The future of AI is already live
Twenty-five thousand people flocked to San Jose this week for Nvidia GTC, the chipmaker’s annual AI industry conference, often called “the Woodstock of AI.” (Think fewer guitars, more GPUs.) It’s where developers and executives gather to discuss what’s next and, increasingly, to gawk at what’s already in motion.
This year, there was a growing sense that things are getting real… and not just because you couldn’t take three steps without running into an actual robot. Here’s what kept coming up:
Your AI coworkers are already onboarding (and they never log off)
Nvidia CEO Jensen Huang (leather jacket-clad, as per usual) used his keynote speech to double down on agentic AI — basically AI that doesn’t just answer questions but goes off and does the work on its own.
Right now, a lot of that momentum is coming from OpenClaw, a fast-growing open source framework that’s taken off in Silicon Valley in recent months. It lets developers build AI agents (“claws”) that can plan tasks, take actions, and keep running continuously in the background without human supervision. Huang announced a new tool, NemoClaw, which makes that framework more usable inside companies by layering in security and privacy controls.
You could begin to see all of this in action on the floor at the packed Build-a-Claw tent, which felt a bit like a science fair where everyone was building autonomous AI agents instead of volcanoes. One developer built a claw that could read all your newsletters and turn them into a personalized briefing. Another created one that edits your resume and applies to relevant jobs for you.
Huang described a future where there are hundreds of AI workers for every one human employee. “They’ll be working around the clock,” he said. “So hopefully our people don’t have to keep up with them.” (But will your new AI coworkers go for coffee with you? TBD.)
Tokens, tokens, tokens
If you had a nickel for every time someone said “tokens” at GTC, you’d have… a lot of nickels.
Tokens are the units AI models use to process information, and they’re how usage gets priced. When you have agents running continuously, those tokens start to add up quickly. (Read: $$$) Teams are starting to track how many tokens their workflows consume and how to make usage more efficient, similar to how they already manage cloud costs.
Glamorous? No. But suddenly a big part of the conversation, especially as we reach what Huang calls the “inference inflection point”— when AI shifts from being trained to “inferring” (reading, generating, making decisions) in real time.
Yes, there were robots — including your toddler’s favorite snowman
Some were doing useful work, like sorting objects or navigating warehouse-style setups. Others were clearly there for the photo ops.
There were even cameos from a robotic version of Olaf, the beloved snowman from Disney’s Frozen. Not the most technically revealing example, but he did give people something fun to text their kids.
And the takeaway was clear: Nvidia is pushing what it calls “physical AI,” bringing these systems beyond the screen into the real world.
That push says a lot about where the company is headed. Nvidia isn’t just making chips anymore. It’s building the systems these machines run on, from the models to the infrastructure underneath them. If AI is going to show up in factories, warehouses, and vehicles, all of those pieces need to work together—and Nvidia wants to provide them.
Speaking of vehicles, robotaxis are almost here. For real this time.
Autonomous vehicles made another appearance, with more structure than in past years.
Nvidia is working with Uber, Lyft, and a lineup of automakers like Hyundai and Nissan to power robotaxi fleets, with deployments expected to start in major cities around 2027.
There’s still a lot to figure out (don’t let your driver’s license lapse just yet), but it feels more organized — and more imminent.
TL;DR
GTC this year felt less like a preview and more like a progress report.
Systems keep running after you log off. Costs are getting tracked more closely. Robots are showing up in practical (and occasionally theatrical) ways.
And across all of it, the takeaway was clear: This technology is moving out of the demo phase and into real life.
