Well, it sure seems like robotics is on the verge of mainstream. GTC 2025, NVIDIA’s huge once-a-year tech and developer conference, was overflowing with AI and robotics hype and cool stuff.

Robotics companies like Agility Robotics, 1x Robotics, NEURA, AGILE, and Field AI all made appearances.

First, Robots

In general, there were a lot less robots here than I was hoping to see — but what we did see was pretty impressive, including a first look at the Neo Gamma from 1x, demo by Agility Robotics, and some incredibly cool work from Boston Dynamics:

Boston Dynamics has now joined the reinforcement learning train; this video is an example of DeepMimic-style whole body reinforcement learning guided from human demonstrations, and shows just how amazing the hardware is.

Upcoming Stuff from NVIDIA

NVIDIA CEO Jensen Huang’s keynote showed a bunch of other upcoming stuff:

• New products for AI compute, particularly the DGX Spark line

• Forthcoming Omniverse and Isaac Lab releases targeted at improving robotics capabilities; these include things like Newton, new physics being developed with Google Deepmind and Disney.

• New large-scale simulation tools coming for building digital twins

I think the NVIDIA foray into smaller and more “affordable” DGX machines is interesting; it foresees a future where developers have a proper “AI desktop” to run large models and simulations on locally, which I think definitely makes sense in some applications.

They also demonstrated the adorable walking robots from Disney which, like everything else, are running policies trained sim-to-real.

The Business of AI

There are a ton — and I mean a ton — of AI cloud providers now. Akash, Lambda Labs, and more. All the big players were there as well: AWS (Amazon Web Services), Dell, Oracle, etc. And so many more that I really can’t go into much detail, so instead will paint with broad strokes.

Enterprise AI was everywhere at GTC. Everyone there wanted you to buy their gpus-as-a-service, or their other AI cloud services, or was offering to somehow facilitate robot training or data processing. It’s amazing to me how fast this whole ecosystem has exploded, and honestly it feels unclear what the value add of a lot of these companies is. How much room for middlemen exists in this space? That remains to be seen.

A lot of the challenges of AI seem as present as ever, though, with very little seeming to focus on open-world perception. Note those robot demos: the Boston Dynamics running video linked above didn’t have anything like perception in the loop, and neither do the ubiquitous Unitree dancing videos that show up every week. A lot of the core problems of physical AI don’t seem significantly closer.

Conclusions and Takeaways

• It’s amazing how big this event is - 25,000 attendees, massive sprawling venue, Jensen Huang is basically a celebrity now.

• Security was weirdly tight, with bag checks and detectors between many of the sessions and the exhibit hall.

• Humanoids are going strong: lots of incredible videos and demos. More companies entering the space, especially with more stripped-down “humanoids” (multi-armed integrated manipulators) like AGILE or Cobot.

• BUT the prominence was much subdued from last year, with less prominently displayed robots, and more of a focus on enterprise AI from anything outside the keynote. Some of the biggest robotics players like Figure and Tesla, of course, do not attend, and the biggest booths in the robotics sections were very old-school companies like Teradyne/Universal Robots and Intuitive Surgical (maker of the Da Vinci minimally invasive surgery robot).

• AI continues to eat the world: AI is everywhere here, and robotics is really prominently displayed as probably the coolest single example.

All in all, it was an interesting event, and I’m really curious how different it ends up looking next year.