NVIDIA CEO Jensen Huang’s definition of AGI is telling

Artificial General Intelligence, or AGI, has spent the last year or so as the AI ​​industry’s favorite buzzword. As the sector’s leading companies burn through capital at historic rates, racking up energy costs and investor expectations that grow harder to meet by the quarter, the promise of imminent human-level machine intelligence has become a useful thing to have in your back pocket.

Whether we’re actually close to that milestone depends almost entirely on how you define it. That definitional flexibility, it turns out, is doing a lot of work.

Take, for example, Jensen Huang, the CEO of NVIDIA — a company currently valued at roughly $4 trillion, built largely on the GPU hardware that powers the AI ​​boom — who recently sat down with podcaster Lex Fridman for a wide-ranging conversation covering data centers, geopolitics, and the question of whether AGI has already arrived. Huang thinks it has. The reasoning behind that claim, however, is fairly dubious.

As Fridman points out, Huang has previously said the timeline for AGI depends on what defines it. At the 2023 New York Times DealBook Summit, Huang defined AGI as software capable of passing tests that approximate normal human intelligence at a reasonably competitive level. He expected AI to clear that bar within five years.

For his part, Fridman offered Huang a generous definition to work with: true AGI, in Fridman’s framing, would look like an AI capable of starting, growing, and running a technology company worth more than a billion dollars. He asked whether that was achievable in the next five to 20 years, given the recent proliferation of agentic AI tools like OpenClaw.

Huang didn’t need five to 20 years. “I think it’s now. I think we’ve achieved AGI,” he replied to Fridman.

That, however, is based on a narrow interpretation of what Fridman asked. The way Huang sees it, the AI ​​doesn’t need to build anything lasting. It doesn’t need to manage people, navigate a board, or sustain a business. It just needs to hit a billion dollars once.

“You said a billion,” Huang told Fridman, “and you didn’t say forever.”

The through-line in both cases isn’t a consistent theory of machine intelligence. It’s a consistent pattern of defining the threshold in whatever way makes “yes, we’re there” the easiest possible answer. His illustration of what that might look like is telling.

After his initial answer, Huang lays out his thoughts, describing a scenario in which an AI creates a simple web service — some app that goes viral, gets used by a few billion people at 50 cents a pop, and then quietly folds. He then points to the dot-com era as precedent, arguing that most of those websites were no more sophisticated than what an AI agent could generate today.

Huang was also candid about the ceiling of that vision. “The odds of 100,000 of those agents building NVIDIA,” he said plainly, “is zero percent.” That’s not a small caveat. It’s the whole ballgame.

What Huang is actually describing — a viral app that monetizes briefly and dies — is a far cry from the transformative, economy-reshaping AGI that dominates the public conversation. So, by his own admission, the kind of compound institutional intelligence required to build something like NVIDIA is nowhere in the picture yet.