NVIDIA’s new ‘NemoClaw’ wants to put an AI engineer in your pocket (and your robot)
Jensen Huang’s latest pitch sounds like a sci-fi flick where the robots don’t just take the jobs—they do the math, too. At GTC Taipei, NVIDIA unveiled NemoClaw, an "open blueprint" designed to build autonomous AI engineers that can handle the grueling, soul-crushing workflows of industrial simulation.
We’re talking about the high-level janitorial work of the engineering world: computer-aided design (CAD), meshing, and debugging simulations. Tasks that usually take human engineers weeks to navigate are supposedly being crushed into hours by these specialized agents.
Corporate titans like Siemens and Cadence are already jumping on the bandwagon. Cadence is using NemoClaw to slash RTL verification times for chip design from weeks to a single afternoon, while Siemens is plugging it into their "Fuse EDA AI Agent." If you’re an engineer who enjoys the quiet solitude of waiting for a simulation to finish, I have some bad news for your coffee breaks.
But NVIDIA isn't content with these agents just sitting in a server rack. The company also announced JetPack 7.2, bringing NemoClaw support to the Jetson platform. This effectively gives "agentic AI" a physical body, landing it inside humanoid robots, drones, and even smart vending machines.
The technical upgrades here are actually substantial for once, cutting through the usual AI hype. JetPack 7.2 adds Yocto project support, which is a massive win for industrial customers who need a leaner, more customizable Linux foundation.
The real-world stats are where things get interesting. SandStar, which runs AI vending machines in 30 countries, managed to use these optimizations to migrate from 16GB to 8GB devices. That’s a 40% memory reduction that actually saves a company cold, hard cash on hardware deployments.
Of course, letting an autonomous agent run wild on your factory floor sounds like a security nightmare. NVIDIA is trying to get ahead of that with "OpenShell," a runtime that acts as a digital bouncer, governing how agents access files, networks, and tools.
Whether these "AI engineers" can actually handle creative problem-solving when a simulation inevitably explodes remains to be seen. For now, they seem mostly focused on automating the boring stuff so humans can focus on the big picture—or at least, that’s what the PR says.
The era of the "physical AI agent" is officially here, whether we’re ready for "reasoning" traffic lights or not. Only time will tell how long it takes for your office to get its first autonomous "colleague."
Sources: NVIDIA Blog (NemoClaw), NVIDIA Blog (Jetson).
