Blog

By Patrick Wheeler and Cara Harrington

In mid-March, we traveled to San Jose for NVIDIA GTC 2026, the annual conference that has become one of the most important gatherings in AI. Over 30,000 attendees packed the San Jose Convention Center and SAP Center for four days of keynotes, panels, demos, and hallway conversations about where AI is headed next. As representatives of the Center for Digital Strategies at Tuck School of Business, we went to learn and came back with more questions than answers, which is exactly what a good conference should do.

Here are five themes that stuck with us.

1. The AI industry is pivoting from training to inference — and that changes everything

CEO Jensen Huang’s two-hour keynote drove home a point that should matter to every business leader: the AI industry is shifting from a world focused on training models to one focused on inference, actually running them at scale. NVIDIA now frames its data centers as “AI token factories,” industrial-scale operations optimized for producing outputs rather than just building models.

Why does this matter outside of a server room? Because inference is where AI meets the real world. It’s the moment a customer gets a response, a doctor gets a recommendation, or an agent completes a task. As this shift accelerates, the companies that figure out how to deploy AI effectively, not just build it, will have the advantage.

2. The bottleneck isn’t intelligence, it’s infrastructure

One of the most grounding conversations at GTC was a fireside chat between Google’s Jeff Dean and NVIDIA’s Bill Dally. Their core message: networking is the real constraint in scaling AI systems right now. It’s not that models aren’t smart enough. It’s that the pipes connecting everything together, the physical infrastructure, can’t keep up.

This extends beyond hardware. A recurring theme across multiple sessions was that most enterprise tools were designed for human speeds, but AI agents operate roughly 50 times faster. When an agent can process information in milliseconds but has to wait on a tool built for someone clicking through a menu, the tool becomes the bottleneck. This is a massive opportunity for software companies and a wake-up call for organizations deploying agentic AI: your infrastructure has to be rebuilt for machine speed, not human speed.

3. OpenClaw was everywhere at GTC, and agentic AI is just getting started.

If there was a single technology that defined the energy at GTC, it was OpenClaw — the open-source agentic AI framework that has exploded in popularity since its launch earlier this year. NVIDIA went all in, dedicating an entire bootcamp and maker space to OpenClaw throughout the week, and introducing NemoClaw, an enterprise-grade reference stack designed to make agentic AI safer for corporate deployment.

But the bigger conversation was about what agents mean for how companies operate. During an open-source panel that included Jensen Huang, one founder offered a striking framework: in any modern enterprise, there are three critical capabilities: executing code, accessing sensitive data, and communicating with external audiences. No single person should do all three, except the CEO. That’s a useful mental model for thinking about how to govern AI agents, which can theoretically do all three simultaneously and at scale.

The example of prior authorization from healthcare made this concrete. Today, when a doctor’s request gets denied by an insurer, someone has to manually gather data and file an appeal. This process eats enormous amounts of time and drives physician burnout. An AI agent can handle this workflow automatically, collecting the right data and aligning the appeal to the specific rejection reason. These aren’t hypothetical use cases. They’re in development now.

4. “LLM” might not be the buzzword for long: World models are the next frontier

Just as “LLM” entered mainstream vocabulary over the past two years, “world model” is poised to follow. Multiple panels and demos explored how AI is moving beyond language into models that understand and simulate the physical world — critical for robotics, autonomous vehicles, industrial applications, and creative tools.

Runway, which started as a video generation company, is a telling example. Anastasis Germanidis, Runway Co-Founder and CTO, gave a talk on Runway’s GWM-1 world model and why they believe video diffusion-based world models are the most direct path to general-purpose simulation. When a video generation startup pivots to world modeling, it tells you where the center of gravity is moving.

For business leaders, the implication is that AI’s impact is about to expand well beyond text-based knowledge work. The physical world is next.

5. Creativity is not being replaced, it’s being redistributed

Cara’s background in entertainment drew her to the panels on media, entertainment, and creativity. The conversation at GTC wasn’t about AI replacing creative professionals, it was about democratization and responsible use. Panelists discussed how these tools can lower barriers to entry for independent creators while enhancing the capabilities of established production teams.

The tone was striking. This wasn’t techno-utopianism or doom. It was working professionals in entertainment talking honestly about how to integrate powerful new tools without losing what makes creative work meaningful. For anyone navigating a career that sits at the intersection of content and technology, that conversation is essential to follow.

The real conference happened between sessions

Some of our most valuable moments at GTC happened not in keynote halls but on the exhibitor floor talking with founders, watching demos, and hearing how people across industries are thinking about AI in their own work. The energy was unmistakable: while the big companies dominated the headlines, it was the individual conversations about what this technology is and how people want to use it that will ultimately shape how AI shows up in everyday life.

GTC 2026 showed us an industry that is simultaneously maturing and wide open. Enterprise best practices are forming, world models are gaining momentum, and agentic AI is moving from concept to deployment. But companies are still very much figuring things out in real time, and the founders building the future are learning as they go. That combination of ambition and humility,  of progress and uncertainty, is what makes this moment so compelling to study, to teach, and to be part of.

Patrick Wheeler is Executive Director of the Center for Digital Strategies at Tuck School of Business, Dartmouth. Cara Harrington is a CDS Fellow conducting original research on AI adoption in professional settings.