NVIDIA

Jensen Huang on the Factory That Turns Electricity Into Intelligence

Jensen Huang· Founder and CEO of NVIDIA at NVIDIA
·~15 min·English·NVIDIA
AI InfrastructureGPUInferenceBusiness Strategy
TL;DR

In a fireside chat at Wistron's U.S. plant, Jensen Huang recasts data centers as AI factories that turn electricity into intelligence tokens, argues this new industrial layer will become as fundamental to society as agriculture and railroads, and ties it to reindustrializing American manufacturing.

01Core Mental Model

Data Centers Stored Information. AI Factories Generate Intelligence.

Old data centers stored the past; AI factories generate a fresh answer every time you ask — the shift from retrieval to generative computing is why Huang stops calling the box a data center and starts calling it a factory.

So that retrieval based computing model is now being replaced by a generative model where we generate intelligence based on your question. So instead of doing search we now just ask it a question. And it would go off and do the search.

— Jensen Huang, NVIDIA
Key Insight
Reframing retrieval as generation is really a reframe of cost: storing a file is almost free once written, but generating an answer burns compute every single time — which is exactly why Huang can call the box a factory, and why AI economics look nothing like the web that came before it.
02The Physical Build-Out

A $4 Million Supercomputer, Produced Like a Phone

The GB300 is a supercomputer produced at consumer-product volume — 1.5 million parts, roughly two tons, about $4 million each, assembled by robots and cranked out at Wistron.

It has 1.5 million parts inside. These robots have to assemble 1.5 million parts. It's two tons to a ton. $4 million. And we're producing them. You know, like, like phones, you know.

— Jensen Huang, NVIDIA
Key Insight
By calling a $4 million supercomputer something you crank out like phones, Huang collapses chip design into heavy manufacturing — the message to Washington isn't abstract AI leadership, it's the plants, jobs, and resilient supply chain that come from building the machines on U.S. soil.
03The Reframe

AI Isn't the Chatbot — It's a Whole New Industry

The chatbot is just the visible tip — underneath, Huang argues, AI is a whole new industry spanning energy, chips, data centers, and models, whose real payoff is diffusion into every other industry.

However, inside it's important to realize AI is actually a whole new industry. AI is incredibly important to the energy industry.

— Jensen Huang, NVIDIA
Key Insight
Calling the chatbot merely the UI is a deliberate redirect — away from the crowded app layer and toward the infrastructure layers NVIDIA sells into — and the claim that diffusion into every industry is where the big win is quietly makes NVIDIA's addressable market the whole economy, not just tech.
04The Commodity

Tokens Have Value Because Humanity Already Runs on Them

Tokens have value because humanity already runs on them — every Netflix pixel came from a token, and AI factories now generate tokens that become words, images, songs, and answers.

It's not illogical that tokens have value. And in fact, humanity is largely token driven. Today when you watch Netflix, every one of those pixels came from a token.

— Jensen Huang, NVIDIA
Key Insight
The Netflix comparison does real work: you already pay for pixels without a second thought, so paying for intelligence-tokens feels no stranger — it normalizes a brand-new unit of economic value by anchoring it to one you never noticed you were buying.
05The Economic Shift

Every Company Becomes a Higher-CapEx Company

Every company becomes a higher-CapEx company — even a law firm's future intelligence is coffee plus hamburgers plus electricity, because intelligence can no longer be produced by cows and grass alone.

And I'll just give you an example of a very small industry legal. And so in the future their intelligence will be a combination of hamburgers and coffee and electricity. And those electricity will generate tokens which will be part of their intelligence.

— Jensen Huang, NVIDIA
Key Insight
Adding electricity to the law firm's input mix turns white-collar productivity into a capital-allocation question: firms must decide how much compute to pair with human expertise, moving AI from optional software toward productive infrastructure they invest in.
06The Thesis

AI Factories Will Become Fundamental Infrastructure

AI factories will become fundamental infrastructure — Huang places them beside agriculture, roads, and railroads and says they will add a CapEx layer worth trillions of dollars.

just as agriculture is a fundamental infrastructure of society, in the future, these AI factories will be a fundamental infrastructure of society.

— Jensen Huang, NVIDIA
Key Insight
Ranking AI factories beside roads, agriculture, and railroads is a claim about permanence, not just scale: infrastructure is what a society can't switch off, so if the analogy holds, the trillions in CapEx read as a baseline to maintain rather than a bubble to time.
07Sovereignty

No Country Can Outsource Its Intelligence

No country can outsource its intelligence — use off-the-shelf intelligence wherever possible, but create and cultivate your own, which is why Huang says countries and companies need both closed and open models.

Off the shelf movies, off the shelf intelligence, off the shelf music. But whenever you must or you desire, create your own.

— Jensen Huang, NVIDIA
Key Insight
The import-everything-but-build-your-own framing turns an apparent contradiction into a moat — arguing for open and closed models at once, because NVIDIA profits from both, while handing every nation a reason to build sovereign compute on NVIDIA hardware rather than rent it.