On Wednesday’s earnings call, Nvidia's CEO announced that the tech industry had reached its holy grail, then demolished the claim in the very next sentence. Jensen Huang stated the company had “achieved AGI,” before promptly labeling the accomplishment “senseless.”
He’s not wrong. That is precisely the issue.
Artificial general intelligence is meant to mark the end of the AI race. Yet nobody agrees on its definition, much less on how anyone would recognize the moment it arrives, which renders the achievement itself just as arbitrary. Declare victory at any moment you choose. There is no one to audit the claim.
What Huang actually said
Questioned about OpenAI’s chase for AGI, Huang replied that where Nvidia is concerned, “for many tasks, we could say that we’ve already achieved AGI.” He offered no definition. No benchmark. Only a dismissal: “I think of all of those milestones and all those, you know, they’re kind of senseless at this point.”
His focus instead landed on AI shifting beyond basic prompt-and-response into autonomous agents that pick up new skills and refine themselves “recursively.” The thing that counts, Huang argued, is AI “doing productive and useful work” and “generating profitable tokens.” Compute begets tokens, tokens beget profit. “This is the exact phase where we’re at. Which is the reason why everybody’s leaning in.”
That closing line gives the game away. AGI is the garnish. Token economics is the actual business.
He’s said it before
Huang has made this pronouncement previously. Speaking on the Lex Fridman podcast in March, he said plainly, “I think we’ve achieved AGI.” On that occasion, too, he left the meaning unspecified.
Fridman countered with a curiously precise definition of his own: an AI capable of “essentially do your job” — that is, launching, scaling and operating a successful tech firm valued above $1 billion. Huang retreated. “The odds of 100,000 of those agents building Nvidia is zero percent,” he said.
AGI has arrived, in other words, right up until someone bolts a testable yardstick onto it. At that point it hasn’t.
Everyone’s definition is different, and that’s on purpose
For years, tech executives have exploited the term’s haziness to generate an assortment of definitions and benchmarks, each orbiting the same hazy concept: AI that equals or exceeds human intelligence across many domains. That “intelligence” itself lacks any universally accepted definition is treated as beside the point.
Building AGI was OpenAI’s stated founding purpose, and the company’s charter still permits vast interpretive latitude, describing it as “highly autonomous systems that outperform humans at most economically valuable work.” Sam Altman has granted as much, acknowledging last year that AGI is “not a super useful term.”
There is also the definition OpenAI reportedly negotiated with Microsoft, which is monetary: systems capable of producing a minimum of $100 billion in profits. Elsewhere, chief research officer Mark Chen put OpenAI at “80% of the way” to AGI, while Altman predicted the company would possess something he’d be willing to call AGI before the year was out.
Eighty percent of the way to a destination with no coordinates.
The rebranding hasn’t helped
The vagueness isn’t being concealed. Anthropic CEO Dario Amodei has described AGI as “imprecise,” even a “marketing term,” favoring “powerful AI” instead. Rivals have minted their own labels for roughly the same concept. The distinctions are supposed to carry weight. In reality they blur into one another.
Meta prefers “personal superintelligence.” Microsoft offers “humanist superintelligence.” Amazon opts for “useful general intelligence.” Demis Hassabis of Google DeepMind speaks of reaching the “foothills of the singularity.” And OpenAI cofounder Ilya Sutskever — who reportedly had staff chanting “feel the AGI” — is now at the helm of a company named Safe Superintelligence.
The fresh vocabulary hasn’t made the concept any more concrete. It has simply made it easier to shift.
Why the talk won’t stop
For as long as AGI remains loosely defined and loosely deployed, the entire exercise is senseless. Unless, naturally, what you want is a convenient device for inflating your progress — in which case an undefined milestone you can claim on an earnings call is precisely the right tool for the job.
More of this is coming from the industry, Huang very much included. Perhaps an AGI will turn up one day and explain to us what AGI actually means.