GTC 2025: The AI Moments That Mattered

From tokens to robots to a historical tribute, these moments defined Nvidia’s biggest event of the year.

GTC is where Nvidia sets the stage for AI’s next leap forward every year.

But this year’s event felt different.

Maybe it was the scale—thousands packed into the SAP Center, with watch parties spilling into plazas and university campuses. Maybe it was the energy—not just another year of GPU announcements, but a sense that AI is breaking into the real world in ways we haven’t seen before.

Or maybe it was the opening video—because nothing gets a crowd fired up like… tokens.

The Keynote Opened With a Video About Tokens—And People Loved It

Before Jensen Huang even walked on stage, Nvidia kicked off the keynote with a cinematic, sci-fi-esque video about tokens—the fundamental building blocks of AI.

In large language models (LLMs), tokens are how AI understands and processes information. More efficient tokenization means better, faster AI models, so Nvidia is obsessed with scaling AI beyond raw compute power.

And if the audience's reaction was any indication, Nvidia knows how to make even abstract AI concepts feel exciting.

Watch the full tokens video here:

But that was just the start.

The Robots Stole the Show

I’ll be honest—I came for the robots.

I don’t want it is about robots, but I LOVE them.

Video note: Yes, that’s Jensen Huang standing in front of the screen. I recorded this live as it was being shown.

So when Nvidia announced Newton, an open-source physics engine built in collaboration with Google DeepMind and Disney Research, it had my full attention.

Newton is designed to accelerate robot learning by creating hyper-realistic physics simulations, allowing AI-powered robots to train in virtual environments before ever touching the real world.

Why This Matters

• Bridging the Sim-to-Real Gap – Robotics AI has long struggled to translate simulated learning into real-world adaptability. Newton is designed to fix that.

• Broad Industry Applications – Disney is already using Newton to improve its robotic character platform, but the tech could have major implications for humanoid robots, automation, and AI-driven industrial robotics.

At every AI conference, robots are the crowd favorite—and this year was no exception.

Insert photo of Newton slide with Disney DeepMind partnership

Rubin: A Tribute That Hit Differently

One of the most powerful moments of the keynote came when Huang introduced the Rubin architecture—the successor to Blackwell, launching in 2026.

Rubin is named after Vera Rubin, the astronomer who confirmed the existence of dark matter. And unlike most tech product names, this one actually felt meaningful.

Huang paused to acknowledge Rubin’s family members in the audience, thanking them for being there. It was a rare moment of humanity in an industry that moves fast and often forgets to honor the people who paved the way.

In a space dominated by names of male physicists, this was a well-earned and long-overdue tribute.

AI Is Powering Self-Driving Cars—But Can It Make Them Safer?

The biggest business move in the keynote was Nvidia’s expanded partnership with General Motors—a major step toward AI-powered autonomous driving.

The key technology behind this? Halos, an AI-driven safety and transparency system designed to make self-driving technology more explainable and trustworthy.

Why This Matters

• Third-Party Safety Assessments – Every line of Halos’ AI code is independently reviewed for transparency.

• Shaping the Future of Autonomous Driving – With GM’s partnership, Nvidia is solidifying itself as the backbone of AI-powered transportation.

I’ll admit—enterprise AI isn’t what excites me most. But this is a big deal.

DGX Spark: AI Workstations That Fit on a Desk

Huang also introduced DGX Spark, a 20-petaflop AI workstation that could reshape who has access to high-performance AI computing.

Launched with his signature humor, he joked:

"This is something you might want to put on your Christmas list."

I captured exclusive footage of the moment—watch below.

Unlike Nvidia’s massive DGX systems built for data centers, DGX Spark is designed to be more accessible to AI developers, startups, and research labs. It will be available through OEM partners like HP, Dell, and Lenovo.

If Spark delivers on its promise, it could become the go-to workstation for AI development outside of cloud computing, making high-performance AI more accessible than ever.

Blackwell Ultra and Rubin: The AI Engines Powering It All

Of course, no Nvidia keynote would be complete without new silicon.

• Blackwell Ultra GB300: 50 percent more memory capacity and significantly higher performance.

• Rubin GPU Platform: Launching in 2026, doubling performance over Hopper. Rubin Ultra follows in 2027.

These advancements in GPU computing are what make everything else in AI possible—from robotics to LLMs to self-driving cars.

Final Thoughts: AI Is Moving Faster Than Anyone Expected

Nvidia’s GTC 2025 keynote wasn’t just about new chips. It was about how AI is becoming infrastructure.

• DGX Spark puts 20 petaflops of AI power on a desk.

• Newton is giving robots a new way to learn.

• Halos is making self-driving AI more explainable.

• Rubin is pushing AI compute to new limits.

The AI revolution isn’t something happening years from now. It is happening right now. We are past the point of speculation. The era of AI has begun, and what we build with it will define the decades to come.

Watch the full keynote here:

---

Vocabulary Key

• Tokens – The building blocks of AI processing, used in LLMs and generative models.

• Newton – Nvidia’s open-source robotics physics engine, built with DeepMind and Disney.

• Halos – An AI-powered safety system for autonomous driving.

• DGX Spark – A 20-petaflop AI workstation for developers and researchers.

• Blackwell Ultra – Nvidia’s latest high-performance AI chip.

• Rubin GPU – The next-generation AI architecture, launching in 2026.