A Conversation with Jensen Huang
Here's what he had to say.
The AI conversation has moved fast. Two years ago, generative AI was the story. A year ago, it was reasoning. Now, we're somewhere else entirely — and Jensen Huang has a clear read on where.
In an exclusive conversation with HumanX CEO Stefan Weitz, the NVIDIA CEO laid out where the real opportunities are for enterprise leaders — and why the companies that treat this moment with too much caution may be the ones that fall behind.
AI Is Bigger Than LLMs
Huang pushed back on the tendency to equate AI with large language models. "There's a vast number of 'species' of different models," he said — models that understand proteins, physics, chemistry, not just language. Wherever there's encoded information, AI can learn to represent it.
"Chatbots are a very important part of it, language is a very important part of it, but AI is a gigantic industry with a gigantic number of applications."
The Three Waves — And We're Entering the Third
Huang mapped recent AI history into three phases. The first was generative AI — the ChatGPT moment. Transformative, but noisy. Models hallucinated and trust was low. The second was reasoning — OpenAI's o1 and o3 made outputs grounded and reliable. "Our trust level went up. As a result, it became more useful."
The third wave — happening now — is agentic AI. AI that doesn't just answer questions but does things. Huang pointed to Claude Code and OpenAI Codex as early signals: systems that set up environments, connect to tools, and iterate toward solutions autonomously.
"The prompts used to be 'What,' 'When,' 'Where,' 'How.' Now the first words are 'Create something for me,' 'Build something for me.'"
Stop Measuring Your Way Into Innovation
When asked what companies should not do with AI, Huang was direct: don't over-measure and don't be cynical.
"You can't measure your way into innovation, you can't measure your way into breakthroughs, you can't measure your way into some discontinuous level of success."
Not every company can let a thousand flowers bloom — but even with three, the mindset matters more than the scale. Treat failure as a data point, not a verdict.
The IT Industry Is About to Be Reinvented
Today's IT industry was built for human users. That's changing. "AI is software that uses software," Huang said. When a hundred billion AI agents become tool users, the addressable market for software explodes.
The question for enterprise leaders isn't just how to deploy AI internally — it's how your products and platforms will serve a world where AI agents are the primary users.
The Five-Layer Stack — And Where It Breaks
Huang's five-layer AI infrastructure stack — energy, chips, infrastructure, models, and applications — came up when Weitz asked which layer faces the most stress. Applications are the most important, Huang said, but energy is the bottleneck.
"Every single data center will have to be extraordinarily careful about the tokens-per-watt." The AI buildout isn't just a software story — it's an energy story, and every watt needs to translate to revenue.
Two Years Away From a Different World
When asked how long until a non-technical CEO can describe a business problem and have AI execute it end-to-end, Huang didn't hesitate: one to two years.
"Anything that you could ask an assistant to do today, or an employee to do today, you're not much further away from one or two years of being able to say that to a Claude, and have Claude do that."
The Bottom Line
Huang's message was clear: the window for cautious experimentation is closing. AI is entering a phase where it doesn't just inform decisions — it executes them. The companies that lean in now will define the next era of enterprise. The ones that wait for certainty may not get a second chance.
Watch the full conversation on-demand.
Register for HumanX 2026 Amsterdam (September 22–24) or HumanX 2027 Las Vegas (March 8–10).