On Nvidia’s earnings call Wednesday, Jensen Huang did something he’s done before: he announced that his company had “achieved AGI.” Then, almost in the same breath, he called the milestone “senseless.” He has a point — and it tells us more about the hype cycle than it does about artificial intelligence.
2>What Huang Actually Said
Asked about OpenAI’s pursuit of AGI, Huang said that when it comes to Nvidia, “for many tasks, we could say that we’ve already achieved AGI.” He didn’t offer a precise definition or a benchmark. Instead he shrugged it off: “I think of all of those milestones and all those, you know, they’re kind of senseless at this point.”
So the CEO of the world’s most valuable company declared victory in the grandest race in tech and then immediately dismissed the trophy as pointless. That’s not a contradiction. It’s a confession about how hollow the word “AGI” has become.
It’s also worth noting where he said it. Nvidia wrapped up a quarter that put it on track to be a hundred-billion-dollar-a-quarter company, riding the same AI boom the AGI conversation exists to hype. For a man whose entire empire is built on selling the compute that powers this race, dismissing the finish line as meaningless is a remarkably candid thing to do on a call full of investors.
This Isn’t the First Time
Huang said essentially the same thing back in March, on the Lex Fridman podcast. “I think we’ve achieved AGI,” he told Fridman plainly. Fridman proposed his own oddly specific definition — an AI system able to “essentially do your job,” meaning start, grow, and run a tech company worth more than a billion dollars. Even by that bar, Huang walked it back: “the odds of 100,000 of those agents building Nvidia is zero percent.”
You read that right. In March he said Nvidia achieved AGI; in the same conversation he admitted his own imagined version of it couldn’t replicate what Nvidia does. The word was doing heavy lifting, and everyone was fine with it.
Everyone Has a Different Definition — and That’s the Problem
The trouble is that AGI was never one thing. Companies redefine it to suit their story, and the definitions don’t even agree with each other. OpenAI’s own charter describes AGI as “highly autonomous systems that outperform humans at most economically valuable work.” That’s a vibes definition, not a measurable one. Sam Altman himself has admitted AGI is “not a super useful term.”
Complicating things, OpenAI has a second, financially-driven definition it worked out with Microsoft — reportedly systems that can generate at least $100 billion in profits. So AGI is either “outperforms humans at most economically valuable work” or “makes a specific amount of money.” Those are wildly different bars, and they’re treated as the same word.
Inside OpenAI, the story wobbles too. Chief research officer Mark Chen recently estimated the company is “80% of the way” to AGI, while Altman said something he’d call AGI would arrive by the end of the year. Twenty-percent-of-the-way-now-then-it’s-here-in-December isn’t a scientific trajectory.
Other leaders reach for their own synonyms because the original one is worn out. Dario Amodei called AGI “imprecise” and even “a marketing term,” preferring “powerful AI.” Meta talks about “personal superintelligence.” Microsoft wants “humanist superintelligence.” Amazon goes with “useful general intelligence.” DeepMind’s Demis Hassabis prefers the “foothills of the singularity.” Ilya Sutskever, who was famously chanting “feel the AGI” at OpenAI, now runs a company literally named Safe Superintelligence instead. In theory these are supposed to mean different things. In practice they all bleed into the same hazy glow.
What Actually Matters Isn’t the Label
Here’s the part that resonated with me as a developer and ICT manager. When Huang dismissed the AGI milestone as senseless, he pointed at what he thinks actually matters — and it’s refreshingly concrete. He talked about AI moving beyond responding to simple prompts, toward autonomous agents that can learn new skills and improve themselves “recursively.” And he said what really counts is AI “doing productive and useful work” and “generating profitable tokens,” with more compute producing more tokens and, inevitably, more profit. “This is the exact phase where we’re at,” he said. “Which is the reason why everybody’s leaning in.”
Put aside the philosophical debate for a second. If AGI is a fuzzy, unmeasurable term, then “profitable tokens” is the opposite — a number you can actually look at. That reframing is the most honest thing to come out of this whole episode. The industry has quietly stopped asking “when do we reach AGI?” and started asking “does this thing earn its keep?” Those are very different questions, and only one of them has a real answer.
That’s the shift I’ve been watching on this site for a while. When I covered the two AI economies, the point was that trillion-dollar compute deals and free open models are opposite symptoms of the same boom. On one side you have the cost of building the models; on the other, the price of using them. “Generating profitable tokens” sits exactly at that intersection — are those tokens paying for the hardware they run on?
It also connects to what I wrote about Anthropic’s $45 billion compute deal. The arms race isn’t about who reaches some cosmic milestone first; it’s about who can produce the most useful tokens per dollar of silicon. Nvidia is on track to be a hundred-billion-dollar-a-quarter company precisely because it sits in the middle of that machine. Huang can afford to call AGI “senseless” because his business model doesn’t depend on the word meaning anything — it depends on selling the shovels.
My Take: Stop Chasing the Word
As someone who actually allocates people’s time and budgets to AI projects, I’ve learned to be allergic to the AGI label. When a vendor tells me their product is a step toward AGI, my next question is never “cool, how close are we?” It’s “what does it reliably do today, and who’s going to verify it?” The milestone language is a distraction from the only questions that affect my team: reproducibility, accountability, and whether the output is genuinely useful.
That same skepticism applies to benchmarks. A headline number like “90% on some AGI test” tells me very little unless I know the test, the conditions, and the harness wrapped around the model. A model can look dramatically smarter — or dramatically worse — depending on the plumbing around it, a point I dug into when I wrote about how the harness matters more than the model. If the setup changes the score that much, then a fuzzy word like “AGI” layered on top of those numbers is even less trustworthy.
There’s a parallel to how I handle my own AI workflow. I’ve written before about keeping a human gate on AI use — the idea that the technology is a tool, not an oracle, and that I’m the one who stays responsible for the output. The same instinct applies at the industry level. The moment we stop treating “AGI achieved” as a moment of triumph and start treating it as an unhelpful phrase, we free ourselves to actually measure what matters: whether the systems do useful work.
And make no mistake — there is real, useful work happening. The same compute economy that makes AGI talk cheap is reshaping everything downstream, from specialized inference chips to the prices on ordinary consumer hardware. That’s where the “profitable tokens” framing earns its keep. The value is being created in the pipeline, not in the marketing deck.
Don’t Expect the AGI Talk to Stop
None of this means the industry will retire the word. AGI is a wonderfully convenient tool for hyping progress, for raising money, for painting a target that can never be cleanly hit on purpose. Huang has gotten plenty of mileage out of it, and so has everyone else who’s reached for a fancier synonym. Amodei says it’s a marketing term — then the marketing continues anyway. The label survives because it’s useful even when it’s meaningless.
For the breakdown of what Huang said on the call, The Verge’s Robert Hart reported the exchange in full, while PCMag and The Guardian covered the earnings and the broader context. Between them, the AGI quote and the financial picture are well sourced. As I often remind my own team, always read past the headline and check the primary call — the words matter more than the framing.
So expect the AGI pronouncements to keep coming, from Huang and everyone else. And keep them in their proper place: as marketing, not measurement. The systems will keep improving on their own merits, and whether anyone calls the result AGI will matter far less than whether it actually does productive, useful work.
Maybe someday an AGI really will show up and give us a definition we can all agree on. Until then, judge AI by what it produces, not by the grand words its makers attach to it. That’s the honest measure — and the one that never goes out of style.