Nvidia has shown off a new generation of AI-powered robots that can teach themselves to perform delicate, hands-on tasks — including snapping a graphics card into a PC motherboard.
According to Tom's Hardware, Nvidia revealed "agentic" robots that learn high-precision, dexterous work in the real world rather than being painstakingly programmed for each step. A demonstration video shows a robot "solve high-precision tasks like… installing GPUs all by itself," the outlet reports.
Installing a graphics card is a useful test case because it is genuinely fiddly: the card has to line up precisely with a slot and be seated with just the right amount of force — too little and it doesn't connect, too much and something breaks. Getting a machine to figure that out on its own, as Tom's Hardware describes, is a meaningfully harder problem than the repetitive, pre-scripted motions most factory robots perform today.
The key word in Nvidia's framing is "agentic" — robots that adapt and learn a task themselves instead of waiting for an engineer to spell out every movement. That shift, from machines that follow instructions to machines that work out how to do something, is what Nvidia is highlighting here.
The sources covering this announcement are limited to Tom's Hardware's report and its syndication via Google News, so the demonstration's full scope, reliability, and timeline for any real-world use remain unclear from the available material.
Why it matters: if robots can teach themselves the kind of careful, hands-on assembly work that has long required human fingers, it points toward automation reaching tasks that until now were considered too delicate for machines.