Jensen Huang has watched — and largely built — the infrastructure on which the AI era runs. This interview, conducted at MIT in early 2024, covers his long view on the technology transition underway, his unorthodox views on product strategy, and what he tells other CEOs when they ask him how to think about the next decade.
You've been building GPU infrastructure for decades. At what point did you realise this would be the computing substrate for AI?
The connection between parallel computation and machine learning became clear to me around 2012 when we saw what happened with AlexNet. Geoffrey Hinton and his team achieved a breakthrough in image recognition by running their neural network on two of our GPUs. The performance gap over traditional CPU-based approaches wasn't incremental — it was an order of magnitude. That was the moment.
What I didn't predict clearly enough was the pace. I thought we had more time. The capability curve has been steeper than even our most optimistic internal projections from a decade ago. The demand for compute has grown faster than our ability to supply it, and our ability to supply it has been growing at an extraordinary rate. That's a remarkable problem to have.
You've said AGI could arrive in five years. That's a bold claim. What's the basis?
I should be precise about what I mean. There are tests we use to measure human intelligence — standardised exams, reasoning benchmarks, reading comprehension tasks. In five years, I believe AI will be able to pass essentially every such test with human-level or better performance. That's not the same as saying AI will have human-like general understanding or consciousness — those are harder and more contested questions.
But in terms of measurable cognitive task performance, the trajectory is clear if you plot it. The capability has roughly doubled every twelve months on standard benchmarks. Five years of that compounding takes you somewhere very interesting. The honest answer is that I don't know exactly when or exactly what form it takes. But I think the people who say we're decades away are not looking at the data.
“The companies that will win are the ones that build institutional muscle for continuous AI integration — not ones treating it as a project with a completion date.”
What does this mean for how companies should be operating today?
The computer industry is completely reinventing itself. Every layer of the stack is being reimagined — hardware, software, applications, workflows. This doesn't happen cleanly or uniformly. Some parts move faster than others. But the direction is clear and the rate is fast enough that the companies that aren't actively working on AI integration now will be behind in a way that's hard to recover from.
What I tell CEOs: the question isn't whether to integrate AI. The question is which workflows to prioritise first, and how to build the organisational capability to do it continuously. This is an ongoing process, not a one-time migration. The companies that will win are the ones that build the institutional muscle for continuous AI integration rather than treating it as a project with a completion date.
NVIDIA's position is unique — you make the hardware that everyone else needs. How do you think about that responsibility?
We think about it a great deal. We're infrastructure. Roads, electricity — this is how I think about what we build. Roads don't decide who uses them or where they go. Electricity doesn't decide what it powers. Our role is to make the infrastructure as capable, as accessible, and as efficient as possible and then trust that the people using it will create things we haven't imagined.
At the same time, we're not passive about the downstream implications. We have relationships with the major AI labs. We participate in conversations about safety and policy. We're not in a position of saying our only job is to ship chips. The infrastructure you build shapes what's possible. That's a form of influence that comes with responsibility.
“Never give up. The number of times NVIDIA should have rationally given up is significant.”
You built NVIDIA over thirty years. What's the one leadership principle that has held constant?
Never give up. It sounds simple and it is simple. But the number of times NVIDIA should have — by rational calculation — given up is significant. Early on we were nearly bankrupt. We've been disrupted and had to reinvent. We've had product failures that caused real harm to the company. Each time, the answer was the same: understand what happened, learn from it, and keep going.
The leaders I've watched fail — and I've watched many — the failure is usually not that they made one catastrophic mistake. It's that they gave up at the wrong moment, or they pivoted away from something before it had a real chance to work. Conviction plus persistence is the formula. The conviction is harder to define, but it's that sense that you're working on something that matters and that you're building toward something real.