Nvidia GTC 2025: The Good and The Bad of My First Time at the Super Bowl of AI Link to heading
Welcome to the jungle. My first steps into GTC 2025.
Intro: My First Time at GTC Link to heading
Why did I go to GTC? Honestly… curiosity. The kind that builds when you keep hearing about a thing—AI this, Nvidia that—and realize it’s not just noise anymore. It’s movement. It’s momentum. And I wanted in.
I’ve been orbiting this world from the outside for a while now, but this year felt like the right time to jump in. To learn the language, feel the pulse, and get close to a technology that’s just beginning to hit its stride. And what better place than a conference hosted by the company shaping the curve?
There were signs early on that this wasn’t going to be a small affair—like the lack of nearby hotel rooms and the cost of flights into San Jose. I flew into San Francisco instead to save some cash, only to be greeted by a $100 Uber estimate just to get to the San Jose airport. Fortunately, a $28 shuttle saved the day. Crisis #1 averted.
Once I landed in San Jose, things started clicking. I picked up my badge right at the airport—a convenience I’d never experienced at a conference before. The hotel turned out to be a win, too. A mobile app for check-in and checkout, a decent hot breakfast, groceries and restaurants within walking distance, and yes—a Starbucks on the corner. I got settled, took a hot shower, let my people know I’d landed safely, and closed my eyes, ready for the long and exciting week ahead.
The Good: Mind-Blown Moments and Game-Changers Link to heading
Right from the start, GTC 2025 felt massive. I heard whispers that it was the largest it had ever been, and Jensen Huang himself called it the Super Bowl of AI. It showed.
A Thousand Paths to Explore Link to heading
The sheer number of sessions—over 1,000—was wild. Quantum computing, agentic AI, healthcare, food, the auto industry, robotics, full-day workshops, even talks running in the middle of the night. Some might say it was too much. But to me, it showed just how healthy and diverse the AI ecosystem really is. Every topic had beginner-to-advanced content, and there was space to either lock into a track or hop between them without getting lost.
Everywhere you turned, there was something happening. And then three more things happening.
Training Labs: Where the Real Learning Happened Link to heading
The Training Labs were the crown jewel of my week. No fluff, no filler—just sharp, 2-hour sessions with clear slides, working code, and instructors who knew their stuff.
Here’s a quick rundown of the labs I attended and why they hit home:
- Fundamentals of Agentic AI
A condensed 8-hour course brought down to two intense hours. A solid foundation to kick things off, especially with Agentic AI as a core theme throughout the week. - Automate 5G Network Configurations with Agents
I knew little about 5G networks going in, but left understanding how agents can manage and model complex systems. Bonus: learned about an open-source tool that made it all click. - Learn to Build Agentic AI Workflows for Enterprise Applications
This one felt like crossing the bridge from enthusiast to enterprise. It gave me a clearer view into how enterprises might structure and implement agentic workflows—and even gave me ideas for better conversations with non-engineers and stakeholders back in the real world. - Domain Adaptive Pre-Training, Evaluating RAG and Semantic Search Systems, Make Retrieval Better
These sessions explored different angles of RAG (Retrieval-Augmented Generation). From tuning LLMs for specific domains to improving retrieval pipelines, they gave me the full picture of how crucial RAG is for AI that actually works. - Accelerating Linguistic Diversity
This one was personal. I work on a project capturing the stories and voices of Black Americans, and this lab helped me imagine how AI could highlight the nuance, rhythm, and beauty of that dialogue in ways that are both respectful and powerful.
Not just theory—actual tools, actual code, actual learning.
And the best part? Recordings, slides, and code are all available post-conference. The learning doesn’t stop when you go home.
Paper Reception: Science Fair Vibes for Grown-Ups Link to heading
One of the most unexpectedly amazing moments was the Paper Reception on Monday night. Held at the San Jose Civic Center, it gave me science fair flashbacks—but instead of baking soda volcanoes, there were 50+ brand-new academic papers, with the authors standing by to break them down.
It was packed. High energy. Researchers lit up as people gathered around their posters, asking thoughtful questions and diving deep into the bleeding edge of AI research. You could feel the pride in their work, and it was contagious.
The Keynote: Jensen’s Grand Vision Link to heading
The keynote was pure spectacle, held at the SAP Center with a full house—standing room only. Jensen Huang didn’t just present—he performed. He introduced the Blackwell architecture, laid out Nvidia’s vision of agentic and physical AI systems, announced partnerships with GM for autonomous vehicles and next-gen factories, and sketched a roadmap for AI factories, software stacks, and open-source model development.
His message? This isn’t just about building smarter models—it’s about reshaping infrastructure, strategy, and how intelligence interacts with the physical world.
The Bad: A Little Too Big to Breathe Link to heading
Overall, I had a great time. But the logistics? Rough. And for many of us, they left a mark.
Session Chaos Link to heading
On day one, I left my first training lab with 15 minutes to spare before the next one. No problem, right? Wrong. The lower level of the convention center was gridlocked—people everywhere, stairs jammed, and lines that snaked around corners like a theme park ride. I quickly realized: if I wasn’t in line at least an hour early, I wasn’t getting in.
This was 9:55 a.m. Day one. I didn’t make it into the next session.
Rain + Park Lunch = Nope Link to heading
Lunch was outdoors… in a park… on a rainy Monday. Lines were long, the ground was soggy, and while the sun peeked out occasionally, it didn’t stick around. A few of us found shelter under vendor tents, but many just bailed. They eventually opened lunch early to manage the situation, but by then, a lot of folks were already drenched and disappointed.
Shuttles, Apps, and a Bit of a Mess Link to heading
The shuttle system looked good on paper. The app let you select pickup and drop-off points and showed you a schedule. But once you reached the shuttle area, none of the buses had matching info. You had to find a tiny, easy-to-miss tent and talk to a staffer with a constantly freezing laptop. She tried her best, but tech issues made it a struggle.
Wi-Fi Blackouts & SAP Center Dead Zones Link to heading
The internet flat-out died multiple times during the week. For a conference about cutting-edge AI infrastructure, the basic networking failed hard. No Wi-Fi. No mobile signal. One day, the SAP Center was a complete dead zone—no service until you walked outside and down the street. Not ideal when your entire schedule, map, and contact list live in your phone.
Looking Ahead: Nvidia’s Strategic Playbook Link to heading
After soaking in the week and letting the content marinate, one thing stands out: Nvidia knows its future won’t rely solely on hardware.
It’s been clear to me since the launch of Omniverse back in 2021—Nvidia is building a moat not just with silicon, but with a sprawling, robust software ecosystem. GTC 2025 made that vision unmistakable. From acceleration libraries for nearly every industry vertical, to frameworks that make deploying complex systems more manageable, they’re playing the long game.
Case in point: their CES announcement of Digits—a desktop supercomputer roughly the size of a Mac Studio—made an appearance at the expo. So did their RTX Pro 6000 Blackwell series, in workstation, Max-Q, and server configurations. But the system that really caught my attention? The DGX Station. Quiet, powerful, compact—and a sign of where the personal AI workstation is heading.
And then there’s Agentic AI—a dominant theme all week. Nvidia doubled down by announcing AgentIQ, a new open-source framework for building AI agents. If you’re familiar with CrewAI or PydanticAI, you’ll feel right at home. What really caught my eye is its compatibility with Anthropic’s Model Context Protocol (MCP). According to Nvidia’s docs, AgentIQ supports tools served by MCP Servers as AgentIQ functions—making it a flexible, standards-aware platform for anyone thinking about agent infrastructure seriously.
Final Thoughts: Would I Go Again? Link to heading
Yes. But I’d come prepared—early arrival, packed patience, and a few snacks in my backpack for those outdoor lunches.
For all its hiccups, GTC 2025 delivered: in content, in community, in future-facing energy. It was overwhelming at times, but that’s kind of the point. AI is expanding fast, and this conference was a snapshot of just how wide and wild that expansion really is.
If you’re trying to get your bearings in this space, there’s no better place to dive in.