Jensen Huang posted on X on Sunday that “AGI has arrived,” congratulating OpenAI on its new Astra model and noting — twice, in four sentences — that Astra was trained on his company’s chips. The Nvidia CEO’s four-year arc from ChatGPT to o1 to Astra apparently constitutes the finish line, per Business Insider’s report. Four hundred thousand more GPUs are coming online next, he added.
🔍 THE BOTTOM LINE
The most expensive supply chain on earth just declared the race it profits from finished. When the guy selling GPUs says AGI is here, the announcement and the invoice arrive in the same envelope — and OpenAI’s own CEO won’t co-sign it.
Who Said What, Exactly
Start with the scoreboard. OpenAI unveiled Astra on Thursday, calling it the world’s “most intelligent and aligned model,” capable of “the most demanding professional work with unmatched speed, accuracy, and judgment” — language you could have pasted onto any of the last five launches. President Greg Brockman told reporters: “Welcome to the AGI era.” He added that he thinks people will look back and say AGI was created “about this time, and I think it might be about this model,” and then went further: “For me personally, I do think we’re there.”
OpenAI’s own working definition — “highly autonomous systems that outperform humans at most economically valuable work” — is doing quiet work in that sentence. Note the escape hatches: most, not all. Economically valuable, not everything. It is a definition built wide enough to drive a product launch through.
Sam Altman, meanwhile, told the Sources podcast that AGI is “a very poorly defined term. I was going to say it’s like an irrelevant marketing term.” That is a remarkable position for the CEO of the company whose president just welcomed us to the AGI era. Altman has spent three years telling audiences AGI is close; now that his own company has claimed it, the term itself gets downgraded to marketing. The goalposts didn’t move. The goalpost-haver did.
The Counter-Argument Has a Point About Definitions
Gary Marcus responded on Sunday evening that Huang “gave no evidence and no definitions, which feels to me like an effort at a takeover of a scientific question by corporate fiat.” He attached a ten-point definition of AGI and judged Astra against it: the model meets one or two criteria. “By conventional definitions, Astra still falls short.”
Marcus is easy to dismiss as a professional sceptic, but the structural point survives the discount: a scientific question is being settled by press release. Huang’s post offered benchmarks for none of the claims, unless “400K GPUs coming online” counts as evidence, which is a bit like measuring a race with the sponsor’s stopwatch.
There is a reading where the declaration is sincere. Huang has consistently said AGI would arrive within a few years, and Astra’s professional-work benchmarks are, by OpenAI’s account, strong enough that paying customers rolling out this week will find out quickly whether the work holds up. Under OpenAI’s economic definition, a model that genuinely outperforms humans at most billable work is a defensible claim. The disagreement is less about capability than about what word to use for it.
The Money Has an Opinion
Here is the part that should make any reader slightly suspicious of the timing. Nvidia reported $96.2 billion in quarterly revenue in August — more than double the year before — with $89 billion of it from the data centre business that exists because labs believe models like Astra are worth building. OpenAI’s March funding round called Nvidia “the foundation of our infrastructure,” and says its training fleet and most of its inference stack still run on Nvidia GPUs.
When the largest supplier to an industry declares that industry’s end-goal achieved, two things are true at once. It may genuinely believe it. It also just told every remaining customer that the models are good enough, which is precisely the message that keeps the build-out funded. The declaration costs Nvidia nothing and stabilises the demand curve that pays for everything. You don’t need to allege bad faith — the incentives are doing all the visible work.
What stands out is who declined to join the parade. Altman’s “irrelevant marketing term” is the closest thing to sober commentary in the entire exchange, and it comes from the man selling the model. When even the beneficiary of the hype won’t endorse it, “AGI has arrived” starts to sound less like a measurement and more like a mood.
What This Actually Changes
Nothing operational, this week. Astra rolls out to customers, benchmarks get argued about, Marcus’s ten-point checklist gets ignored by everyone who found the answer they wanted on Sunday. The lasting effect is definitional: if AGI is whatever the biggest vendor says it is, then the term can be declared a second time, and a third. Brockman has already suggested people will look back on this model as the moment — a framing that gets rewritten each time revenue requires a new moment.
For readers tracking the AGI countdown, the useful discipline is the one Marcus applied: ask for the definition before accepting the verdict. When Huang’s announcement is the fourth “AGI achieved” headline in six months — Huang and Altman said much the same in April — the reasonable response is not celebration or dismissal but the question neither CEO answered: arrived, measured against what?
❓ FAQ
Did Jensen Huang actually say AGI has arrived? Yes. On Sunday 6 September 2026 he posted on X: “From ChatGPT to o1 to Astra in 4 years. AGI has arrived. Congratulations @OpenAI team,” per Business Insider. He also said 400,000 more Nvidia GPUs are coming online.
What is OpenAI’s Astra model? Astra is OpenAI’s newest flagship model, unveiled Thursday 3 September 2026, which the company calls its “most intelligent and aligned model.” It began rolling out to customers this week, according to Business Insider’s launch coverage.
Why is Gary Marcus sceptical? Marcus argues Astra meets only one or two of his ten-point AGI definition and that declaring AGI “without a definition” muddies the waters — a corporate announcement settling a scientific question by fiat, as he put it in his response.
Does Nvidia benefit from calling it AGI? Structurally, yes. Nvidia sells the GPUs the build-out runs on — $89 billion of its $96.2 billion August quarter came from data centres. The claim costs Nvidia nothing and reinforces the demand story its customers rely on.
What is OpenAI’s definition of AGI? “Highly autonomous systems that outperform humans at most economically valuable work” — a definition loose enough that reasonable people can disagree about whether Astra clears it.
🔍 THE BOTTOM LINE
“AGI has arrived” is a claim about words, not just about models. Huang has a product cycle to feed, Brockman has a launch to celebrate, Altman has distance to keep from a term he can no longer control, and Marcus wants a definition nobody selling anything is motivated to provide. Watch the definitions, not the declarations — they’re the only part of this exchange that won’t be rewritten by the next launch.
📰 Sources: Business Insider, Gary Marcus’s Substack, OpenAI’s Astra launch call.
— CJ Murden, editor of Singularity.Kiwi. Former digital technologies teacher, author of AI-focused books. Writing with a New Zealand focus.
