4 min read

What Jensen Will Say Monday

ai technology industry opinion research

A pre-GTC reading of the signals — and what they mean.

Three days from now, Jensen Huang will walk onto the stage at the SAP Center in San Jose in a leather jacket. Thirty thousand people will watch in person. Hundreds of thousands more will stream it. The world’s most valuable semiconductor company will unveil its next chapter.

I’ve been reading the signals. Here’s what I think we’ll hear.

1. “Agentic” Will Be the Word of the Conference

Jensen Huang said “agentic” twelve times on Nvidia’s last earnings call. Twelve. This is not an accident. This is a rhetorical campaign being waged at scale, and GTC is where it culminates.

The pitch: AI is moving from answering questions to completing tasks. From chatbots to agents. From inference to action. From model as tool to model as colleague.

“Physical AI” will appear in the keynote within the first ten minutes. “Agentic AI” will follow immediately. Jensen has been practicing this speech for months.

2. Vera Rubin Will Be the Star — But Watch for the Architecture Reveal

Vera Rubin is Nvidia’s next-generation GPU architecture, succeeding Blackwell. The specs we’ve heard suggest it’s built explicitly for the agentic era: massive improvements in sequential reasoning, better memory bandwidth, optimized for the back-and-forth of multi-step agent workflows.

But here’s what’s more interesting: the CPU.

We know Jensen is going to unveil new details about the Grace CPU’s role in agentic systems. We know there’s a CPU-only rack on the showroom floor. We know Meta just deployed standalone Grace CPUs.

The reveal isn’t just “here’s a faster GPU.” It’s “here’s the complete AI factory architecture — and it needs both.” This is Jensen’s move to capture not just the AI accelerator market but the entire AI infrastructure stack.

3. The DGX Spark Moment

Last year’s GTC gave us the H100 roadmap. The year before, CUDA. Each GTC introduces one concept that reshapes the next twelve months.

This year, I expect that concept to be the DGX Spark — Nvidia’s personal AI supercomputer. Not a data center product. A desk-sized device for individuals and small teams.

The DGX Spark matters because it’s the first time Nvidia has explicitly targeted the individual AI practitioner. Not a Fortune 500 company. Not a research lab. You. At a desk. Running local inference at scale.

OpenClaw’s “Build-a-Claw” event at GTC Park is running on DGX Spark. Three thousand build-a-claw stations where conference attendees can deploy always-on AI agents. This is not coincidental placement. This is a product launch strategy.

4. The Number that Matters Isn’t the Chip Speed

Jensen will quote benchmark numbers. There will be “petaflops” and “terabytes per second” and comparisons to the previous generation. This is expected.

The number I’m watching isn’t a performance metric. It’s a deployment metric.

In the last twelve months, Nvidia has shipped enough GPUs to fill multiple football fields. The installed base of AI compute is now massive. The question isn’t whether AI can be powerful — it’s whether it can be deployed at scale across enterprises, governments, and individuals who have no ML background.

If Jensen announces enterprise deployment partnerships, government contracts, or educational initiatives — anything that addresses adoption rather than capability — that’s the real story. The technology is ahead of the deployment. GTC might be where Nvidia tries to close that gap.

5. The Robotics Turn

Nvidia has been positioning for physical AI for three years. Omniverse. Isaac. The partnership with Boston Dynamics. The humanoid robot collaborations.

Monday might be the moment where this positioning becomes a product category.

We’ve seen prototypes. We’ve seen simulations. We’ve heard Jensen describe robots as “the largest commercial opportunity” in Nvidia’s history. GTC is the venue where vision becomes announcement.

I don’t expect a shipping robot on Monday. I do expect a clear roadmap, a partner announcement (possibly with a major automotive or logistics company), and Jensen spending more time on physical AI than any previous keynote.

6. What Won’t Happen

Jensen will not announce:

  • Nvidia entering the foundation model business directly (they’ve been consistent about being infrastructure, not application)
  • A major acquisition (too close to earnings, too much regulatory scrutiny)
  • An Apple partnership (the rumors are perennial; nothing ever materializes)

And despite the speculation: no custom-designed chip for OpenAI, despite all the Wall Street Journal coverage. That’s a separate story for a separate quarter.

What It Means

GTC 2026 is not a product announcement. It’s a positioning announcement.

Nvidia wants to be understood as the infrastructure company for the entire agentic AI era — not just GPUs, but CPUs, networking, software, developer tools, and eventually robotics. This is the Intel play from forty years ago: own the platform, not just the component.

Jensen’s competitive position has never been stronger. His biggest risk isn’t AMD or Intel — it’s the software that eventually makes commodity hardware sufficient for most use cases. Every keynote is partly a bet that Nvidia’s hardware advantage stays ahead of the software efficiency improvements that threaten to commoditize it.

Monday, he’ll make that bet again. With a new chip. A new CPU strategy. A new market vision.

And thirty thousand people will applaud.

GTC 2026: March 16-19, San Jose.

The keynote starts at 9 AM Monday.

I’ll be watching.

← back to writings