters because always-on capture turns ordinary spaces into data collection zones whether people nearby agreed to participate or not.
That discomfort is already familiar in the living room. A Walmart executive at CES explained the retailer’s $2.3 billion Vizio acquisition as a play for ACR: “Automatic Content Recognition data” that can identify what is on screen and connect viewing behavior with shopping behavior. “We were recently acquired by Walmart, so we have the ability to take that viewership data… [and] marry that with Walmart data,” Vizio’s Mike O’Donnell said, outlining an end-to-end ad attribution pipeline that treats the TV as an analytics sensor.
Outside the booth lights, autonomy is getting more physical and more maintainable. Waymo used CES to highlight its next robotaxi vehicle, the Zeekr-built Waymo Ojai, describing a sensor suite designed for real-world grime: heaters, wipers, and fluid to keep cameras and lidars usable when weather stops being polite. Scaling robotaxis is not just software; it is operations, uptime, and hardware built for dirty streets.
Even entertainment’s AI talk tracked the same direction. Arcana CEO Jonathan Yunger said AI is moving toward a blended workflow rather than wholesale replacement: “Right now, you need real actors,” he said, while describing AI as a tool that can compress parts of visual effects work. Different industry, same theme: fewer “AI demos,” more systems that have to hold up under use.
CES still has plenty of spectacle. The most lasting surprises, though, were the products quietly trying to become a new default interface voice, sensing, and assistance across machines that dig, drive, film, and sit in the home.
