It was crowded. I mean it was really, really crowded and for a city that comfortably accommodates conferences of 12-15,000 attendees, San Jose struggled with GTC’s 34,000 or more. It was great to see the energy, and enthusiasm, but the crowds were horrific (and the 90-degree days didn’t help the situation). But San Jose is beautiful this time of year, so when thousands of attendees are being bused for 30-40 minutes in traffic from Cupertino and beyond to get to the San Jose convention center (because all of the local hotels were sold out), at least they could enjoy the view!

But for business, it was stellar! GTC is now one of the core venues to host/challenge the new era of IT, and the feeling of excitement, pride and innovation was everywhere. (I became so hypnotized by the frenzy, that I even started daydreaming how cool it would look if everyone was given a soft black Italian leather jacket!) Importantly, as I listened to customers and partners, it became clear that this GTC show is all about business rather than boondoggle (like so many trade shows have become). End-users, developers and partners alike came to learn. Learn from NVIDIA, learn from the ecosystem of partners and learn each other. Meetings were being held and the complexion of the Global 1000 IT function felt like it was being re-written in near-real time at the show!

So with so much buzzing in my head after I returned home, I jotted down my list of takeaways from the show:

1) Putting AI to Work Has Become the Primary Battleground

The conference made it clear the industry is shifting its headlines from the Hyperscalers training their trillion parameter AI models → Enterprise learning how to create their own models and then running them at scale (inference).

• Companies like HPE were demonstrating their AI Factory solutions that turn power into tokens to do valuable AI business work.
• NVIDIA emphasized inference repeatedly as the next trillion-dollar opportunity (not surprising since that is where the volume will be.)
• Architectures, chips, and software announcements during the show were heavily optimized for real-time and agentic AI workloads.

Implication: The competitive landscape is broadening beyond GPUs into full-stack inference systems.

---

2) The “$1 Trillion AI Infrastructure Opportunity” Is Now the Narrative

A headline theme was NVIDIA projecting ~$1T in AI infrastructure over the next few years. (Jensen’s Keynote)

Implication:

• AI is no longer experimental—it is being framed as core economic infrastructure and the upgrade to traditional computing.
• This narrative is driving hyperscaler, enterprise, service provider and sovereign investment cycles.

---

3) AI Factories Replace Traditional Data Centers

Jensen’s framing of “AI factories” was widely being discussed in the halls. Anything that consumes power and generates AI tokens is an AI factory in his talk-track.

• “AI Factories” are production facilities that convert power → compute → tokens → revenue.

Implication:

• Power is tied to Compute, which becomes directly tied to revenue generation, with energy availability being a gating factor to revenue
• Capex justification shifts from IT cost to industrial production economics. The direct line to revenue is being written in INK.

---

4) Agentic AI Is Moving From Concept to Deployment

“Agentic AI” (autonomous, tool-using systems) was a dominant theme as everryone was discussing how they could put AI to work:

• Systems that reason, plan, and execute tasks independently were showcased across demos and platforms. EXECUTE being the most exciting (and scary) word since there are a slew of implications when tasks can be dynamically authored and then executed automatically.

Implication:

• Software is evolving from hardwired tools that provide correlated information → dynamic autonomous digital workers.
• Infrastructure must support persistent, stateful, long-running AI processes (and the energy to add millions of digital workers is non-trivial.)

---

5) Full-Stack Expansion Beyond GPUs (CPU, Networking, Storage)

NVIDIA is no longer positioning itself as just a GPU “chip” vendor:

• Introduction of Vera CPUs and new Vera Rubin and rack-scale systems signals direct competition with a much wider range of suppliers
• New architectures (e.g., accelerated storage, DPUs) target system bottlenecks and a number of vendors were discussing ASIC-level answers to this growing concern.

Implication:

• NVIDIA is building a vertically integrated AI platform company and the Ecosystem will become more critical.
• This puts pressure on incumbents across CPUs, networking, and storage to deliver more, not incrementally more, but AI-specific more.
• Complementary providers and consultants/advisors will see significant uptake from the consumers that are trying to “get in the game.”

---

6) Physical AI and Robotics Are Entering a Commercial Phase

“Physical AI” (robots, autonomous systems, digital twins) were heavily emphasized. In fact, human-like robots were walking (and dancing) all over the convention center. The innovation hook is they were autonomous, making decisions on their own. It was very cool:

• AI is moving from digital-only “ChatGPT” value → real-world deployment via simulation and physical robotics.

Implication:

• Commercial, Industrial, automotive, and logistics sectors become major AI consumers, and the costs drive down as adoption and volume increases.
• Simulation (digital twins) becomes a prerequisite for real-world AI, and the sensing demonstrations were everywhere.

---

7) Open Models and Ecosystem Strategy Are Strategic Levers

There was significant discussion around open vs. closed AI models:

• Open ecosystems are enabling faster innovation and broader adoption which is the same discussion we heard 20 years ago with open-source operating systems (like Linux) versus closed software architectures (like Windows). I lived it. And it goes that “Open” means more people/developers are involved, and even in a well funded AI startup, the number of engineers working on a closed piece of software pales in comparison to the open source community of enthusists. (dozens or hundreds, versus thousands kind of thing)

Implication:

• NVIDIA is positioning itself as the foundational platform layer regardless of which framework/application model choice will be made. 
• Control shifts from models → infrastructure + developer ecosystem (CUDA, tools, frameworks). This is a core foundation discussion.

---

Bottom Line

The dominant meta-theme from GTC 2026 which I walked away with:

AI *HAS NOW* transitioned from a curious technology wave to an ready-for production industrial system for the next generation of IT. Those adopters will be rewarded. Those procrastinators will be in deficit and at a rapidly increasing competitive disadvantage.

Across all seven takeaways, the consistent pattern is:

• Shifts from pilot and POC to value through inference and production-scale model deployments
• Emergence of AI-native infrastructure (AI factories, sovereign and commercial, not just bigger data centers)
• Expansions to full-stack computing dominance; foundation, ecosystem, collaboration, trusted advisors, etc
• Movements into real-world, autonomous systems and putting AI to work DOING things, rather than simple reporting