Jensen Huang: Intelligence Is a Commodity, Agents Are the iPhone of Tokens

Why Jensen Huang Thinks We’ve Already Achieved AGI

Jensen Huang, the longest-serving tech CEO in the world, sat down with Lex Fridman for a sweeping conversation about Nvidia’s trajectory from a gaming GPU company to the engine of the AI revolution. At a $4 trillion valuation, Nvidia now builds what Huang calls “AI factories” — and his mental model has shifted from holding up a chip to envisioning planetary-scale computing infrastructure.

On agents as the new computing paradigm: “OpenClaw is the iPhone of tokens. It is the fastest growing application in history. It went straight up.” Huang draws a direct line from the four scaling laws — pre-training, post-training, test-time, and agentic — to a world where agents spawn sub-agents, use tools, access file systems, and do research autonomously. He architected Nvidia’s Vera Rubin rack specifically for agentic workloads two years before OpenClaw launched.

On intelligence becoming a commodity: “Intelligence is a commodity. I’m surrounded by intelligent people more intelligent than I am in each one of the spaces they’re in. And yet I have a role in that circle.” Huang argues that humanity, character, and compassion are the real superpowers — and that society has mistakenly elevated “intelligence” to a status it doesn’t deserve. With AI commoditizing cognition, what matters is what you do with it.

On the radical AGI claim: When asked if an AI system could start and run a billion-dollar company, Huang surprised Fridman: “I think we’ve achieved AGI.” His reasoning is pragmatic — a claw agent could create a viral web service that briefly reaches massive scale, much like internet-era startups. But building something like Nvidia? “100,000 of those agents building Nvidia — 0%.”

On why coding expands to a billion people: Huang reframes programming as specification — describing what you want a computer to build. “The number of software engineers at NVIDIA is going to grow, not decline… We just went from 30 million to probably 1 billion.” Every carpenter, accountant, and pharmacist becomes a coder when natural language is the programming interface. The radiologist analogy is instructive: despite superhuman computer vision since 2020, the number of radiologists grew because the purpose of the job expanded.

On extreme co-design and 60 direct reports: Huang runs Nvidia with 60 direct reports, no one-on-ones, and everything discussed in group settings. This mirrors the product: extreme co-design means optimizing across GPU, CPU, memory, networking, storage, power, cooling, and software simultaneously. “No conversation is ever one person. We present a problem and all of us attack it.”

6 Key Takeaways from Jensen Huang on AI’s Future

• Token factories replace data warehouses — Computing shifted from file retrieval to token generation. AI factories are revenue-generating infrastructure, not cost centers. Token pricing is segmenting like iPhones, with $1,000/million-token products “not if, but when.”
• Four scaling laws feed each other — Pre-training, post-training, test-time compute, and agentic scaling form a virtuous loop. Agents generate experiences, the best get memorized back into pre-training, refined in post-training, enhanced at inference, and deployed again.
• Supply chain is the real moat — Nvidia’s 1.3-million-component racks involve 200 suppliers. Huang personally visits CEOs of DRAM, TSMC, and infrastructure companies to shape their investment decisions years in advance. “We don’t have a contract with TSMC. Three decades, hundreds of billions of dollars.”
• CUDA installed base is everything — More than any technical advantage, it’s the developer install base that makes Nvidia unassailable. “If somebody came up with a GUDA or a TUDA, it wouldn’t make any difference.”
• China is the fastest innovating AI nation — 50% of world’s AI researchers, insane internal competition among provinces, open-source culture driven by social networks (“their schoolmates are their brothers for life”), and a builder-nation mentality.
• Grid power waste is the real opportunity — Power grids run at ~60% capacity 99% of the time. Huang advocates contractual agreements where data centers gracefully degrade during peak demand, rather than waiting years for new energy infrastructure.

What Jensen Huang’s Vision Means for AI-Powered Organizations

Jensen’s core thesis is that we’ve crossed the threshold from AI as a tool to AI as a worker — and the infrastructure implications are staggering. Token factories are the new manufacturing plants, agents are the new employees, and intelligence itself is table stakes. The organizations that win won’t be the ones with the smartest AI, but the ones that orchestrate AI workers most effectively — which is, notably, exactly how Huang describes his own role at Nvidia: a “dishwasher sitting in the middle of superhumans,” orchestrating 60 brilliant specialists through collective reasoning. The agentic era isn’t coming. According to the man building the hardware for it, it’s already here.