Caterpillar is wiring up its yellow iron with Nvidia silicon.
The heavy equipment giant has tapped Nvidia to power a new AI assistant, “Cat AI,” that’s being piloted on its mid-size Cat 306 CR Mini Excavator and demoed at CES 2026.
Cat AI runs on Nvidia’s Jetson Thor platform, part of the chipmaker’s push into what it calls “physical AI” — the blend of large-scale models, simulation and edge compute that lives inside real-world machines instead of just in the cloud.
What Cat AI actually does
This isn’t full autonomy for construction sites. Cat AI is more like a smart copilot for human operators.
Brandon Hootman, vice president of data and AI at Caterpillar, told TechCrunch the system is built on a fleet of AI agents that sit in the cab with the operator and:
- Answer questions from the operator
- Surface manuals and other resources on demand
- Offer safety tips in context
- Help schedule service and maintenance
“Our customers don’t live in front of a laptop day in and day out; they live in the dirt,” Hootman said. “The ability to get the insights and take the action that they need while they’re doing the work is very important to them.”
In other words, Caterpillar isn’t trying to drag operators into a dashboard; it’s pushing the dashboard into the machine.
Data firehose meets digital twins
The bigger play is the data flowing through all of this.
Hootman said Caterpillar’s machines already send around 2,000 messages back to the company every second. That telemetry is now feeding a new layer of simulation.
Caterpillar is piloting digital twins of construction sites using Nvidia’s Omniverse simulation tools. Those twins let the company and its customers:
- Test different scheduling scenarios before work starts
- Better estimate how much material a project will actually need
- Iterate on site plans in software instead of on the dirt
The idea: fewer surprises, tighter bids, less wasted material.
A step beyond autonomous mining trucks
Caterpillar isn’t new to automation. The company already runs fully autonomous vehicles in the mining sector, where controlled environments and repeatable routes make self-driving more tractable.
Construction sites are messier and more variable, so Cat AI is a next step rather than a leap to driverless excavators.
“The reason that we started here was it was a real challenge of our customers today that needed to be addressed, and also something that we had some real momentum on and we felt like we could bring to market pretty quickly,” Hootman said. “What we also liked is that it provided a kind of a technology foundation for us to then build upon.”
That foundation spans the operator in the cab, the data streaming off the machine, and the digital twin running in the cloud.
Nvidia’s ‘physical AI’ play
For Nvidia, Caterpillar is another proof point that its AI stack isn’t just for web companies and data centers.
Bill Dally, Nvidia’s chief scientist, told TechCrunch in 2025 that physical AI is the company’s next big frontier. At CES this year, Nvidia detailed a full‑stack ecosystem for that vision, including:
- Open AI models such as its Cosmos model family
- Simulation libraries like Omniverse
- Hardware and developer kits to deploy models at the edge
Deepu Talla, Nvidia’s vice president of robotics and edge AI, argues that physical AI goes far beyond traditional robotics labs.
“Physical AI is the next wave of AI,” Talla said. “Nvidia is pioneering that with computers that train the models, that do the simulation to test the models and deploy the models into the robots, whether [that’s] an autonomous car or a Caterpillar machine.”
In that framing, a mini excavator with Cat AI is just another robot — one that happens to move dirt instead of packages.
Why this matters
Put the pieces together and Caterpillar’s partnership with Nvidia looks less like a flashy CES demo and more like an on‑ramp:
- From connected machines to continuously learning fleets
- From static plans to simulated, optimized job sites
- From manual workflows to assistive, eventually more automated operations
For construction, an industry that still runs on paper plans and phone calls on many sites, that’s a big shift. For Nvidia, it’s one more foothold in the physical world its chips are rapidly colonizing.
